ProgrammableMarketing/subvocal.company (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 7
  • This quarter · 11
  • Big bets · 3
  • Wildcards · 2
  • How we can help
Generated Nov 10, 2026
  1. Programmable
  2. Playbooks
  3. The Subvocal Company
The Subvocal Company logo

growth playbook · Nov 10, 2026

subvocal.company (opens in a new tab)

The Subvocal Company growth playbook

Diagnosis

Subvocal had a big launch: roughly 750 paid preorders, about $150,000 in refundable deposits, over 12 million views on X and an 847-point Show HN in its first nine days. The website hasn't caught up. It has five pages, and the preorder, terms and privacy pages show only a heading and one sentence to anything that reads the page text.

So the press filled the gap with three different price stories, and the first Hacker News comment asked whether it works at all. The playbook starts by fixing that: official facts and an honest FAQ on /preorder, batch positions for buyers past the 500-unit first run, a device-data page for the most common objection, a waitlist, and a published error rate.

Next it turns the 750 buyers into survey answers, paid data participants and a calibration-first onboarding, opens the device to any app, and builds the hiring and build-log pages needed to ship by Q1 2027. The big bets are on-device decoding, a first market outside the US and a developer SDK.

Growth Score

7/10

Biggest Opportunity

Turn launch attention into trust and data: one honest source of truth for price, delivery and accuracy, and 750 committed buyers who can shape and train the product.

Biggest Risk

A long wait for unproven hardware: if accuracy, privacy answers or delivery dates disappoint before Q1 2027, refundable deposits can leave as fast as they came, while AlterEgo and dictation apps compete for the same users.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Put the price, deposit terms and a "does it work" FAQ in plain text on /preorder

    Write the eight facts (deposit, refund, early-bird total, balance timing, delivery quarter, countries, exclusions, batch) into the /preorder page's HTML today.

    High impact First result in 1 week
  2. 2

    Post an uncut accuracy video with the error rate on screen

    Pick 50 unseen sentences and record one take with someone who hasn't used the device.

    High impact First result in 2 weeks
  3. 3

    Tell all 750 preorder holders which batch they're in, now that the 500-unit first run is oversubscribed

    Sort the preorder list by payment time and mark the 500th order.

    High impact First result in 1 week

02

Your idea portfolio

23 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 8Big bets · 6Fill-ins · 7Reconsider · 2
Effort →

Showing 23 of 23 ideas

This week 7

Quick wins with a clear first action.

  • 01

    Subvocal's /preorder page promises "answers to common questions about the first production batch", but its readable text is one heading and one sentence: no price, deposit amount, delivery window or FAQ. /preorder-terms and /privacy are the same. Search engines, link previews and AI assistants that don't run scripts have nothing to quote, and the press has filled the gap with different stories. The Register says $499 in total ($199 now, $300 on shipping). IBTimes adds a $699 regular price and US-only preorders. The YC launch describes a $199 reservation with $200 off a final price it doesn't name. And the first Hacker News comment on the site was "Faq doesn't seem to indicate whether it works or not".

    High impactLow effort First result in 1 week Founder/CEO

    First action

    Write the eight facts (deposit, refund, early-bird total, balance timing, delivery quarter, countries, exclusions, batch) into the /preorder page's HTML today.

    How to measure it

    Visit-to-deposit conversion on /preorder

    Now
    –
    Target
    +20%
    By
    30 days

    Steps

    1. 1

      Agree the facts

      One official version of the reservation, refund, early-bird total, balance timing and delivery window.

      Founder/CEO 1 day

    2. 2

      Put it in the HTML

      Facts block and full terms in the page source, not only inside a script or checkout widget.

      Founder/CTO 1 day

    3. 3

      Write the FAQ

      Ten questions from the Hacker News threads and the inbox, starting with accuracy and calibration.

      Founder/CEO 1 day

    4. 4

      Add structured data

      Product with offer price, and FAQPage markup that matches the visible text.

      Founder/CTO 2 hours

    5. 5

      Check what's read

      Load the pages with scripts off and run Search Console's URL inspection.

      Founder/CTO 1 hour

    6. 6

      Ask for corrections

      Send the facts to the outlets that published different prices.

      Founder/CEO 1 day

    7. 7

      Measure

      Visit-to-deposit rate and support emails for two weeks before and after.

      Founder/CEO 2 weeks

    First result
    1 week
    Cost
    $0 (founder time)
    Owner
    Founder/CEO

    Smallest test

    Add a five-line facts box (deposit, refund, total, delivery quarter, where you ship) to /preorder today and watch conversion for a week.

    Why it works

    Put one plain-HTML facts block at the top of /preorder, and the full terms on /preorder-terms, that does not depend on scripts. It should state the official facts: the reservation amount and how refunds work, the early-bird total and when the balance is charged, whether the regular price is set, the expected delivery quarter and what it depends on, where you ship today, what's not included (shipping, tax, prescription lenses) and which batch a new order joins. Below it, add an FAQ for the questions buyers actually ask: does it work today and how accurately, how long calibration takes, where decoding happens, lenses and fit, battery, and which apps it types into. Add Product and FAQPage structured data with the same numbers, so search results and assistants quote Subvocal rather than a news story.

    Most visitors after a 12-million-view launch arrive from a shared link, already half-convinced. What stops a $199 decision is not knowing the total cost, the refund rules or whether it works, and right now only third-party articles answer those questions, and they disagree. One source of truth lifts conversion, cuts support email and makes the next press story correct.

    Ready to use

    ChecklistFacts box for /preorder

    [ ] Reservation amount, refundable: how and how fast
    [ ] Early-bird total and when the balance is charged
    [ ] Regular price: set, or not yet final
    [ ] Expected delivery quarter and what it depends on
    [ ] Where we can ship today
    [ ] Not included: shipping, tax, prescription lenses
    [ ] Which batch a new order joins
    [ ] Does it work today? Error rate and calibration time
    [ ] Where decoding happens and what is stored
    [ ] Visible in the page HTML with scripts off
    [ ] Product and FAQPage structured data match the text

    Risks and how to handle them

    • Publishing a final price before costs are known

      Fix the deposit and early-bird total, and say plainly that the regular price is not final yet.

    • An honest accuracy answer puts some buyers off

      Show the number with its trend and the refundable deposit; buyers who stay will cancel less later.

    Evidence

    • Your websitePreorder page, read twice on November 10, 2026: the readable text is the heading "Preorder Subvocal" and one sentence promising "preorder details, and answers to common questions about the first production batch", about 19 words in all. No price, deposit amount, delivery date, FAQ or structured data appear. /preorder-terms likewise shows only a line saying it covers "refundable deposits, final payment, cancellations, and estimated delivery", and /privacy only a heading and a one-line introduction.Source
    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Hacker News"Talk Without Talking" (October 30, 2026), linking to subvocal.company: its only comment reads "Faq doesn't seem to indicate whether it works or not".Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 02

    The Show HN post on November 2 offered preorders for 500 units. Six days later the launch update counted roughly 750 paid preorders. The YC launch sold the $199 reservation on "earliest shipping" in the first limited batch. If the first run is still 500 units, around 250 buyers paid for a first-batch slot they won't get, and nothing on the site says how batches are allocated.

    High impactLow effort First result in 1 week Founder/CEO

    First action

    Sort the preorder list by payment time and mark the 500th order.

    How to measure it

    Deposit refund rate

    Now
    –
    Target
    Under 5 per 100
    By
    60 days

    Steps

    1. 1

      Rank orders

      By payment time, with any agreed exceptions.

      Founder/CEO 2 hours

    2. 2

      Set batch sizes

      Match the production plan for the first and second runs.

      Founder/CTO 1 day

    3. 3

      Write the email

      Batch, window, what could move it, refund link.

      Founder/CEO 2 hours

    4. 4

      Send and log replies

      Tag questions and refund requests.

      Founder/CEO 1 day

    5. 5

      Update /preorder

      Show the batch now reserving and its window.

      Founder/CTO 2 hours

    6. 6

      Update monthly

      Batch status in every build-log email.

      Founder/CEO monthly

    First result
    1 week
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Email the buyers after order 500 first and track replies and refunds for a week.

    Why it works

    Email every buyer their order number, their batch and that batch's expected window, and show the same on /preorder for new buyers ("Now reserving batch 2"). Give later-batch buyers something real for waiting, such as early access to the companion app or a calibration session, plus a one-click refund link.

    Telling people where they stand before they ask keeps refunds low, and it turns a sold-out first run into proof of demand for the second.

    Risks and how to handle them

    • Later-batch buyers cancel

      Offer a perk for waiting and an honest window.

    • Hardware dates slip

      Give ranges, not dates, and update every month.

    Evidence

    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
  • 03

    The most frequent concern in the 395-comment Show HN thread is cloud decoding of silent speech. The founders answered in the thread with safeguards and a hope for local inference later, but that answer lives only in a comment. The published privacy page describes itself as covering website visits, preorders and contacting the team; it says nothing about the glasses' sensor data. AlterEgo already tells visitors it responds only to deliberate silent speech, and Superwhisper sells offline use.

    High impactLow effort First result in 2 weeks Founder/CTO

    First action

    Turn the founders' Hacker News reply on safeguards into a first draft.

    How to measure it

    Visit-to-deposit rate for /data viewers

    Now
    –
    Target
    At or above the site average
    By
    45 days

    Steps

    1. 1

      Map the data

      Every signal from the glasses to the app, the cloud and storage.

      Founder/CTO 1 day

    2. 2

      Draft in plain English

      Ten short answers, with today's facts and the plans labelled as plans.

      Founder/CEO 2 days

    3. 3

      Legal review

      Check the page and an updated privacy policy.

      Outside counsel 1 week

    4. 4

      Publish and link

      From /preorder, the FAQ and the footer.

      Founder/CTO 2 hours

    5. 5

      Answer the threads

      Link the page when privacy comes up on X and Hacker News.

      Founder/CEO ongoing

    6. 6

      Device policy

      Full device and app privacy policy before shipping.

      Founder/CEO 1 month

    First result
    2 weeks
    Cost
    A few hours of legal review
    Owner
    Founder/CTO

    Smallest test

    Add a five-question privacy section to /preorder this week.

    Why it works

    Publish a /data page, linked from /preorder and the FAQ, that covers: what the glasses capture and when (only while the trigger is held?), what goes to the cloud, what's stored and for how long, who can access raw signals, whether buyers' data trains models and how to opt out, how to delete it, and the plan for local decoding. Extend the privacy policy to cover the device and app before the first batch ships.

    Privacy is the objection people raise in public. A written, specific answer turns a reason to wait into a reason to trust, and it beats rivals who only promise.

    Risks and how to handle them

    • Promising more than the system does

      Describe only what is built today and date the page.

    • Legal exposure from loose wording

      Have counsel review before publishing.

    Evidence

    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source
    • Your websitePreorder page, read twice on November 10, 2026: the readable text is the heading "Preorder Subvocal" and one sentence promising "preorder details, and answers to common questions about the first production batch", about 19 words in all. No price, deposit amount, delivery date, FAQ or structured data appear. /preorder-terms likewise shows only a line saying it covers "refundable deposits, final payment, cancellations, and estimated delivery", and /privacy only a heading and a one-line introduction.Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
  • 04

    The site has five pages and no newsletter or waitlist signup. The only ways in are a paid reservation or an email address. Preorders are reported as US-only, and the founders told a European commenter "Not immediately, no". A launch seen over 12 million times reached many people who can't or won't pay $199 yet, and they leave with nothing. AlterEgo, with no product on sale, still collects "Get Notified" emails.

    Medium impactLow effort First result in 1 week Founder/CEO

    First action

    Add an email and country form below the preorder button.

    How to measure it

    Waitlist signups

    Now
    –
    Target
    5,000
    By
    60 days

    Steps

    1. 1

      Build the form

      Email, country and use, with a consent checkbox.

      Founder/CTO 2 hours

    2. 2

      Place it

      Homepage, /preorder and the launch update.

      Founder/CTO 1 hour

    3. 3

      Welcome email

      Demo video, facts page and what happens next.

      Founder/CEO 2 hours

    4. 4

      Add to the build log

      Waitlist gets the monthly update.

      Founder/CEO monthly

    5. 5

      Country report

      Signups by country each month.

      Founder/CEO monthly

    6. 6

      Convert

      Invite US signups when a batch opens.

      Founder/CEO per batch

    First result
    1 week
    Cost
    $0 to $50 a month for an email tool
    Owner
    Founder/CEO

    Smallest test

    Add the form to /preorder only and count signups for a week.

    Why it works

    Add an email-only form to the homepage and /preorder with two optional questions: country, and where you'd use it. Send a welcome email with the demo video and the monthly build log (idea 14). Use the country counts to pick the first market outside the US (idea 20), and invite undecided US signups to reserve when each batch opens.

    Capturing intent costs nothing, and the list becomes demand data for certification, a second batch and launch day.

    Risks and how to handle them

    • The list goes cold before shipping

      Send a short build log every month.

    • Consent rules for non-US signups

      Clear consent wording and a line in the privacy policy.

    Evidence

    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 05

    Subvocal says it decodes 200+ words per minute with a "reasonably low error rate", but gives no number. The Register lists accuracy as undisclosed, the first Hacker News comment asked whether it works, and Show HN commenters worked out their own figure of roughly 5% error from the demo video. Right now the only accuracy number in circulation comes from commenters, not from Subvocal.

    High impactLow effort First result in 2 weeks Head of ML

    First action

    Pick 50 unseen sentences and record one take with someone who hasn't used the device.

    How to measure it

    Word error rate, new user after calibration

    Now
    –
    Target
    Published monthly, falling each month
    By
    90 days

    Steps

    1. 1

      Choose the test set

      50 everyday sentences: emails, prompts, a line of code.

      Head of ML 1 day

    2. 2

      Record one take

      New user, calibration included, no cuts.

      Founder/CTO 1 day

    3. 3

      Overlay the numbers

      Reference, decoded text and running error rate.

      Founder/CEO 1 day

    4. 4

      Publish

      /preorder, launch update and X.

      Founder/CEO 2 hours

    5. 5

      Repeat monthly

      A new person each month, same test set.

      Head of ML monthly

    6. 6

      Track

      Deposits from viewers of the video.

      Founder/CEO monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Head of ML

    Smallest test

    Post one uncut 60-second clip with the error count this week.

    Why it works

    Record a five-minute single take: a person who has never used the device calibrates, then silently reads 50 sentences they haven't seen, with the reference text, the decoded text and a running word error rate on screen. Publish it on /preorder and in the launch update, and repeat it monthly with a new person, so a falling error rate becomes the progress story.

    Showing errors is more believable than claiming none, and a number that falls every month is proof of progress few prototype-stage rivals publish.

    Risks and how to handle them

    • The number looks worse than hoped

      Show it with calibration time and the trend; publish conditions.

    • Compared with dictation apps unfairly

      State the test conditions on screen.

    Evidence

    • Hacker News"Talk Without Talking" (October 30, 2026), linking to subvocal.company: its only comment reads "Faq doesn't seem to indicate whether it works or not".Source
    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
  • 06

    Real visitors already pass Core Web Vitals (LCP 1.7s), so this isn't urgent. But in Google's mobile lab test the largest element is the demo video, which appears at 4.7s, and nine third-party hosts make up 452KB of a 1,205KB page, with 234KB of unused JavaScript. Launch traffic from X arrives on phones and slower connections too.

    Low impactLow effort First result in 1 day Founder/CTO

    First action

    Add a poster attribute with a compressed still to the hero video.

    How to measure it

    Mobile performance score

    Now
    –
    Target
    90+
    By
    2 weeks

    Steps

    1. 1

      Poster image

      Compressed still from the demo.

      Founder/CTO 1 hour

    2. 2

      Defer the stream

      Load the video after first paint.

      Founder/CTO 2 hours

    3. 3

      List the tools

      Each third-party script, its owner and purpose.

      Founder/CTO 1 hour

    4. 4

      Remove duplicates

      Drop tools that overlap or are unused.

      Founder/CTO 2 hours

    5. 5

      Fix contrast

      The one flagged text colour.

      Founder/CTO 30 minutes

    6. 6

      Re-test

      PageSpeed mobile and desktop.

      Founder/CTO 30 minutes

    First result
    1 day
    Cost
    $0
    Owner
    Founder/CTO

    Smallest test

    Add the poster image and re-run PageSpeed.

    Why it works

    Give the video a compressed poster image and load the stream after first paint. List the nine third-party tools with an owner and a purpose for each, and remove duplicates and anything nobody uses. Fix the one colour-contrast issue, then run PageSpeed again.

    A faster first frame on phones keeps more launch visitors on the page long enough to see the demo, at almost no cost.

    Risks and how to handle them

    • Removing a tool someone relies on

      Check each one's owner before removing it.

    Evidence

    • PageSpeedGoogle PageSpeed (November 10, 2026): mobile score 78, desktop 95. Real visitors pass Core Web Vitals (LCP 1.7s, INP 106ms, CLS 0). In the mobile lab test the largest element is the demo video, shown at 4.7s; suggested fixes are unused JavaScript (about 0.9s, 234KB), image delivery (0.6s, 193KB) and cache lifetimes (0.4s). Nine third-party hosts make up 452KB of the 1,205KB page; one colour-contrast issue.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 07

    The YC launch asks people to share where they would use Subvocal, and names offices, cafes, transit, meetings and coding. Show HN commenters add libraries and noisy places. With about 750 paying buyers, those answers can decide the first use-case pages, the companion app's default modes, the calibration sentences and who to recruit for data sessions. Today they arrive scattered across comments.

    High impactLow effort First result in 1 week Founder/CEO

    First action

    Write the two questions and send them to the first 100 buyers.

    How to measure it

    Preorder holders who answer

    Now
    –
    Target
    40%
    By
    2 weeks

    Steps

    1. 1

      Write two questions

      Where, and into what.

      Founder/CEO 1 hour

    2. 2

      Send

      To all preorder holders, with one reminder.

      Founder/CEO 1 day

    3. 3

      Tag buyers

      Store answers against each order.

      Founder/CTO 2 hours

    4. 4

      Share results

      Top answers in the next launch update.

      Founder/CEO 1 day

    5. 5

      Feed the product

      Calibration sentences and app modes from the top uses.

      Head of ML 2 weeks

    6. 6

      Repeat

      Again after the first batch ships.

      Founder/CEO per batch

    First result
    1 week
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Send it to 100 buyers today.

    Why it works

    Send a two-click email survey: "Where will you use Subvocal most?" (open office, meetings, coding, commuting, library or study, at home while others sleep, other) and "What will you type into most?" (email, chat, AI assistants, code editor, notes). Tag each buyer with their answers and use them in ideas 8, 9, 13 and 21.

    Buyers who just paid are the most willing to answer, and their answers turn guesses about the market into a ranked list.

    Risks and how to handle them

    • Low response

      Keep it to two clicks and share the results back.

    Evidence

    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source

This quarter 11

The core plays for the next three months.

  • 08

    Subvocal's cross-user model is trained on data from paid participants, and the plan calls for thousands of hours of population-scale RF data. The Head of ML must make it work across users, sessions and devices with about 15 minutes of calibration, and the Head of Hardware is building portable prototypes for large-scale data collection. Meanwhile 750 people have already paid to be early: the most motivated participant pool there is, and exactly the users the model has to work for.

    High impactMedium effort First result in 1 month Head of ML

    First action

    Add a "Bay Area? Want to help train Subvocal?" question to the buyer survey.

    How to measure it

    Participant hours recorded

    Now
    –
    Target
    1,000 hours
    By
    1 quarter

    Steps

    1. 1

      Consent terms

      Separate data consent, payment terms, deletion path.

      Founder/CEO 2 weeks

    2. 2

      Find participants

      Location and availability from the buyer survey.

      Founder/CEO 2 days

    3. 3

      Ready the rigs

      Portable prototypes reliable enough for sessions outside the lab.

      Head of Hardware 1 month

    4. 4

      Run a pilot

      Ten participants, two sessions each on different days.

      Head of ML 2 weeks

    5. 5

      Track coverage

      Spread across fits, sessions and speaking styles; fill gaps from panels.

      Head of ML weekly

    6. 6

      Report back

      Share held-out accuracy with participants monthly.

      Head of ML monthly

    First result
    1 month
    Cost
    Participant payments and session staff time
    Owner
    Head of ML

    Smallest test

    Run five sessions with local buyers next month.

    Why it works

    Invite preorder holders and waitlist signups in the Bay Area to paid recording sessions with the portable rigs, paid in cash or as credit against the balance. Track how well the pool covers different face shapes, glasses fits and speaking styles, and fill gaps from paid panels. Use a separate, clear consent form with an opt-out and deletion path, and report held-out results back to participants.

    Recruiting from your own buyers is cheaper and faster, gets more careful sessions, and keeps the people waiting for the product involved in building it.

    Risks and how to handle them

    • Consent and privacy concerns

      Separate consent from the purchase, with an easy opt-out and deletion.

    • A narrow pool of early adopters

      Measure coverage and fill gaps from paid panels.

    Builds on

    Evidence

    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • HiringHead of ML (YC Work at a Startup, $150K-$200K, in person in San Francisco): owns turning noisy physiological signals into continuous language, generalising across users, sessions and devices, adapting to a new user with about 15 minutes of calibration data, and designing held-out user, session and device evaluations. The company page lists a team of 3, backed by Y Combinator and Afore Capital.Source
    • HiringHead of Hardware ($125K-$200K): portable prototypes reliable enough for large-scale data collection, miniaturisation into glasses, and ownership of the BOM, DFM/DFT, regulatory preparation, contract manufacturers and the path from EVT to production. Head of Sensing ($150K-$250K): new sensing geometries, calibration across people, sessions and devices, and work with external RF, fabrication, simulation and regulatory partners.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 09

    The Head of ML post sets a goal of adapting to a new user with about 15 minutes of calibration data, and The Register reports a YC partner calibrated the prototype by reading 20 sentences. For a buyer, calibration is the first thing they ever do with the glasses. If it feels like homework or ends with poor accuracy, the first impression and the refund decision happen in the same quarter of an hour.

    High impactHigh effort First result in 1 quarter Head of ML

    First action

    Define the accuracy threshold that counts as "calibrated".

    How to measure it

    New users reaching target accuracy in session one

    Now
    –
    Target
    80%
    By
    First batch shipping

    Steps

    1. 1

      Set the threshold

      What error rate counts as ready.

      Head of ML 1 week

    2. 2

      Write scripts

      Calibration sentences per use case from the survey.

      Founder/CEO 1 week

    3. 3

      Live meter

      Accuracy shown as the user goes.

      Founder/CTO 2 weeks

    4. 4

      First real message

      End by sending something real.

      Founder/CTO 1 week

    5. 5

      Test with 20 people

      Watch where they stall.

      Head of ML 2 weeks

    6. 6

      Instrument

      Time, drop-off and accuracy per step.

      Founder/CTO 1 week

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Head of ML

    Smallest test

    Run the current calibration with five new people and time it.

    Why it works

    Build calibration into onboarding: sentences from the buyer's own use case (from the survey: emails, prompts, code), a live accuracy meter that rises as they go, and an ending where they silently send a real message. Measure time to the first accurate sentence and the share of new users who reach the target accuracy in their first session.

    The moment a person sees their own silent words appear correctly is the product's whole promise; designing for it cuts early returns and creates the clips people share.

    Risks and how to handle them

    • Calibration takes longer than 15 minutes

      Let users stop early and keep improving in the background.

    Builds on

    Evidence

    • HiringHead of ML (YC Work at a Startup, $150K-$200K, in person in San Francisco): owns turning noisy physiological signals into continuous language, generalising across users, sessions and devices, adapting to a new user with about 15 minutes of calibration data, and designing held-out user, session and device evaluations. The company page lists a team of 3, backed by Y Combinator and Afore Capital.Source
    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
  • 10

    The launch says people put up with gooseneck whisper mics and soundproof masks today. Wispr Flow says it can handle a whisper if the mic is close to the mouth, Superwhisper sells offline dictation, and AlterEgo promises silent speech but isn't on sale. Show HN commenters compared Subvocal with AlterEgo, Cornell research and lip-reading projects. The five-page site has no page that answers "why this instead of what I use now?"

    Medium impactLow effort First result in 1 month Founder/CEO

    First action

    Verify each rival's current claims on their own homepage and date the page.

    How to measure it

    Deposits from /compare visitors

    Now
    –
    Target
    50
    By
    90 days

    Steps

    1. 1

      Check facts

      Each alternative's own current claims.

      Founder/CEO 1 day

    2. 2

      Build the table

      Seven rows, no spin.

      Founder/CEO 1 day

    3. 3

      When not to buy

      Honest limits today.

      Founder/CEO 2 hours

    4. 4

      Review claims

      Trademarks and wording.

      Outside counsel 3 days

    5. 5

      Publish

      Linked from the FAQ.

      Founder/CTO 2 hours

    6. 6

      Refresh

      Every quarter.

      Founder/CEO quarterly

    First result
    1 month
    Cost
    Staff time
    Owner
    Founder/CEO

    Smallest test

    Add a five-row comparison to the FAQ first.

    Why it works

    Publish a /compare page with a fair table: works in silence, works with lips nearly closed, needs a camera, works offline, works in any app, price, available now. Add a short "when not to buy Subvocal" section, and a "use both" note for dictation-app users (see idea 11).

    Buyers are already making this comparison in public; a fair page wins the search and the argument, and the honesty builds trust.

    Risks and how to handle them

    • Claims go out of date

      Date the page and cite each source.

    Evidence

    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
  • 11

    Show HN commenters asked for an input-only device that works with any agent or app rather than a closed one. The launch is aimed at people who already rely on Wispr Flow and Superwhisper. Those tools run on every major platform and live in buyers' daily workflows. If Subvocal only works inside its own app, buyers have to choose; if it types anywhere, they don't.

    High impactMedium effort First result in 1 quarter Founder/CTO

    First action

    List the top 20 apps from the buyer survey and test text entry in each.

    How to measure it

    Weekly active users in 3+ apps

    Now
    –
    Target
    60%
    By
    60 days after first shipping

    Steps

    1. 1

      Pick platforms

      Mac and Windows first, then phones.

      Founder/CTO 1 day

    2. 2

      Build text entry

      Type into the focused field.

      Founder/CTO 3 weeks

    3. 3

      Trigger behaviour

      Hold to speak, release to type.

      Founder/CTO 1 week

    4. 4

      Test top apps

      From the buyer survey.

      Founder/CTO 1 week

    5. 5

      Publish compatibility

      A list on the FAQ.

      Founder/CEO 1 day

    6. 6

      Measure

      Apps per user per week.

      Founder/CTO weekly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Founder/CTO

    Smallest test

    Make the prototype app paste into the focused field on one Mac.

    Why it works

    Ship a mode in which the companion app types decoded text into whatever field has focus, on Mac, Windows and phone where the platform allows, held by the planned trigger key. Test it against the top 20 apps from the buyer survey and publish a compatibility list. Later, document how to pass Subvocal text into dictation tools' cleanup features.

    Buyers keep their workflow and add silence to it, and tools that look like rivals become reasons to buy.

    Risks and how to handle them

    • Platform permission limits on phones

      Start on desktop, where text entry is allowed.

    Builds on

    Evidence

    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
  • 12

    Show HN commenters with prescriptions, wide noses or narrow bridges asked whether the glasses will fit and work with their lenses; the founders are "quite set on nose pads". IBTimes reports that optional prescription lenses are not included in the price. Glasses wearers can't make a $199 decision, or confirm one, without knowing.

    Medium impactMedium effort First result in 1 month Head of Hardware

    First action

    Add the frame dimensions and a lens-option line to the FAQ.

    How to measure it

    Completed fit profiles

    Now
    –
    Target
    80% of buyers
    By
    Before first shipping

    Steps

    1. 1

      Publish dimensions

      Lens width, bridge, temple length, nose pads.

      Head of Hardware 1 day

    2. 2

      Fit survey

      Send to preorder holders.

      Founder/CEO 1 day

    3. 3

      Pick a lens lab

      Partner for prescription inserts or lenses.

      Head of Hardware 1 month

    4. 4

      Price the option

      Add it to the facts box.

      Founder/CEO 1 week

    5. 5

      Add to FAQ

      Fit and lenses answers.

      Founder/CEO 2 hours

    6. 6

      Collect at balance

      Lens details with the final payment.

      Founder/CTO at shipping

    First result
    1 month
    Cost
    Partner setup costs
    Owner
    Head of Hardware

    Smallest test

    Send the fit survey to 100 buyers.

    Why it works

    Publish the frame's dimensions and nose-pad options, a "Will it fit me?" guide, and the prescription-lens path: partner lab, price range and how to submit a prescription. Send a fit survey to preorder holders and collect lens details at balance time.

    Fit is a classic reason glasses get returned; answering it before shipping protects both conversion and returns.

    Risks and how to handle them

    • Design not final

      Publish ranges and update when the design locks.

    Evidence

    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • HiringHead of Hardware ($125K-$200K): portable prototypes reliable enough for large-scale data collection, miniaturisation into glasses, and ownership of the BOM, DFM/DFT, regulatory preparation, contract manufacturers and the path from EVT to production. Head of Sensing ($150K-$250K): new sensing geometries, calibration across people, sessions and devices, and work with external RF, fabrication, simulation and regulatory partners.Source
  • 13

    The whole site is one homepage pitch plus four short pages. The launch names offices, meetings, coding and walking around, and is aimed at heavy dictation users writing emails, prompts, messages and code. Superwhisper already markets dictation to coding-agent users. Each setting has its own question ("can I use it in an open office?") and no page that answers it.

    Medium impactMedium effort First result in 6 weeks Founder/CEO

    First action

    Wait for the survey results, then pick the top three uses.

    How to measure it

    Deposits from use-case pages

    Now
    –
    Target
    50
    By
    90 days

    Steps

    1. 1

      Pick three

      Top answers from the survey.

      Founder/CEO 1 day

    2. 2

      Film clips

      30 seconds in each real setting.

      Founder/CTO 1 week

    3. 3

      Write pages

      Situation, clip, FAQ, button.

      Founder/CEO 1 week

    4. 4

      Publish and link

      From the homepage and FAQ.

      Founder/CTO 1 day

    5. 5

      Share where they gather

      Developer and remote-work communities.

      Founder/CEO 2 weeks

    6. 6

      Measure

      Visits and deposits per page.

      Founder/CEO monthly

    First result
    6 weeks
    Cost
    Staff time
    Owner
    Founder/CEO

    Smallest test

    Publish one page for the top use and compare its conversion.

    Why it works

    Pick the three most common uses from the buyer survey. For each, build a page with the situation, a 30-second clip filmed in that setting, what people type there, setting-specific FAQs (noise, privacy, latency) and the preorder button.

    A person who sees their own situation solved converts faster, and specific pages match the searches and posts people share.

    Risks and how to handle them

    • Clips overpromise prototype performance

      Film uncut and label prototype footage.

    Builds on

    Evidence

    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
  • 14

    Buyers have paid a refundable $199 and face a wait of several months; The Register calls the product a proof of concept. Delivery depends on hardware milestones the Head of Hardware will own (EVT to production, regulatory preparation) and, per IBTimes, on FCC approval. The only update on the site is dated November 8, and no sitemap page has a last-modified date.

    High impactLow effort First result in 1 month Founder/CEO

    First action

    Write the template and publish the December issue.

    How to measure it

    Refunds per 100 preorders

    Now
    –
    Target
    Under 5
    By
    First batch shipping

    Steps

    1. 1

      Template

      Milestone, accuracy, certification, batch, one question.

      Founder/CEO 2 hours

    2. 2

      First issue

      Published and emailed.

      Founder/CEO 1 day

    3. 3

      Sitemap dates

      Add lastmod for every page.

      Founder/CTO 1 hour

    4. 4

      Answer a buyer

      Pick one real question each month.

      Founder/CEO monthly

    5. 5

      Share

      Post the log on X and LinkedIn.

      Founder/CEO monthly

    6. 6

      Track

      Opens, replies and refunds after each issue.

      Founder/CEO monthly

    First result
    1 month
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Send one update to buyers this month and watch refunds.

    Why it works

    Publish a dated build-log post on the first Monday of each month and email it to buyers and the waitlist: the hardware milestone reached, the latest published error rate (idea 5), certification status, batch timing, and one buyer question answered. Add lastmod dates to the sitemap.

    Silence is what makes preorder buyers ask for refunds; a steady, specific log keeps them invested and gives press and waitlist a reason to come back.

    Risks and how to handle them

    • Bad news in an update

      Say it early and plainly, with the new plan.

    Evidence

    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • HiringHead of Hardware ($125K-$200K): portable prototypes reliable enough for large-scale data collection, miniaturisation into glasses, and ownership of the BOM, DFM/DFT, regulatory preparation, contract manufacturers and the path from EVT to production. Head of Sensing ($150K-$250K): new sensing geometries, calibration across people, sessions and devices, and work with external RF, fabrication, simulation and regulatory partners.Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
  • 15

    Within a week of launch, outlets described the price three ways: $499 in total with $300 due on shipping (The Register), a $699 regular price with US-only preorders (IBTimes), and a $199 reservation with $200 off an unnamed final price (the YC launch). The site has no press page, no fact sheet and no press contact, so each journalist builds the story from whatever they find.

    Medium impactLow effort First result in 2 weeks Founder/CEO

    First action

    Turn the /preorder facts box (idea 1) into a one-page fact sheet.

    How to measure it

    Share of new coverage with correct pricing

    Now
    –
    Target
    90%+
    By
    60 days

    Steps

    1. 1

      Write the sheet

      Ten dated facts.

      Founder/CEO 2 hours

    2. 2

      Gather assets

      Photos, renders and video links.

      Founder/CTO 1 day

    3. 3

      Publish /press

      With a press email.

      Founder/CTO 2 hours

    4. 4

      Footer link

      Visible from every page.

      Founder/CTO 30 minutes

    5. 5

      Send it

      To outlets that covered the launch.

      Founder/CEO 1 day

    6. 6

      Keep it current

      Update with each build log.

      Founder/CEO monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Publish the fact sheet as a section of the launch update first.

    Why it works

    Publish /press with a dated fact sheet (what it is, how it works in two sentences, price and deposit, delivery window, where it ships, team size and backers as disclosed), approved photos of the prototype and renders, the demo video links and a press email. Link it from the footer and send it to outlets planning follow-ups.

    Reporters copy what they can find; giving them one page removes the source of the conflicting numbers and makes follow-up coverage cheaper to win.

    Risks and how to handle them

    • Fact sheet goes stale

      Date it and update it with each monthly log.

    Builds on

    Evidence

    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
  • 16

    Subvocal is a team of three, hiring a Head of ML, a Head of Sensing and a Head of Hardware in San Francisco, and delivery to 750 buyers depends on those hires. The roles are listed only on YC's job board: subvocal.company has no careers page or link. The launch reached over 12 million views on X and 847 points on Hacker News, an audience that includes the RF, ML and hardware engineers the company needs.

    High impactLow effort First result in 2 weeks Founder/CEO

    First action

    Add a "We're hiring" footer link that points to the YC jobs page today.

    How to measure it

    Qualified applicants per Head role

    Now
    –
    Target
    20
    By
    30 days

    Steps

    1. 1

      Footer link

      Point to the YC jobs page today.

      Founder/CTO 30 minutes

    2. 2

      Write /careers

      Three roles and the problems behind them.

      Founder/CEO 1 day

    3. 3

      Show the work

      Prototype photos and a short technical note.

      Founder/CTO 1 day

    4. 4

      Hiring thread

      Post in Hacker News' monthly thread.

      Founder/CEO monthly

    5. 5

      Ask buyers

      Referral request in the build log.

      Founder/CEO monthly

    6. 6

      Track

      Applicants by source.

      Founder/CEO weekly

    First result
    2 weeks
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Add the footer link and count applicants from it.

    Why it works

    Publish /careers with the three roles, the hard problems in plain language (generalising across users, shrinking a lab rig into glasses), the data and prototypes candidates would work with, and the YC apply links. Link it in the footer, mention it in every build log, post in Hacker News' monthly hiring thread, and ask preorder holders for referrals.

    The people most excited by the launch are often the best candidates, and a clear page turns attention into applications at no cost.

    Risks and how to handle them

    • Too many unqualified applicants

      State the experience bar and a short technical question.

    Evidence

    • HiringHead of ML (YC Work at a Startup, $150K-$200K, in person in San Francisco): owns turning noisy physiological signals into continuous language, generalising across users, sessions and devices, adapting to a new user with about 15 minutes of calibration data, and designing held-out user, session and device evaluations. The company page lists a team of 3, backed by Y Combinator and Afore Capital.Source
    • HiringHead of Hardware ($125K-$200K): portable prototypes reliable enough for large-scale data collection, miniaturisation into glasses, and ownership of the BOM, DFM/DFT, regulatory preparation, contract manufacturers and the path from EVT to production. Head of Sensing ($150K-$250K): new sensing geometries, calibration across people, sessions and devices, and work with external RF, fabrication, simulation and regulatory partners.Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 17

    The Register reports the early-bird deal as $499 in total: $199 upfront and $300 due on shipping. About $150,000 in refundable deposits sits on roughly 750 orders, but the real revenue depends on how many buyers pay the balance months from now. The terms page, as published, says it covers "final payment" and "cancellations", yet shows none of that text.

    High impactLow effort First result in At first shipping Founder/CEO

    First action

    Check how deposits were taken and whether a card is on file for the balance.

    How to measure it

    Deposits that pay the balance

    Now
    –
    Target
    90%
    By
    First batch shipping

    Steps

    1. 1

      Confirm the terms

      When, how and how much, in the published terms.

      Founder/CEO 2 days

    2. 2

      Check payment setup

      Saved cards and the charge flow.

      Founder/CTO 1 day

    3. 3

      Early balance option

      Optional, with a non-price perk.

      Founder/CEO 1 week

    4. 4

      Reminder sequence

      30, 7 and 1 day before the charge.

      Founder/CEO 1 day

    5. 5

      Failed payments

      Card update link and retries.

      Founder/CTO 2 days

    6. 6

      Measure

      Conversion by batch.

      Founder/CEO per batch

    First result
    At first shipping
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Email 50 buyers offering an optional early balance payment and see who takes it.

    Why it works

    Decide and publish exactly when and how the balance is charged, with notice ahead of time. Offer an optional early balance payment with a non-price perk (for example, priority fit help), run a reminder sequence before the charge, add a card-update flow and a gentle failed-payment path, and track deposit-to-balance conversion by batch.

    Preorder businesses lose more revenue to expired cards and surprise charges than to people changing their minds; planning it now keeps the $499 orders whole.

    Risks and how to handle them

    • Changing terms after purchase

      Only add optional choices; keep the original terms for everyone.

    Builds on

    Evidence

    • PressThe Register (November 2, 2026) calls it a proof of concept: an early-bird price of $499 in total, $199 upfront and $300 due on shipping; shipping expected in Q1 2027; full retail price not announced; accuracy rate, battery life and app pairing not disclosed. A YC partner needed to read only 20 sentences to calibrate the prototype.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
    • Your websitePreorder page, read twice on November 10, 2026: the readable text is the heading "Preorder Subvocal" and one sentence promising "preorder details, and answers to common questions about the first production batch", about 19 words in all. No price, deposit amount, delivery date, FAQ or structured data appear. /preorder-terms likewise shows only a line saying it covers "refundable deposits, final payment, cancellations, and estimated delivery", and /privacy only a heading and a one-line introduction.Source
  • 18

    The YC launch asks people to send it to "the most voice-dependent people you know", but there is no way to do that beyond copying a URL, and no way to know who did. About 750 buyers are the most enthusiastic advocates Subvocal has, and many know other heavy dictation users.

    Medium impactLow effort First result in 2 weeks Founder/CEO

    First action

    Generate a unique link for each order.

    How to measure it

    Deposits from referral links

    Now
    –
    Target
    150
    By
    60 days

    Steps

    1. 1

      Create links

      One per order.

      Founder/CTO 1 day

    2. 2

      Choose rewards

      Non-price perks only.

      Founder/CEO 1 day

    3. 3

      Email buyers

      Link, reward and a short message to forward.

      Founder/CEO 1 day

    4. 4

      Track

      Deposits per link.

      Founder/CTO weekly

    5. 5

      Thank referrers

      In the build log.

      Founder/CEO monthly

    6. 6

      Review

      Keep or change rewards after 60 days.

      Founder/CEO 60 days

    First result
    2 weeks
    Cost
    Small perk costs
    Owner
    Founder/CEO

    Smallest test

    Send links to 100 buyers and count referred deposits.

    Why it works

    Give every buyer a personal link. Each referred deposit earns the referrer a non-price reward, such as early access to the companion app, a calibration session with the team, or a name in the build log. Don't use queue jumps, so batch order stays fair (idea 2). Show referral counts in the buyer email.

    People who just bought want others to join them; a simple link and a reward that costs little turns that into tracked demand.

    Risks and how to handle them

    • Spammy sharing

      Reward deposits, not clicks.

    Builds on

    Evidence

    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source

Big bets 3

Larger plays that can change the trajectory.

  • 19

    Cloud decoding is the most frequent concern in the Show HN thread, and the founders hope to offer local inference "at some point in the future". Superwhisper already sells dictation that works offline, and AlterEgo leads on what its device won't pick up. For a device that reads facial signals, where decoding happens may decide who buys.

    High impactHigh effort First result in 2 quarters Head of ML

    First action

    Measure today's model size and the latency on a current laptop.

    How to measure it

    Users decoding locally

    Now
    –
    Target
    Available to all users by batch 2
    By
    3 quarters

    Steps

    1. 1

      Measure

      Model size, compute and latency today.

      Head of ML 2 weeks

    2. 2

      Distil

      A smaller model within an accuracy budget.

      Head of ML 1 quarter

    3. 3

      Laptop first

      Local mode in the desktop app.

      Founder/CTO 1 quarter

    4. 4

      Verify

      Independent review of what leaves the device.

      Founder/CTO 1 month

    5. 5

      Launch

      Update /data, /compare and the FAQ.

      Founder/CEO 1 week

    6. 6

      Phone next

      Port when hardware allows.

      Head of ML 2 quarters

    First result
    2 quarters
    Cost
    Significant ML engineering time
    Owner
    Head of ML

    Smallest test

    Run the current model offline on one laptop and time it.

    Why it works

    Set a goal of a local decoding mode on the user's laptop (then phone) by the second batch: a smaller distilled model, a latency target, and a clear switch in the app. Publish independent verification of what leaves the device. Make it the privacy headline on /data and /compare.

    It turns the loudest objection into the main reason to choose Subvocal over cloud-first rivals, and it matters most to the professional users the launch targets.

    Risks and how to handle them

    • Lower accuracy locally

      Let users choose, and show the accuracy of each mode.

    Builds on

    Evidence

    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
    • HiringHead of ML (YC Work at a Startup, $150K-$200K, in person in San Francisco): owns turning noisy physiological signals into continuous language, generalising across users, sessions and devices, adapting to a new user with about 15 minutes of calibration data, and designing held-out user, session and device evaluations. The company page lists a team of 3, backed by Y Combinator and Afore Capital.Source
  • 20

    Preorders are reported as US-only, and the founders told a European commenter "Not immediately, no", citing separate RF approvals. The Head of Hardware post includes regulatory preparation, and the Head of Sensing works with regulatory partners. The launch got global coverage, but the company has no data on where demand outside the US is strongest.

    High impactHigh effort First result in 2 quarters Head of Hardware

    First action

    Add the country field to the waitlist and read the counts after 30 days.

    How to measure it

    Signups in the first non-US market

    Now
    –
    Target
    1,000
    By
    2 quarters

    Steps

    1. 1

      Rank countries

      From waitlist data.

      Founder/CEO 1 month

    2. 2

      Ask a test lab

      The approval path and costs for that market.

      Head of Hardware 2 weeks

    3. 3

      Design for it

      Build the requirements into EVT.

      Head of Sensing 1 quarter

    4. 4

      Budget

      Testing, certification and logistics.

      Founder/CEO 2 weeks

    5. 5

      Local waitlist page

      In the market's language if needed.

      Founder/CTO 1 week

    6. 6

      Open preorders

      Once approval is clear.

      Founder/CEO when approved

    First result
    2 quarters
    Cost
    Certification and test-lab fees
    Owner
    Head of Hardware

    Smallest test

    Get one test-lab quote for the top waitlist country.

    Why it works

    Use waitlist country counts (idea 4) to pick one market. Ask a test lab what approval path it needs, build those requirements into the hardware now rather than after the US launch, budget for it, and open a local waitlist page that converts to preorders once approval is clear.

    Designing for a second market's approval during EVT is far cheaper than redesigning afterwards, and a ready waitlist means revenue on day one.

    Risks and how to handle them

    • Distraction from the US launch

      Only design for it now; spend on certification after US approval.

    Builds on

    Evidence

    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • PressIBTimes (November 4, 2026): regular price $699 and an early price of $499 with a refundable $199 deposit; preorders US-only; deliveries expected from Q1 2027, dependent on FCC approval; shipping, taxes and optional prescription lenses not included.Source
    • HiringHead of Hardware ($125K-$200K): portable prototypes reliable enough for large-scale data collection, miniaturisation into glasses, and ownership of the BOM, DFM/DFT, regulatory preparation, contract manufacturers and the path from EVT to production. Head of Sensing ($150K-$250K): new sensing geometries, calibration across people, sessions and devices, and work with external RF, fabrication, simulation and regulatory partners.Source
    • Your websiteSitemap (checked November 10, 2026): five URLs (/, /preorder, /preorder-terms, /privacy, /launch-update), none with a last-modified date. There is no blog, FAQ, about, contact, press or careers page, and no newsletter or waitlist signup; the only published contact is one founders' email address on the privacy page.Source
  • 21

    Show HN commenters want a device that feeds any agent or app. The launch lists prompts and commands alongside emails and code as outputs, and the founders describe a trigger key rather than a wake word. Typing into text fields (idea 11) covers writing; an SDK would let developers use silent commands, confidence scores and the trigger itself.

    High impactHigh effort First result in 2 quarters Founder/CTO

    First action

    Ask buyers in the survey whether they'd build on an SDK.

    How to measure it

    Published third-party integrations

    Now
    –
    Target
    10
    By
    2 quarters after first shipping

    Steps

    1. 1

      Define the API

      Text stream, trigger, confidence, commands.

      Founder/CTO 2 weeks

    2. 2

      Pick ten developers

      From preorder holders who code.

      Founder/CEO 1 week

    3. 3

      Private beta

      Weekly feedback calls.

      Founder/CTO 1 quarter

    4. 4

      Docs

      Quickstart and reference.

      Founder/CTO 1 month

    5. 5

      Samples

      Editor, notes and assistant integrations.

      Founder/CTO 1 month

    6. 6

      Showcase

      Feature builds in the build log.

      Founder/CEO monthly

    First result
    2 quarters
    Cost
    Engineering time
    Owner
    Founder/CTO

    Smallest test

    Expose the decoded text stream locally to one developer.

    Why it works

    Define a small API (a text stream, trigger events, confidence and a command mode), run a private beta with ten developers from the preorder list, publish docs and three sample integrations (a code editor, a notes app, an assistant), and showcase what they build in the build log.

    Developers who build for a device make it more useful for everyone and become its loudest advocates; that's how input hardware turns into a platform.

    Risks and how to handle them

    • Distracts from core accuracy work

      Keep the API small and start after keyboard mode ships.

    Builds on

    Evidence

    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Hacker NewsShow HN (November 2, 2026; 847 points, 395 comments): the post describes a cross-user model trained on data from paid participants, a companion app that decodes in the cloud, a Q1 2027 ship target and preorders for 500 units with a $200 discount. The most frequent concern in the thread is cloud decoding of silent speech; the founders reply that they "have built many safeguards to prevent even us from accessing the raw data in the cloud" and hope for local inference "at some point in the future".Source

Wildcards 2

Unexpected experiments, each with a cheap way to test it.

  • 22

    Subvocal's pitch rests on a number: 200+ words per minute, decoded silently. Wispr Flow sells with a similar one, 220 wpm speaking against 45 typing. Typing has speed tests and leaderboards; silent speech has none yet, and the launch already showed how well this product travels on X.

    Medium impactLow effort First result in 1 month Founder/CEO

    First action

    Record the team's own silent wpm and accuracy on one passage.

    How to measure it

    Monthly challenge entries

    Now
    –
    Target
    100
    By
    2 months after beta starts

    Steps

    1. 1

      Write passages

      Fresh text each month.

      Head of ML 1 day

    2. 2

      Scoring page

      wpm and accuracy from the app.

      Founder/CTO 1 week

    3. 3

      Leaderboard

      Opt-in names, median shown.

      Founder/CTO 2 days

    4. 4

      Launch

      Announce on X and in the build log.

      Founder/CEO 1 day

    5. 5

      Clip the best

      Short videos of top runs.

      Founder/CEO monthly

    6. 6

      Repeat

      Monthly, same conditions.

      Founder/CEO monthly

    First result
    1 month after beta
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Post the team's own scores as the first entries.

    Why it works

    Once beta users have devices, run a monthly challenge: silently read a fresh passage and get words per minute and accuracy, with a leaderboard (opt-in names) and short clips of the best runs. Publish the median as well as the top scores, under the same conditions as the accuracy video (idea 5).

    A number people can beat is shareable, and publishing the median keeps it honest, which builds the credibility the product needs.

    Risks and how to handle them

    • Scores look worse than the claim

      Publish conditions and the trend; update the claim if needed.

    Builds on

    Evidence

    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Competitor researchRivals' and adjacent tools' own homepages: AlterEgo, an MIT-developed wearable with "Silent Sense" for silent speech, still in development with a "Get Notified" signup, says it responds only to deliberate silent speech. Wispr Flow, "Don't type, just speak", is free for 2,000 words a week, claims 220 wpm against 45 for typing, and says it can handle speech as quiet as a whisper with the mic close to the mouth. Superwhisper, free tier, works offline on Mac, Windows, iOS, Android and Linux, with 30-day refunds.Source
    • Your websiteLaunch update (November 8, 2026): "roughly 750 paid preorders" and "about $150,000 in refundable deposits" since launching on October 30; the launch "reached over 12 million views on X"; "We're working toward our first production run."Source
  • 23

    The problem Subvocal solves is places where you can't speak: open offices, cafes, transit, and the libraries and quiet offices Show HN commenters mentioned. The team works in person in San Francisco. Hardware demos usually happen at loud events, but this product is most convincing in the quietest room in the building.

    Medium impactLow effort First result in 1 month Founder/CEO

    First action

    Ask two coworking spaces near the office for a quiet-zone slot.

    How to measure it

    Signups per session

    Now
    –
    Target
    30
    By
    3 sessions

    Steps

    1. 1

      Book spaces

      Two coworking quiet zones.

      Founder/CEO 1 week

    2. 2

      Prep the demo

      A reliable rig and a calibrated demonstrator.

      Founder/CTO 1 week

    3. 3

      Signup flow

      QR to the waitlist and /preorder.

      Founder/CTO 1 hour

    4. 4

      Run the session

      90 minutes, silent demos.

      Founder/CEO monthly

    5. 5

      Film clips

      For use-case pages and X.

      Founder/CEO each session

    6. 6

      Recruit participants

      Invite locals to data sessions.

      Head of ML each session

    First result
    1 month
    Cost
    Space fees, if any
    Owner
    Founder/CEO

    Smallest test

    One session in one coworking space this month.

    Why it works

    Book monthly 90-minute sessions in coworking quiet zones and phone booths. Let visitors try the prototype (or watch a team member type silently next to them), collect waitlist and preorder signups on the spot, and film short clips for the use-case pages (idea 13).

    Seeing someone type without a sound, in a room where talking is banned, is the proof no video fully gives, and it also recruits local data participants.

    Risks and how to handle them

    • Prototype fails live

      Use a calibrated team member and show the error rate openly.

    Evidence

    • Launch YCLaunch YC post: a $199 reservation secures one of the first units with founders pricing and earliest shipping, plus a $200 early-bird discount on the final price; reservations are limited. It describes voice input failing in open offices, cafes and transit, where people try gooseneck whisper mics and soundproof masks, and targets heavy dictation users (it names Wispr Flow and Superwhisper) writing emails, prompts, messages, code and notes. It claims 200+ words per minute with a "reasonably low error rate" and plans thousands of hours of population-scale RF data so a model needs only short calibration. Its asks: reserve a unit, share where you would use it, and send the launch to the most voice-dependent people you know.Source
    • Hacker NewsShow HN thread requests: commenters cite a roughly 5% error rate in the demo video and 500ms latency (founders: "unoptimized"); ask about prescription lenses and nose-bridge fit (founders are "quite set on nose pads"); want an input-only device that works with any agent or app; ask about Europe (founders: "Not immediately, no", citing separate RF approvals) and battery life; compare it with AlterEgo, Cornell research and lip-reading projects; and suggest uses in libraries, offices and noisy places.Source
    • HiringHead of ML (YC Work at a Startup, $150K-$200K, in person in San Francisco): owns turning noisy physiological signals into continuous language, generalising across users, sessions and devices, adapting to a new user with about 15 minutes of calibration data, and designing held-out user, session and device evaluations. The company page lists a team of 3, backed by Y Combinator and Afore Capital.Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • Operation Autopilot growth strategyAutonomous cleaning and facility robotics delivered as an all-inclusive subscription: site assessment, deployment, monitoring, maintenance…22 growth ideas
  • Anduril Industries growth strategyA defense technology company building autonomous systems across air, land, sea and space, all connected by Lattice, its AI-powered command…21 growth ideas
  • Inferact growth strategyThe company founded by vLLM's creators and maintainers: a production inference platform for open models, run in the customer's cloud or…21 growth ideas
  • Power Digital growth strategyA growth marketing agency organised in vertical divisions (B2B, Consumer Services, CPG, Fashion, Healthcare via Cardinal, Lifestyle) with…21 growth ideas
  • Timeless Vapes growth strategyAn Arizona cannabis brand founded in 2013 (Timeless Refinery, Timeless Select LLC) making vape cartridges, T1 and T2 all-in-ones, NOIR…21 growth ideas
  • Brumby growth strategyAutonomous drones and a phone app for cattle properties: the drone musters and moves herds, flies scheduled checks, counts animals…22 growth ideas
  • Fashion Nova growth strategyA Los Angeles trend-to-market fashion brand selling affordable, size-inclusive clothing online for women, plus, men and kids, plus beauty…22 growth ideas
  • Mintel growth strategyA global market intelligence company: consumer research, the GNPD product launch database, market data, expert analysis, AI tools (Leap…21 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on November 10, 2026, and cites them throughout. The Subvocal Company did not commission, review or endorse it, and Programmable is not affiliated with The Subvocal Company. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how The Subvocal Company runs its business.

Something wrong, or do you represent The Subvocal Company? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 308 ideas

    One source of truth: the preorder facts and FAQ, batch positions, the device-data page, a waitlist, an accuracy video, a press fact sheet, the buyer survey and a lighter hero video.

  2. Days 31 to 6012 ideas

    Use the buyers: paid data sessions, calibration-first onboarding, fit and lenses, the balance plan, referrals and a build log; publish the comparison and use-case pages, the careers page and keyboard mode; run quiet-room demos and the speed challenge.

  3. Days 61 to 903 ideas

    Commit to the big bets: on-device decoding, a first market outside the US and a developer SDK.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value dictating where talking out loud isn't possible. The most cited problem: voice input is awkward in open offices, cafes and transit.

  • A new interface that feels like science fiction
  • Not knowing whether it works today, or how accurately
  • In their words: “talk without talking”

Market

Competitive landscape

4 competitors compared: AlterEgo, Wispr Flow, Superwhisper and Whisper mics and masks.

  • Gap to exploit: Truly silent text input that works with closed or barely moving lips
  • AlterEgo: Not on sale, no price
  • Wispr Flow: Needs audible speech, or a whisper with a close mic

Hiring signals

A team of three is hiring the three leaders it needs to turn a lab prototype into shipped glasses.

  • Head of ML: Cross-user generalisation and 15-minute calibration. Ideas 5, 8, 9, 19.
  • Head of Hardware: Data-collection rigs, EVT to production, regulatory prep. Ideas 8, 12, 14, 20.
  • Head of Sensing: New sensing designs and regulatory partners. Ideas 16, 20.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence