ProgrammableMarketing/armature.tech (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 3
  • How we can help
Generated Nov 10, 2026
  1. Programmable
  2. Playbooks
  3. Armature
Armature logo

growth playbook · Nov 10, 2026

armature.tech (opens in a new tab)

Armature growth playbook

Diagnosis

Armature is a three-person YC P26 team in Paris with a sharp idea (be the tool coding agents pick) and a big research asset: 43 leaderboards, about 291 "Do coding agents recommend X?" pages, a study with 300 points on Hacker News, and agent.reviews.

Its own GTM post names the gap: research brings customers, but there is no repeatable way to find the next ones. The site rarely asks for anything. Vendor pages end in generic links, leaderboards have no 'add my tool', there is no newsletter, and the eight vendor chapters sit behind 'request early access'.

There is no published customer result either. This playbook starts by turning the research into a front door for each vendor, adding capture and proof, then building automated outbound, a fixed-price audit, agent-readable pages of its own, and a self-serve pick-rate monitor.

Growth Score

7/10

Biggest Opportunity

Every vendor named in the library and leaderboards is a warm lead looking at its own numbers. Giving each one a way to ask why it loses turns published research into a repeatable pipeline.

Biggest Risk

The category is new and crowded at once (Gauge, Amplifying.ai, AI visibility trackers such as Profound and Peec AI), and developers on Hacker News questioned the business model. Trust in the method has to grow as fast as the sales motion.

01

Start here

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

  1. 1

    Turn the 291 "Do coding agents recommend X?" pages into each vendor's front door, with its own lost sessions and a one-click request for the full breakdown

    Hand-add an "Is this your product?" form to the ten library pages with the most traffic this week.

    High impact First result in 2 weeks
  2. 2

    Publish a first real before-and-after, anonymised if necessary, with the recorded runs behind it

    Ask the customer with the largest pick-rate gain whether they can be named.

    High impact First result in 1 month
  3. 3

    Add "Add my tool" and "Request this category" to every leaderboard, and let the requests steer the next study

    Put a two-field "Add my tool" form under the payments board.

    Medium impact First result in 2 weeks

02

Your idea portfolio

22 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.

  • 9 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 · 5Big bets · 5Fill-ins · 6Reconsider · 6
Effort →

Showing 22 of 22 ideas

This week 6

Quick wins with a clear first action.

  • 01

    Armature has published about 291 library pages. Each one asks whether coding agents recommend a named product and answers with its install share, its rank, splits by agent and persona, and how many sessions considered it but picked a rival. The page for the payments leader shows 75 such sessions. The person most likely to read a page like that works at the vendor it names. Yet every page ends with the same three generic links ("See the service", "Read the results", "All library pages"). There is no form, no email capture and no link to the 45-check readiness scan. The GTM job post says research brings customers today but there is no repeatable way to find the next ones. These pages, already published and indexed, can be that repeatable way.

    High impactLow effort First result in 2 weeks Founding GTM

    First action

    Hand-add an "Is this your product?" form to the ten library pages with the most traffic this week.

    How to measure it

    Library-page report requests

    Now
    –
    Target
    30 a month
    By
    60 days

    Steps

    1. 1

      Pick the fields

      Install share, rank, considered-but-not-chosen count and the top rival, from the existing page data.

      Founding Engineer, Product 1 day

    2. 2

      Write the block

      One line showing the vendor's own loss, one promise, one work-email field.

      Founding GTM 1 day

    3. 3

      Build the template

      Generate the block on all of the roughly 291 vendor pages and on every leaderboard row.

      Founding Engineer, Product 3 days

    4. 4

      Prepare the report

      A private page with the losing runs, the winning rival and the docs the agents read, drawn from recorded traces.

      Founding Engineer, Applied AI Research 1 week

    5. 5

      Route requests

      Match work-email domains to vendors; low-share vendors reach a founder within a day.

      Founding GTM 2 days

    6. 6

      Follow up

      Offer a call built on the report that leads into the week-1 audit.

      CEO (co-founder) ongoing

    7. 7

      Measure

      Requests, calls and Growth contracts by page, category and share band.

      Founding GTM monthly

    First result
    2 weeks
    Cost
    Engineering time only; the data already exists
    Owner
    Founding GTM

    Smallest test

    Hand-add the block to the five lowest-share vendors in one category and see who asks.

    Why it works

    Generate a block on every vendor page from the data the page already has: "Is this your product? Agents considered <product> in <N> sessions and chose someone else. See every losing run: which rival won, for which persona, and which docs the agent read." A single work-email field, with product and category prefilled, sends the vendor a private report built from the traces Armature has already recorded, and offers a call. Put the lowest-share and most-lost vendors first in line for a founder. Add the same block to every vendor row on the 43 leaderboards.

    The reader arrives with their own numbers on screen, the strongest opening a $5,000-a-month service can get. The traces already exist, so each report takes minutes, and the work-email domain tells Armature exactly which vendor is asking.

    Ready to use

    ChecklistVendor block for library and leaderboard pages

    [ ] Headline: "Is this your product?"
    [ ] Their line: "Agents considered <product> in <N> sessions and chose <top rival>"
    [ ] Promise: "See every losing run, the persona and the docs the agent read"
    [ ] One work-email field, product and category prefilled
    [ ] Private report page built from recorded traces
    [ ] Link to the 45-check readiness scan
    [ ] Same block on each leaderboard row
    [ ] Tracking by page, category and share band

    Risks and how to handle them

    • Vendors dislike being profiled

      Leave the public numbers as they are and send traces only to work emails at the vendor's own domain.

    • Free reports replace paid audits

      Share where the vendor loses and keep the fixes for the audit.

    Evidence

    • Your websiteLibrary page for the payments leader (published October 14, 2026): install share 88% (578 of 657 judged sessions), rank 1 of 9, 75 sessions in which agents considered it and did not choose it, with splits by agent and by persona (junior developer 100%, enterprise team 83%, senior engineer 77%); closing links "See the service", "Read the results" and "All library pages"; no form, no email capture, no link to the readiness scan, no author.Source
    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
    • Your websitePayments leaderboard (updated September 29, 2026): 657 runs by five coding agents on 11 codebases; nine providers ranked, the bottom four with 1 to 3 wins each; links to an interactive board showing every run's session, diff and judge verdict, and a Markdown copy; no "is your product here?" prompt and no email capture.Source
  • 02

    The leaderboards index has a single "Evaluate my product" button. There is no way to add a missing tool or ask for a category. On Hacker News, readers did exactly that in the comments, naming tools and categories they wanted covered. The research engineer is hired to decide what Armature studies next.

    Medium impactLow effort First result in 2 weeks Founding Engineer

    First action

    Put a two-field "Add my tool" form under the payments board.

    How to measure it

    Leaderboard requests with a work email

    Now
    –
    Target
    40 a month
    By
    60 days

    Steps

    1. 1

      Design the forms

      Two fields each, with a work email required.

      Founding Engineer, Product 1 day

    2. 2

      Ship on all boards

      The index and all 43 leaderboard pages.

      Founding Engineer, Product 2 days

    3. 3

      Show demand

      A public count per requested category.

      Founding Engineer, Product 1 day

    4. 4

      Rank the queue

      Order the next studies by requests and contract value.

      Founding Engineer, Applied AI Research weekly

    5. 5

      Notify requesters

      Email on the day the board goes live, with their result.

      Founding GTM per launch

    6. 6

      Qualify

      Vendors that request get the report offer from idea 1.

      Founding GTM ongoing

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Founding Engineer, Applied AI Research

    Smallest test

    Add the form to three boards and count requests for two weeks.

    Why it works

    Add two short forms on each leaderboard and on the index. "Add my tool" asks for the product, category and a work email. "Request this category" asks for the category and why it matters. Show the request count publicly ("14 vendors asked for this"), study the most-requested categories first, and email requesters the day their board goes live.

    A vendor asking to be added has said it cares about its pick rate, which is the best possible qualification for the Growth service. Public request counts also give the research roadmap a visible, demand-led order.

    Risks and how to handle them

    • Too many requests to study

      Publish the order and the threshold for a new category.

    Evidence

    • Your websiteLeaderboards index: one "Evaluate my product →" button under "Get featured by coding agents"; no form to add a product or request a category; the results area reads "Loading leaderboard results" to a reader that does not run JavaScript.Source
    • Hacker NewsHacker News thread on the study: 300 points and 153 comments; praise for the data ("The data was cool."), criticism of the business model of selling agent recommendations, complaints about pop-ups and the mobile layout (the author replied the pop-ups were fixed), and readers naming tools and categories they wanted covered.Source
    • HiringFounding Engineer, Applied AI Research: runs the studies behind the public leaderboards, published as open leaderboards with every trace available; publishes findings under their own name and decides what Armature studies next; turns customer questions into better evals.Source
  • 03

    Armature publishes new leaderboards weekly (the latest on November 9, 2026), and its study drew 300 points on Hacker News. Yet the study page has no signup, only an email address for suggesting questions. The site links no newsletter or RSS feed, so every reader who leaves has to come back on their own.

    High impactLow effort First result in 2 weeks Founding GTM

    First action

    Put a signup box at the end of the study post.

    How to measure it

    Research email subscribers

    Now
    –
    Target
    1,500
    By
    90 days

    Steps

    1. 1

      Pick the tool

      A simple email tool with tags for role and category.

      Founding GTM 1 day

    2. 2

      Add signups

      Blog, leaderboards and the library page template.

      Founding Engineer, Product 1 day

    3. 3

      Add an RSS feed

      For new boards and posts.

      Founding Engineer, Product 1 day

    4. 4

      Write issue one

      The three most surprising findings since September.

      Founding Engineer, Applied AI Research 2 days

    5. 5

      Send every two weeks

      One finding, one board, one vendor lesson.

      Founding GTM biweekly

    6. 6

      Track

      Vendor-tagged subscribers who book calls.

      Founding GTM monthly

    First result
    2 weeks
    Cost
    Email tool, under $50 a month
    Owner
    Founding GTM

    Smallest test

    One inline box on the study post for two weeks.

    Why it works

    Add a one-field signup on the blog, the leaderboards and every library page: "New leaderboards and findings, every two weeks." Each issue carries one finding, one new board and one vendor lesson. Ask for a role and category at signup so vendor readers can be told apart and get the offer from idea 1.

    Research is the channel that already works. An owned list turns one-off traffic spikes into an audience the GTM hire can reach on every launch.

    Risks and how to handle them

    • Pop-ups annoy developers, as on Hacker News

      Use an inline box only, never a modal.

    Evidence

    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • Your websiteChannels: the site links a LinkedIn company page and Y Combinator; the X account in its links is a founder's personal profile; no GitHub organisation, YouTube channel, newsletter or RSS feed is linked.Source
    • Hacker NewsHacker News thread on the study: 300 points and 153 comments; praise for the data ("The data was cool."), criticism of the business model of selling agent recommendations, complaints about pop-ups and the mobile layout (the author replied the pop-ups were fixed), and readers naming tools and categories they wanted covered.Source
  • 04

    The Publications page lists eight chapters. Part II is written for vendors: "Measure where you stand", "Publish the right artifacts", "Differentiate", "Survive execution". All eight are marked "Soon" behind "Request early access", and the only routes in are booking a call or sending an email. The GTM hire is asked to turn each study into a lead magnet. These chapters already are one, just without a form.

    High impactLow effort First result in 2 weeks Founding GTM

    First action

    Publish "Measure where you stand" as a draft behind a three-field form.

    How to measure it

    Chapter requests from vendors

    Now
    –
    Target
    100
    By
    60 days

    Steps

    1. 1

      Finish one chapter

      "Measure where you stand" as a dated draft.

      Founding Engineer, Applied AI Research 1 week

    2. 2

      Build the form

      Work email, product, category.

      Founding Engineer, Product 1 day

    3. 3

      Add the appendix

      The category's leaderboard numbers on one page.

      Founding Engineer, Applied AI Research 2 days

    4. 4

      Send in a day

      Chapter plus appendix, with a call link.

      Founding GTM ongoing

    5. 5

      Release the rest

      One chapter every two weeks to the same list.

      Founding Engineer, Applied AI Research biweekly

    6. 6

      Measure

      Requests, calls and audits by chapter.

      Founding GTM monthly

    First result
    2 weeks
    Cost
    Writing time
    Owner
    Founding GTM

    Smallest test

    Email one finished chapter to the people who already asked for early access.

    Why it works

    Replace "Request early access" with a short form: work email, product, category. Send the first draft chapter within a day, with a one-page appendix of that category's own leaderboard numbers. Release the next chapters to the same list, so it becomes the vendor segment of the research email.

    A vendor that asks for "What a vendor should do" has named itself a buyer. Adding the category appendix makes the chapter about their market, which is the shortest path to a call.

    Risks and how to handle them

    • Drafts look unfinished

      Label them as drafts with a date and a changelog.

    Evidence

    • Your websitePublications page: eight upcoming chapters in two parts, Part II "What a vendor should do" (Measure where you stand, Publish the right artifacts, Differentiate, Survive execution), all marked "Soon" behind "Request early access"; drafts and early results go to vendors who ask; the CTAs are a call booking or an email.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
  • 05

    The homepage shows "2% → 11%" and the discoverability page shows "11% to 23%". Both are presented as examples. The site has no case studies, no customer names and no testimonials, and the only mark of backing is "Backed by Y Combinator". A buyer paying from $5,000 a month for a new category needs proof that it works, and Armature already records the agent runs behind every monthly report.

    High impactMedium effort First result in 1 month Founding Forward Deployed Eng…

    First action

    Ask the customer with the largest pick-rate gain whether they can be named.

    How to measure it

    Call-to-contract rate

    Now
    –
    Target
    +25%
    By
    90 days

    Steps

    1. 1

      Pick the result

      The largest, cleanest movement in a known category.

      Founding Forward Deployed Engineer 1 day

    2. 2

      Get consent

      Named, or anonymised with the category.

      CEO (co-founder) 1 week

    3. 3

      Write it

      Baseline, changes shipped, result and traces.

      Founding Forward Deployed Engineer 3 days

    4. 4

      Check the method

      Same panel, same prompts, dated runs.

      Founding Engineer, Applied AI Research 2 days

    5. 5

      Place it

      Homepage, discoverability and pricing.

      Founding Engineer, Product 1 day

    6. 6

      Use in calls

      The first slide of every discovery call.

      Founding GTM ongoing

    First result
    1 month
    Cost
    Staff time only
    Owner
    Founding Forward Deployed Engineer

    Smallest test

    Replace the example numbers on /discoverability with one dated real result.

    Why it works

    Take one customer's movement from the first audit to month two. Publish the category, the baseline pick rate, the changes shipped (docs, SDK guide, listing), the new pick rate and links to sample traces. Name the customer if they agree. If not, describe it as "a US developer tool in <category>". Put the result on the homepage, the discoverability page and the pricing page.

    Armature's whole pitch is measured evidence, so its own proof should be measured too. Recorded runs make an anonymised story believable.

    Risks and how to handle them

    • The customer won't be named

      An anonymised story with real traces still beats illustrations.

    Evidence

    • Your websiteDiscoverability page: Measure, Find changes, Create and ship, Report; first four weeks Audit, First content, First measurement, Report and plan; a monthly report "with the recorded agent runs behind each claim"; figures such as "11% to 23%" and "9 / 32 runs pick you" are shown as examples; no case studies, customer names or testimonials; "Tell us about your product, and we will show you where you stand today."Source
    • Your websiteHomepage: H1 "Be the tool" followed by the rotating names of three coding agents; two offers, Agent discoverability ("Service run with you") and Agent usability (self-serve, "Start for free"); "Backed by Y Combinator"; no customer logos or testimonials; a sample "Your pick rate on the panel: 2% → 11%"; the FAQ counts a pick when the agent installs the product and wires it into the repo, not when it only mentions it.Source
    • HiringFounding Forward Deployed Engineer: builds the customer operations function; designs, runs and presents experiments measuring how often coding agents choose each customer's product; owns customer relationships and the results promised to them; automates delivery with AI agents, one repeated task at a time.Source
  • 06

    /compare names Gauge, Lightsage and Firegoat as agent-led growth tools. But all eight library "Armature vs" pages cover AI visibility trackers, SEO suites or a devrel hire, and none covers Gauge or Amplifying.ai, which also test what coding agents choose. The Profound page calls Profound "one of the strongest products in the AI visibility category" and then never links to /compare.

    Medium impactLow effort First result in 1 month Founding GTM

    First action

    Draft "Armature vs Gauge" from Gauge's own launch pages and the /compare rows.

    How to measure it

    Comparison-page calls

    Now
    –
    Target
    8 a month
    By
    90 days

    Steps

    1. 1

      Check rivals' claims

      Read each rival's current pages.

      Founding GTM 1 day

    2. 2

      Write two pages

      Gauge and Amplifying.ai, using the /compare rows.

      Founding GTM 3 days

    3. 3

      Cross-link

      Every "Armature vs" page to /compare and back.

      Founding Engineer, Product 1 day

    4. 4

      Add CTAs

      The readiness scan and a call on each page.

      Founding Engineer, Product 1 day

    5. 5

      Date them

      An updated date and sources on each page.

      Founding GTM 1 hour

    6. 6

      Refresh

      Recheck rival pages every quarter.

      Founding GTM quarterly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Founding GTM

    Smallest test

    Add the cross-links today and watch /compare visits.

    Why it works

    Write fair "Armature vs Gauge" and "Armature vs Amplifying.ai" pages using the /compare rows (tests coding agents, buyer-like codebases, explains why a rival wins, writes and ships fixes, checks installs). Link every "Armature vs" page to /compare, and link /compare to each of them. Date the pages and cite each rival's own pages.

    Buyers who already know the category compare the specialists, not the AI visibility trackers. These are the highest-intent searches Armature can win.

    Risks and how to handle them

    • Claims about rivals go stale

      Cite their own pages and date every claim.

    Evidence

    • Your website/compare: Armature against AI visibility trackers (Profound, Peec AI, AthenaHQ), agent-led growth tools (Gauge, Lightsage, Firegoat), AI content platforms, content agencies and in-house hires; rows such as "Tests coding agents, not only chat", "Writes and ships the fixes" and "Checks what agents install in the code".Source
    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • Your website"Armature vs Profound" (updated September 24, 2026) calls Profound "one of the strongest products in the AI visibility category" and ends with "See the service" and "Read the results"; it does not link to /compare.Source
    • Competitor researchRivals' own pages: Profound, "The AI marketing platform to win in" AI search platforms, with no mention of coding agents; Peec AI, "AI search analytics for marketing teams"; Amplifying.ai for vendors, Benchmarking ("when coding agents choose your product, what they choose instead, and why") and Ground Control (change docs, rerun the same task); Gauge's launch page, "Agent Led Growth: Get written into every customer's codebase".Source

This quarter 10

The core plays for the next three months.

  • 07

    Each leaderboard ranks up to nine vendors from hundreds of runs. On the payments board, the bottom four have 1 to 3 wins each. The GTM hire is asked to build automations that find, qualify and reach prospects, and the leaderboards already list who is losing and by how much.

    High impactMedium effort First result in 1 month Founding GTM

    First action

    Write the notes by hand for the payments board's vendors ranked two to nine.

    How to measure it

    Calls from leaderboard outreach

    Now
    –
    Target
    10 a month
    By
    90 days

    Steps

    1. 1

      Pick boards

      Categories with high-ACV vendors.

      Founding GTM 1 day

    2. 2

      Generate notes

      Share, top rival, one trace, one cause.

      Founding Engineer, Applied AI Research 1 week

    3. 3

      Find the right person

      Developer relations, growth or product marketing at each vendor.

      Founding GTM 1 week

    4. 4

      Review

      A founder reads each note before it goes out.

      CEO (co-founder) weekly

    5. 5

      Send

      One note per vendor per board refresh.

      Founding GTM per refresh

    6. 6

      Learn

      Replies by category, rank and role.

      Founding GTM monthly

    First result
    1 month
    Cost
    Outreach tooling, under $200 a month
    Owner
    Founding GTM

    Smallest test

    Eight hand-written notes from one board.

    Why it works

    For each new or refreshed board, generate a short note for every vendor except the leader: their share, the rival that beat them most often, one losing trace and one likely cause. Send it from a founder to the right role at that vendor (developer relations, growth or product marketing), with the private report from idea 1 as the call to action. Sequence it so each vendor gets one note per board refresh, not a drip campaign.

    Outreach that opens with the prospect's own measured loss is not cold. The research is already done, so automation makes each note almost free.

    Risks and how to handle them

    • It reads as pressure selling

      Lead with the finding, link the public board, and make one ask.

    Builds on

    Evidence

    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
    • Your websitePayments leaderboard (updated September 29, 2026): 657 runs by five coding agents on 11 codebases; nine providers ranked, the bottom four with 1 to 3 wins each; links to an interactive board showing every run's session, diff and judge verdict, and a Markdown copy; no "is your product here?" prompt and no email capture.Source
    • Your websiteLibrary page for the payments leader (published October 14, 2026): install share 88% (578 of 657 judged sessions), rank 1 of 9, 75 sessions in which agents considered it and did not choose it, with splits by agent and by persona (junior developer 100%, enterprise team 83%, senior engineer 77%); closing links "See the service", "Read the results" and "All library pages"; no form, no email capture, no link to the readiness scan, no author.Source
  • 08

    Discoverability is sold only as a service "From $5,000 / mo", and the first week of every engagement is already a defined audit: pick rate, competitors and reasons. A vendor curious about its pick rate has two options, a monthly commitment or nothing.

    High impactLow effort First result in 1 month CEO

    First action

    Set the price and offer it to the next three calls that stall on the monthly fee.

    How to measure it

    Audit-to-Growth conversion

    Now
    –
    Target
    40%
    By
    2 quarters

    Steps

    1. 1

      Define the scope

      Panel, agents, personas and the deliverable.

      Founding Forward Deployed Engineer 2 days

    2. 2

      Set the price

      A fixed fee, credited against month one.

      CEO (co-founder) 1 day

    3. 3

      Template the report

      The same structure for every category.

      Founding Forward Deployed Engineer 1 week

    4. 4

      Add to pricing

      A card next to Growth.

      Founding Engineer, Product 1 day

    5. 5

      Offer in calls

      For vendors that hesitate on the retainer.

      Founding GTM ongoing

    6. 6

      Track

      Audit-to-Growth conversion and delivery time.

      CEO (co-founder) monthly

    First result
    1 month
    Cost
    Delivery time per audit
    Owner
    CEO (co founder)

    Smallest test

    Offer it by hand to three stalled deals.

    Why it works

    Offer the baseline panel audit as a one-off product at a fixed price: the vendor's pick rate across agents and personas, the rivals that beat it, the top five reasons with traces, and a ranked fix list. Credit the fee against the first month if they sign within 30 days. Put it on /pricing next to Growth.

    A fixed-price, fixed-scope first purchase is easier to approve than a retainer, and the audit already exists inside delivery. It also screens out vendors that won't act on the findings.

    Risks and how to handle them

    • Audits cannibalise retainers

      Credit the fee only when they sign within 30 days.

    Evidence

    • Your websitePricing: Discoverability Growth "From $5,000 / mo" with "A dedicated growth engineer", weekly reviews and five or more new drafts a week, sold as a service with no self-serve plan; Usability Free "$0" with 1,000 credits a month, "Then $50 per 1,000 extra credits" and 7-day retention; the only other tier is Custom (SSO/SAML, audit logs, custom retention).Source
    • Your websiteDiscoverability page: Measure, Find changes, Create and ship, Report; first four weeks Audit, First content, First measurement, Report and plan; a monthly report "with the recorded agent runs behind each claim"; figures such as "11% to 23%" and "9 / 32 runs pick you" are shown as examples; no case studies, customer names or testimonials; "Tell us about your product, and we will show you where you stand today."Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
  • 09

    Usability setup is "a few lines and one deploy", with SDKs for TypeScript, Python, Go and PHP, and the page already suggests asking your coding agent to wire it in. The free tier keeps data for only 7 days, so a user who signs up and doesn't deploy that week never sees their use cases or issues.

    Medium impactMedium effort First result in 1 month Founding Engineer

    First action

    Measure signup-to-first-session today, by SDK.

    How to measure it

    7-day activation

    Now
    –
    Target
    50%
    By
    90 days

    Steps

    1. 1

      Measure

      Signup to first session, by SDK.

      Founding Engineer, Product 1 day

    2. 2

      Write prompts

      One per SDK and framework.

      Founding Engineer, Product 3 days

    3. 3

      Test with agents

      Run each prompt on sample servers in the panel.

      Founding Engineer, Applied AI Research 1 week

    4. 4

      Ship on signup

      Success screen and docs.

      Founding Engineer, Product 2 days

    5. 5

      First-insight email

      Top use cases and one issue.

      Founding Engineer, Product 1 week

    6. 6

      Nudge

      Day 3 with no sessions.

      Founding GTM ongoing

    First result
    1 month
    Cost
    Engineering time
    Owner
    Founding Engineer, Product

    Smallest test

    Add one TypeScript prompt to the success screen.

    Why it works

    Publish a copy-paste install prompt for each SDK on the signup success screen and in the docs, detecting the framework from the user's answer. After the first sessions arrive, send one email: "Your top three use cases, and one issue affecting most users." Nudge anyone with no sessions after three days, linking to the prompt again.

    For a self-serve analytics tool, time to first insight is everything. The product's own output (use cases and issues) is the best activation email it can send.

    Risks and how to handle them

    • Prompts break as SDKs change

      Add the prompts to the nightly evals.

    Evidence

    • Your websiteUsability page and docs: SDKs for TypeScript, Python and Go (the docs add PHP); "It takes a few lines and one deploy"; the page suggests prompting your coding agent to wire it in; PII and secrets redacted by default; use cases ranked "including the use cases you don't support yet"; the docs publish llms.txt and an Armature MCP for asking "what are my top issues this week?"; no customer logos or testimonials.Source
    • Your websitePricing: Discoverability Growth "From $5,000 / mo" with "A dedicated growth engineer", weekly reviews and five or more new drafts a week, sold as a service with no self-serve plan; Usability Free "$0" with 1,000 credits a month, "Then $50 per 1,000 extra credits" and 7-day retention; the only other tier is Custom (SSO/SAML, audit logs, custom retention).Source
  • 10

    Armature's own study found that a "1-day retention" note on a free plan cost one email API against a rival when coding agents chose. Armature's free usability plan says "7-day retention". Above it, the only tier is Custom, behind "Let's talk". So the one plan-limit line the research warns about sits on Armature's own pricing, and a team that outgrows 7 days has no button to press.

    Medium impactLow effort First result in 1 month Founding Engineer

    First action

    Run 50 sessions of the analytics task on the panel with the current pricing page.

    How to measure it

    Self-serve paid accounts

    Now
    –
    Target
    15
    By
    2 quarters

    Steps

    1. 1

      Baseline runs

      Analytics-for-MCP tasks on the panel.

      Founding Engineer, Applied AI Research 3 days

    2. 2

      Read the traces

      Did agents cite the retention line?

      Founding Engineer, Applied AI Research 2 days

    3. 3

      Design the tier

      30 or 90 days, priced per credits.

      Founding Engineer, Product 2 days

    4. 4

      Ship billing

      Self-serve checkout.

      Founding Engineer, Product 1 week

    5. 5

      Rerun

      Same tasks, new pricing page.

      Founding Engineer, Applied AI Research 3 days

    6. 6

      Publish the finding

      A library page on Armature's own test.

      Founding GTM 2 days

    First result
    1 month
    Cost
    Panel runs and engineering time
    Owner
    Founding Engineer, Product

    Smallest test

    The 50-session baseline alone.

    Why it works

    Run Armature's own panel on tasks like "add analytics to my MCP server" and record whether agents read the pricing page and how they react to the retention line. Then add a self-serve paid tier with 30- or 90-day retention at a published price, keeping Custom for SSO and audit logs. Rerun the panel and compare pick rates.

    It tests Armature's own thesis on itself, and a priced middle tier gives growing teams a way to pay without a sales call.

    Risks and how to handle them

    • Longer retention raises storage cost

      Price it per credit, as overage already is.

    Evidence

    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • Your websitePricing: Discoverability Growth "From $5,000 / mo" with "A dedicated growth engineer", weekly reviews and five or more new drafts a week, sold as a service with no self-serve plan; Usability Free "$0" with 1,000 credits a month, "Then $50 per 1,000 extra credits" and 7-day retention; the only other tier is Custom (SSO/SAML, audit logs, custom retention).Source
    • Your websiteUsability page and docs: SDKs for TypeScript, Python and Go (the docs add PHP); "It takes a few lines and one deploy"; the page suggests prompting your coding agent to wire it in; PII and secrets redacted by default; use cases ranked "including the use cases you don't support yet"; the docs publish llms.txt and an Armature MCP for asking "what are my top issues this week?"; no customer logos or testimonials.Source
  • 11

    agent.reviews collects "Tool reviews written by coding agents, right after real tasks", with large counts such as 20,152 reviews for Git. Vendors have no way to claim a listing. The only vendor path is a small "Evaluate your tool" link in the footer.

    Medium impactMedium effort First result in 6 weeks Founding Engineer

    First action

    Add a "Claim this tool" link that opens a two-field form on the ten most-reviewed listings.

    How to measure it

    Claimed listings

    Now
    –
    Target
    50
    By
    2 quarters

    Steps

    1. 1

      Design the claim

      Work-email domain verification.

      Founding Engineer, Product 3 days

    2. 2

      Set the rules

      Vendors can add docs links, never edit reviews.

      CEO (co-founder) 1 day

    3. 3

      Build the summary

      What agents struggled with, privately.

      Founding Engineer, Applied AI Research 1 week

    4. 4

      Ship

      On every listing.

      Founding Engineer, Product 1 week

    5. 5

      Route

      Claims to GTM with any leaderboard data.

      Founding GTM ongoing

    6. 6

      Measure

      Claims, calls and audits.

      Founding GTM monthly

    First result
    6 weeks
    Cost
    Engineering time
    Owner
    Founding Engineer, Product

    Smallest test

    Claim links on ten listings for a month.

    Why it works

    Add "Claim this tool" to every listing. Verify through a work email at the vendor's domain. Claimed vendors can add official docs links and setup notes, and see a private summary of what agents struggled with. Each claim goes to the GTM hire with the vendor's pick-rate data from the leaderboards, if the tool is on one.

    Vendors routinely claim listings on review sites. Here every claim is a developer-tool team asking to see how agents rate them, which is exactly Armature's buyer.

    Risks and how to handle them

    • Looks like pay-to-play

      State publicly that claims never change reviews or rankings.

    Evidence

    • Your websiteagent.reviews (by Armature): "Where agents choose and review software"; "Tool reviews written by coding agents, right after real tasks"; large counts such as 20,152 reviews for Git; no way for a vendor to claim its listing; the only vendor path is an "Evaluate your tool" footer link to armature.tech.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
  • 12

    To a reader that doesn't run JavaScript, which describes many agents' web tools, the agent readiness page shows only "Preparing the 45-test scanner…" and the leaderboards index only "Loading leaderboard results". /compare names the 45-check scan as the starting point for every customer. The individual leaderboards already have Markdown copies and the docs publish llms.txt, so the fix is a pattern Armature already uses.

    Medium impactLow effort First result in 2 weeks Founding Engineer

    First action

    Fetch both pages with a non-JavaScript reader and list what is missing.

    How to measure it

    Key pages with server-rendered content

    Now
    –
    Target
    100%
    By
    30 days

    Steps

    1. 1

      Audit

      Key pages as a non-JavaScript reader sees them.

      Founding Engineer, Product 1 day

    2. 2

      Server-render

      Scanner description and the leaders table.

      Founding Engineer, Product 3 days

    3. 3

      Add Markdown twins

      As on the individual boards.

      Founding Engineer, Product 1 day

    4. 4

      Add llms.txt

      On armature.tech, listing boards, library and scanner.

      Founding Engineer, Product 1 day

    5. 5

      Self-scan

      All 45 checks on armature.tech.

      Founding Engineer, Applied AI Research 1 day

    6. 6

      Publish

      Score, fixes and a before-and-after.

      Founding GTM 2 days

    First result
    2 weeks
    Cost
    Engineering time
    Owner
    Founding Engineer, Product

    Smallest test

    Server-render the scanner description this week.

    Why it works

    Server-render a plain description of the scanner (what the 45 checks cover and what you get) and a static table of category leaders on the index, each with a Markdown twin. Add an llms.txt to armature.tech that lists the leaderboards, library and scanner. Then run the 45 checks on armature.tech and publish the score and fixes as a library page.

    Buyers will ask their own agents about Armature. If those agents can read the scanner and the leaderboards, Armature's best assets reach them, and a public self-score shows it follows its own advice.

    Risks and how to handle them

    • A low self-score embarrasses

      Publish after the fixes, with the before-and-after.

    Evidence

    • Your websiteAgent readiness page: titled "Agent Discoverability Readiness", a "Public domain scanner"; to a reader that does not run JavaScript it shows only "Preparing the 45-test scanner…", with no input or list of checks; /compare names this 45-check scan as the starting point.Source
    • Your websiteLeaderboards index: one "Evaluate my product →" button under "Get featured by coding agents"; no form to add a product or request a category; the results area reads "Loading leaderboard results" to a reader that does not run JavaScript.Source
    • Your websiteUsability page and docs: SDKs for TypeScript, Python and Go (the docs add PHP); "It takes a few lines and one deploy"; the page suggests prompting your coding agent to wire it in; PII and secrets redacted by default; use cases ranked "including the use cases you don't support yet"; the docs publish llms.txt and an Armature MCP for asking "what are my top issues this week?"; no customer logos or testimonials.Source
    • Your websitePayments leaderboard (updated September 29, 2026): 657 runs by five coding agents on 11 codebases; nine providers ranked, the bottom four with 1 to 3 wins each; links to an interactive board showing every run's session, diff and judge verdict, and a Markdown copy; no "is your product here?" prompt and no email capture.Source
  • 13

    Armature's usability products are themselves installed by coding agents: the page tells users to ask their coding agent to wire the SDK in. The site shows no customer results yet, and the sitemap already has evaluation and product-analytics leaderboards. Armature's own pick rate, measured the same way, is the one case study it can publish without anyone's permission.

    Medium impactMedium effort First result in 1 quarter Founding Engineer

    First action

    Write five panel tasks where Armature's usability product is a valid choice.

    How to measure it

    Own pick rate

    Now
    –
    Target
    +10 points
    By
    2 quarters

    Steps

    1. 1

      Write tasks

      Analytics and evals tasks across languages.

      Founding Engineer, Applied AI Research 2 days

    2. 2

      Baseline

      Run them on all panel agents.

      Founding Engineer, Applied AI Research 1 week

    3. 3

      Find causes

      Why agents chose rivals or built in-house.

      Founding Forward Deployed Engineer 1 week

    4. 4

      Ship fixes

      Docs, SDK guides and listings.

      Founding Engineer, Product 2 weeks

    5. 5

      Rerun

      Same tasks, a month later.

      Founding Engineer, Applied AI Research 1 week

    6. 6

      Publish

      Before-and-after with traces and the disclosure.

      Founding GTM 2 days

    First result
    1 quarter
    Cost
    Panel runs
    Owner
    Founding Engineer, Applied AI Research

    Smallest test

    Baseline one task on the panel.

    Why it works

    Add tasks such as "add session analytics to this MCP server" and "add regression evals for this CLI" to the panel. Publish Armature's baseline pick rate, the docs and listing changes it ships, and the new rate a month later, with traces. Mark the conflict of interest clearly and use the same judge and method as every other board.

    It shows the full service loop on a real product, gives the usability product a growth channel through agents, and answers sceptics with Armature's own traces.

    Risks and how to handle them

    • Seen as self-serving

      The same method and judge as every board, with traces public.

    Evidence

    • Your websiteUsability page and docs: SDKs for TypeScript, Python and Go (the docs add PHP); "It takes a few lines and one deploy"; the page suggests prompting your coding agent to wire it in; PII and secrets redacted by default; use cases ranked "including the use cases you don't support yet"; the docs publish llms.txt and an Armature MCP for asking "what are my top issues this week?"; no customer logos or testimonials.Source
    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • Your websiteDiscoverability page: Measure, Find changes, Create and ship, Report; first four weeks Audit, First content, First measurement, Report and plan; a monthly report "with the recorded agent runs behind each claim"; figures such as "11% to 23%" and "9 / 32 runs pick you" are shown as examples; no case studies, customer names or testimonials; "Tell us about your product, and we will show you where you stand today."Source
  • 14

    The study drew 300 points and 153 comments on Hacker News. Readers praised the data but attacked the business model of selling agent recommendations, and some questioned coverage. The study already includes a disclosure, and agent.reviews depends on developers trusting it. Right now that trust is defended thread by thread.

    Medium impactLow effort First result in 1 month CEO

    First action

    Collect the method questions and objections from the Hacker News thread.

    How to measure it

    Inbound links to the method page

    Now
    –
    Target
    25
    By
    2 quarters

    Steps

    1. 1

      Collect objections

      Hacker News and sales calls.

      Founding GTM 1 day

    2. 2

      Write the method

      Panels, personas, judges and what counts as a pick.

      Founding Engineer, Applied AI Research 2 days

    3. 3

      Write the rules

      What Armature will never do.

      CEO (co-founder) 1 day

    4. 4

      Add a challenge path

      How to dispute a result.

      Founding Engineer, Product 1 day

    5. 5

      Link everywhere

      Boards, library, studies, agent.reviews.

      Founding Engineer, Product 1 day

    6. 6

      Point to it

      In the next public thread.

      CEO (co-founder) ongoing

    First result
    1 month
    Cost
    $0
    Owner
    CEO (co founder)

    Smallest test

    Add a "What we won't do" section to the study post.

    Why it works

    Write one page that every study, board and agent.reviews listing links to. It covers how panels, personas and judges work, and that customers and non-customers get the same method. It states what Armature will not do: no paid placement in leaderboards, no edits to agent reviews, and no fixes that mislead agents (only better docs, setup and listings). It ends with how to challenge a result and how to suggest a category.

    Buyers need the research to be credible, because that credibility is what they are buying. A clear line between measuring and selling protects the leaderboards and answers the main objection before it is raised.

    Risks and how to handle them

    • Rules limit future offers

      Write rules for the measurement, not for the services.

    Evidence

    • Hacker NewsHacker News thread on the study: 300 points and 153 comments; praise for the data ("The data was cool."), criticism of the business model of selling agent recommendations, complaints about pop-ups and the mobile layout (the author replied the pop-ups were fixed), and readers naming tools and categories they wanted covered.Source
    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • Your websiteagent.reviews (by Armature): "Where agents choose and review software"; "Tool reviews written by coding agents, right after real tasks"; large counts such as 20,152 reviews for Git; no way for a vendor to claim its listing; the only vendor path is an "Evaluate your tool" footer link to armature.tech.Source
  • 15

    Growth customers pay from $5,000 a month and get weekly reviews and a monthly report "with the recorded agent runs behind each claim". The FDE owns "the results promised". At high contract values, renewal depends on the champion convincing a leader who never reads the full report.

    High impactLow effort First result in 1 quarter Founding Forward Deployed Eng…

    First action

    Build the slide for the current customers' next reports.

    How to measure it

    Month-3 retention

    Now
    –
    Target
    90%
    By
    2 quarters

    Steps

    1. 1

      Design the slide

      Then and now, by agent, rival gap, changes, one trace.

      Founding Forward Deployed Engineer 2 days

    2. 2

      Automate it

      Generated from the report data.

      Founding Engineer, Product 1 week

    3. 3

      Add the note

      Two forwardable lines.

      Founding Forward Deployed Engineer 1 hour

    4. 4

      Send

      With every monthly report.

      Founding Forward Deployed Engineer monthly

    5. 5

      Ask for feedback

      Did it reach their leader?

      CEO (co-founder) quarterly

    6. 6

      Track

      Renewals and expansions.

      CEO (co-founder) quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Founding Forward Deployed Engineer

    Smallest test

    Send the slide to one customer next month.

    Why it works

    Add a single slide to every monthly report. It shows the pick rate then and now, by agent, the rival gap, the changes shipped, and one trace link as proof. Keep the format fixed so a quarter of slides tells a story. Send it to the champion with a two-line note they can forward.

    Renewals are won inside the customer's company. A forwardable result does the champion's work for them and makes Armature visible above them.

    Risks and how to handle them

    • Flat months look bad

      Show the work shipped and what it is expected to move.

    Evidence

    • Your websiteDiscoverability page: Measure, Find changes, Create and ship, Report; first four weeks Audit, First content, First measurement, Report and plan; a monthly report "with the recorded agent runs behind each claim"; figures such as "11% to 23%" and "9 / 32 runs pick you" are shown as examples; no case studies, customer names or testimonials; "Tell us about your product, and we will show you where you stand today."Source
    • Your websitePricing: Discoverability Growth "From $5,000 / mo" with "A dedicated growth engineer", weekly reviews and five or more new drafts a week, sold as a service with no self-serve plan; Usability Free "$0" with 1,000 credits a month, "Then $50 per 1,000 extra credits" and 7-day retention; the only other tier is Custom (SSO/SAML, audit logs, custom retention).Source
    • HiringFounding Forward Deployed Engineer: builds the customer operations function; designs, runs and presents experiments measuring how often coding agents choose each customer's product; owns customer relationships and the results promised to them; automates delivery with AI agents, one repeated task at a time.Source
  • 16

    Real visitors pass Core Web Vitals (field LCP 1.4 s, CLS 0.01), so this is not urgent. But field LCP is 28% slower than in the previous period. Google's phone test scores 67, with lab LCP at 3.5 s, 871 ms of blocking time and a 4.1 MB page with 174 requests. The layout shift comes from the rotating headline, and the page has two accessibility issues (colour contrast, no main landmark).

    Low impactLow effort First result in 1 week Founding Engineer

    First action

    Fix the H1 width and add the main landmark today.

    How to measure it

    PageSpeed mobile score

    Now
    –
    Target
    85
    By
    60 days

    Steps

    1. 1

      Fix the headline

      A fixed-width slot for the rotating name.

      Founding Engineer, Product 1 hour

    2. 2

      Fonts

      Self-host and subset.

      Founding Engineer, Product 2 hours

    3. 3

      Images

      Modern formats, right sizes.

      Founding Engineer, Product 2 hours

    4. 4

      Defer panels

      Load demo panels on scroll.

      Founding Engineer, Product 1 day

    5. 5

      Accessibility

      Contrast and the main landmark.

      Founding Engineer, Product 2 hours

    6. 6

      Watch the trend

      Field LCP monthly.

      Founding Engineer, Product monthly

    First result
    1 week
    Cost
    $0
    Owner
    Founding Engineer, Product

    Smallest test

    Fix the H1 width and rerun PageSpeed.

    Why it works

    Reserve a fixed width for the rotating agent name so the H1 doesn't move. Self-host and subset the fonts (Google Fonts adds 190 KB), compress the images (about 244 KB to save) and defer the demo panels below the fold. Fix contrast and add a main landmark. Keep the field LCP trend on a monthly check.

    Developers notice slow, jumpy pages, and the HN thread already complained about the mobile layout. Holding the speed lead is cheap now and expensive later.

    Risks and how to handle them

    • The design loses its motion

      Keep the rotation, just inside a fixed box.

    Evidence

    • PageSpeedHomepage (November 10, 2026): mobile score 67, desktop 73; lab LCP 3.5 s, Total Blocking Time 871 ms, 4.1 MB and 174 requests; real visitors pass Core Web Vitals (field LCP 1.4 s, CLS 0.01), but field LCP is 28% slower than in the previous period; the layout shift comes from the rotating H1; 2 accessibility issues (colour contrast, no main landmark); Google Fonts 190 KB; about 244 KB of image savings available.Source
    • Hacker NewsHacker News thread on the study: 300 points and 153 comments; praise for the data ("The data was cool."), criticism of the business model of selling agent recommendations, complaints about pop-ups and the mobile layout (the author replied the pop-ups were fixed), and readers naming tools and categories they wanted covered.Source

Big bets 3

Larger plays that can change the trajectory.

  • 17

    Armature names about 291 products in its library, but discoverability is sold only as a service from $5,000 a month. The product engineer is hired to decide what becomes self-serve, and the job post says thousands of customer experiments run on internal infrastructure, mostly delivered by people. Amplifying.ai already sells benchmarking to vendors.

    High impactHigh effort First result in 1 quarter Founding Engineer

    First action

    List the vendors who asked for reports but did not buy Growth, and ask what they would pay to monitor.

    How to measure it

    Self-serve monitor customers

    Now
    –
    Target
    40
    By
    2 quarters

    Steps

    1. 1

      Validate price

      Ten calls with vendors who didn't buy Growth.

      CEO (co-founder) 2 weeks

    2. 2

      Scope

      One category, existing panel, alerts.

      Founding Engineer, Product 1 week

    3. 3

      Automate runs

      Scheduled runs per vendor.

      Founding Engineer, Applied AI Research 1 month

    4. 4

      Build the dashboard

      Rate, rivals, traces, alerts.

      Founding Engineer, Product 1 month

    5. 5

      Billing

      Self-serve checkout.

      Founding Engineer, Product 1 week

    6. 6

      Upgrade path

      "Fix this" opens a Growth call.

      Founding GTM 1 week

    7. 7

      Launch

      To requesters from ideas 1 and 2.

      Founding GTM 2 weeks

    First result
    1 quarter
    Cost
    Engineering time plus panel compute per account
    Owner
    Founding Engineer, Product

    Smallest test

    Sell a manual monthly monitor to five vendors before building it.

    Why it works

    Package the panel as a monthly self-serve monitor for one category. It shows the vendor's pick rate by agent and persona, the top rivals, losing traces and alerts when the rate moves, all at a published price. Fixes and content stay in the Growth service, and the monitor's upgrade button opens it. Start with the categories that already have boards.

    It reaches vendors too small for a retainer, makes delivery scale with software instead of people, and gives Growth a steady stream of vendors who already watch their numbers.

    Risks and how to handle them

    • Compute cost outruns price

      Cap runs per plan and price per category.

    • Cannibalises Growth

      Keep fixes and content human-delivered.

    Builds on

    Evidence

    • HiringFounding Engineer, Product: decides what becomes self-serve and what keeps a human in the loop; Armature runs thousands of experiments for customers on internal infrastructure, mostly delivered through people, and the product should make operations "100x smoother".Source
    • Your websitePricing: Discoverability Growth "From $5,000 / mo" with "A dedicated growth engineer", weekly reviews and five or more new drafts a week, sold as a service with no self-serve plan; Usability Free "$0" with 1,000 credits a month, "Then $50 per 1,000 extra credits" and 7-day retention; the only other tier is Custom (SSO/SAML, audit logs, custom retention).Source
    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • Competitor researchRivals' own pages: Profound, "The AI marketing platform to win in" AI search platforms, with no mention of coding agents; Peec AI, "AI search analytics for marketing teams"; Amplifying.ai for vendors, Benchmarking ("when coding agents choose your product, what they choose instead, and why") and Ground Control (change docs, rerun the same task); Gauge's launch page, "Agent Led Growth: Get written into every customer's codebase".Source
  • 18

    The September study (16,893 runs, 51 codebases, 18 sectors) reached 300 points on Hacker News with findings such as "agents picked the same tool in only 42% of cells". The GTM post says most customers find Armature through this research. But it is one study, and the results change with every agent release.

    High impactHigh effort First result in 1 quarter Founding Engineer

    First action

    Freeze the core panel and set the next edition's date.

    How to measure it

    Calls per edition

    Now
    –
    Target
    25
    By
    2 editions

    Steps

    1. 1

      Freeze the panel

      Codebases, prompts and personas.

      Founding Engineer, Applied AI Research 2 weeks

    2. 2

      Define the index

      One number per category.

      Founding Engineer, Applied AI Research 1 week

    3. 3

      Run

      The full panel each quarter.

      Founding Engineer, Applied AI Research quarterly

    4. 4

      Write

      Headline findings and movers.

      Founding GTM 1 week

    5. 5

      Brief

      Newsletters and press under embargo.

      CEO (co-founder) 1 week

    6. 6

      Notify movers

      Vendors who rose or fell.

      Founding GTM launch day

    First result
    1 quarter
    Cost
    Panel compute and staff time
    Owner
    Founding Engineer, Applied AI Research

    Smallest test

    Rerun three categories and publish the movement.

    Why it works

    Fix a core panel and rerun it every quarter. Publish an index with one headline number per category (leader share, movement, biggest riser and faller), dated and comparable over time. Brief developer-tool newsletters and journalists a week ahead under embargo. Answer the coverage critics from the HN thread by adding requested categories to each edition.

    A recurring, dated number becomes a reference that others cite. Each edition is a fresh news hook, and every vendor that moves has a reason to call.

    Risks and how to handle them

    • Method changes break comparisons

      Version the method and keep a fixed core panel.

    Builds on

    Evidence

    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • Hacker NewsHacker News thread on the study: 300 points and 153 comments; praise for the data ("The data was cool."), criticism of the business model of selling agent recommendations, complaints about pop-ups and the mobile layout (the author replied the pop-ups were fixed), and readers naming tools and categories they wanted covered.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
    • HiringFounding Engineer, Applied AI Research: runs the studies behind the public leaderboards, published as open leaderboards with every trace available; publishes findings under their own name and decides what Armature studies next; turns customer questions into better evals.Source
  • 19

    Armature is in YC's P26 batch and shows "Backed by Y Combinator" as its main trust mark. Its customers are mostly US developer-tool companies with high contract values. Investors and accelerators hold dozens of those companies each, and the 45-check readiness scan is already the stated starting point for a customer.

    High impactMedium effort First result in 1 quarter CEO

    First action

    Ask the YC partner group whether a portfolio scan would be useful to batch-mates selling to developers.

    How to measure it

    Partner-sourced audits

    Now
    –
    Target
    15
    By
    2 quarters

    Steps

    1. 1

      Pick the first partner

      Armature's own accelerator community.

      CEO (co-founder) 1 week

    2. 2

      Package the scan

      Readiness scan plus a one-category snapshot.

      Founding Forward Deployed Engineer 2 weeks

    3. 3

      Run the portfolio

      Every developer-tool company they list.

      Founding Engineer, Applied AI Research 2 weeks

    4. 4

      Deliver

      A private report and a founder talk.

      CEO (co-founder) 1 day

    5. 5

      Follow up

      Each company's own result and the audit offer.

      Founding GTM 2 weeks

    6. 6

      Repeat

      One or two developer-focused funds.

      CEO (co-founder) quarterly

    First result
    1 quarter
    Cost
    Panel compute and founder time
    Owner
    CEO (co founder)

    Smallest test

    Scan ten batch-mates selling to developers and offer each its result.

    Why it works

    Offer a portfolio partner a free readiness scan and one-category pick-rate snapshot for every developer-tool company they hold, delivered as a private portfolio report plus a short talk for founders. Each company gets its own result and an offer for the audit from idea 8. Start with Armature's own accelerator community, then one or two developer-focused funds.

    One relationship opens dozens of warm introductions to the right buyer, and the investor gets a new view of how its companies will be found as agents choose more software.

    Risks and how to handle them

    • Portfolio data is sensitive

      Each company sees only its own results; the investor sees only aggregates unless the company agrees.

    Builds on

    Evidence

    • HiringYC company page: batch P26, founded 2026, team of 3, Paris; four open founding roles (GTM, Forward Deployed Engineer, Applied AI Research, Product) at €60K–€100K plus 0.30%–1.00% equity.Source
    • Your websiteHomepage: H1 "Be the tool" followed by the rotating names of three coding agents; two offers, Agent discoverability ("Service run with you") and Agent usability (self-serve, "Start for free"); "Backed by Y Combinator"; no customer logos or testimonials; a sample "Your pick rate on the panel: 2% → 11%"; the FAQ counts a pick when the agent installs the product and wires it into the repo, not when it only mentions it.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
    • Your websiteAgent readiness page: titled "Agent Discoverability Readiness", a "Public domain scanner"; to a reader that does not run JavaScript it shows only "Preparing the 45-test scanner…", with no input or list of checks; /compare names this 45-check scan as the starting point.Source

Wildcards 3

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

  • 20

    Armature already publishes leaderboard pages for individual new models, the latest on November 9, 2026. Its study found agents agree on the same tool in only 42% of cases. So a model release can move a vendor's share overnight, and only Armature is watching. Today those pages are published, but nobody is told.

    Medium impactMedium effort First result in First release after launch Founding Engineer

    First action

    Diff the two most recent model pages by category and write the first "what changed" page.

    How to measure it

    Calls per release alert

    Now
    –
    Target
    5
    By
    3 releases

    Steps

    1. 1

      Set the trigger

      Which releases count, and the 48-hour target.

      Founding Engineer, Applied AI Research 1 day

    2. 2

      Automate the diff

      Category shares, new model against previous.

      Founding Engineer, Applied AI Research 2 weeks

    3. 3

      Publish

      A short "what changed" page.

      Founding GTM per release

    4. 4

      Alert vendors

      Moves above five points, with traces.

      Founding GTM per release

    5. 5

      Offer the audit

      For vendors that fell.

      Founding GTM per release

    6. 6

      Review

      Calls and press per release.

      CEO (co-founder) quarterly

    First result
    First release after launch
    Cost
    Panel compute per release
    Owner
    Founding Engineer, Applied AI Research

    Smallest test

    Diff the last two model pages by hand and email three vendors.

    Why it works

    Within 48 hours of a major release, rerun the core panel with the new model and compare each category with the previous model. Publish a short "what changed" page, and email each vendor whose share moved more than five points: up (proof their work holds) or down (what the new model reads instead). Pair it with the research email from idea 3.

    A sudden drop is the moment a vendor most needs to know why. Arriving first, with traces, makes Armature the source vendors check after every release.

    Risks and how to handle them

    • Small samples overstate moves

      Alert only above a minimum number of runs and show the error margin.

    Builds on

    Evidence

    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
  • 21

    Armature's persona splits show big gaps. Even the payments leader wins 100% of sessions for a junior developer but 83% for an enterprise team and 77% for a senior engineer. Armature's customers have high contract values, which usually means enterprise buyers, and the library already has a guide on winning the enterprise persona.

    Medium impactMedium effort First result in 1 quarter Founding Forward Deployed Eng…

    First action

    Pull enterprise-versus-junior gaps for every category and list the ten largest.

    How to measure it

    Persona reports sold

    Now
    –
    Target
    6
    By
    2 quarters

    Steps

    1. 1

      Find the gaps

      Enterprise versus junior, by category.

      Founding Engineer, Applied AI Research 3 days

    2. 2

      Build the panel

      Enterprise-style codebases and prompts.

      Founding Engineer, Applied AI Research 2 weeks

    3. 3

      Template the report

      Losses, rivals, reasons and traces.

      Founding Forward Deployed Engineer 1 week

    4. 4

      Price it

      An add-on to the audit or Growth.

      CEO (co-founder) 1 day

    5. 5

      Landing page

      Upgrade the enterprise-persona guide.

      Founding GTM 2 days

    6. 6

      Sell

      To vendors with the widest gaps.

      Founding GTM ongoing

    First result
    1 quarter
    Cost
    Panel compute and staff time
    Owner
    Founding Forward Deployed Engineer

    Smallest test

    Send the enterprise gap from one category to three vendors.

    Why it works

    Offer a report run only with enterprise and senior-engineer personas on enterprise-style codebases (compliance, existing vendors, larger stacks). It shows where the vendor loses those buyers and to whom. Price it as a paid add-on to the audit or Growth, and use the guide as its landing page.

    For a vendor selling to enterprises, losing the enterprise persona costs the most, and few rivals split results by persona. It turns a research detail into a premium product.

    Risks and how to handle them

    • Simulated personas questioned

      Publish the persona prompts in the method page (idea 14).

    Builds on

    Evidence

    • Your websiteLibrary page for the payments leader (published October 14, 2026): install share 88% (578 of 657 judged sessions), rank 1 of 9, 75 sessions in which agents considered it and did not choose it, with splits by agent and by persona (junior developer 100%, enterprise team 83%, senior engineer 77%); closing links "See the service", "Read the results" and "All library pages"; no form, no email capture, no link to the readiness scan, no author.Source
    • HiringFounding GTM (Paris, frequent trips to San Francisco): "Our customers are mostly US devtool companies, with high ACV." "Most of our customers find us through this research." "Our research brings customers to us today, but we haven't yet built a repeatable way to find the next ones." The role turns each study into a lead magnet with the research engineer, builds automations to find, qualify and reach prospects, and runs first calls.Source
    • Your websiteSitemap (465 URLs, checked November 10, 2026): about 404 /library pages, about 291 of them "do-coding-agents-recommend-<product>" pages, plus guides such as "how-to-win-the-enterprise-persona"; 43 leaderboard pages (latest November 9, 2026), including pages for individual new models; 3 blog posts; /compare plus eight library "armature-vs" pages (Profound, Peec AI, Scrunch AI, Otterly, Ahrefs Brand Radar, Semrush AI Toolkit, hiring a devrel, LLM observability tools), none for Gauge or Amplifying; no case study pages.Source
  • 22

    Armature already publishes every trace, prompt and diff on its own site, and the research engineer is hired to publish under their own name. But the site links no GitHub organisation, YouTube channel or newsletter, and its X link goes to a founder's personal profile. Developers and researchers discover and cite datasets on GitHub, not inside an interactive board.

    Medium impactMedium effort First result in 1 month Founding Engineer

    First action

    Export the September study's valid sessions as a single versioned file.

    How to measure it

    GitHub stars on the dataset

    Now
    –
    Target
    500
    By
    2 quarters

    Steps

    1. 1

      Create the organisation

      A company GitHub organisation.

      Founding Engineer, Product 1 hour

    2. 2

      Export

      Sessions, verdicts and task definitions.

      Founding Engineer, Applied AI Research 1 week

    3. 3

      Document

      Schema, loader, citation file and licence.

      Founding Engineer, Applied AI Research 3 days

    4. 4

      Link it

      From every board and study.

      Founding Engineer, Product 1 day

    5. 5

      Announce

      In the research email and developer communities.

      Founding GTM 1 day

    6. 6

      Version

      A new release with each study.

      Founding Engineer, Applied AI Research quarterly

    First result
    1 month
    Cost
    $0 plus staff time
    Owner
    Founding Engineer, Applied AI Research

    Smallest test

    Publish one category's sessions and see whether it gets stars.

    Why it works

    Create a company GitHub organisation with a versioned dataset of the published sessions (task, persona, agent, pick, judge verdict), a loader script and the panel task definitions. Add a citation file and a licence that requires attribution. Link it from every board and study, and announce each new version in the research email.

    Open data earns stars, forks and citations that bring developers back to Armature. It also answers the coverage critics: anyone can check the method or propose a new category with a pull request.

    Risks and how to handle them

    • Rivals reuse the data

      Require attribution; Armature's edge is the panel and the fixes, not the archive.

    Builds on

    Evidence

    • Your websiteStudy "Which tools do coding agents install" (September 3, 2026): 16,893 runs, 5,292 valid sessions across 51 codebases and 18 sectors; the three agents picked the same tool in only 42% of cells; "being mentioned isn't winning" (one payments brand cited 139 times and never picked); a "1-day retention" note on a free plan cost one email API against a rival; full traces published; a disclosure that Armature sells growth services to developer tools; no newsletter signup, only an email address to suggest questions.Source
    • Your websiteChannels: the site links a LinkedIn company page and Y Combinator; the X account in its links is a founder's personal profile; no GitHub organisation, YouTube channel, newsletter or RSS feed is linked.Source
    • HiringFounding Engineer, Applied AI Research: runs the studies behind the public leaderboards, published as open leaderboards with every trace available; publishes findings under their own name and decides what Armature studies next; turns customer questions into better evals.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.

  • Scopely growth strategyA mobile-first game developer and publisher behind MONOPOLY GO!, Stumble Guys, MARVEL Strike Force, Star Trek Fleet Command and, through…21 growth ideas
  • LifeBonder growth strategyA friendship app and community: privacy-first friend matching, interest and emotion group chats, Friend Cards, a personality quiz and…21 growth ideas
  • Reacher growth strategyAI software for running TikTok Shop affiliate programs: creator discovery across a 4M+ creator database, automated outreach, creator CRM…23 growth ideas
  • 906 CMVI Holdings (Pure for Men) growth strategy906 CMVI Holdings is the Texas management company that runs operations, fulfillment (Houston) and marketing (Austin) for Pure for Men, a…22 growth ideas
  • Darwin Robotics growth strategyBuilds general-purpose robot technicians, starting with data centers, where robots that learn from human demonstrations take work orders…24 growth ideas
  • Awin growth strategyA global affiliate and partnership marketing platform: tracking, partner discovery, payments and services connecting advertisers with…21 growth ideas
  • Mercury growth strategyBusiness banking and finance software for startups and growing companies: checking, savings and treasury, cards and expense management…21 growth ideas
  • Vulcan Technologies growth strategySecure AI agents for public work: JUSTINIAN, an agent for government and legal knowledge workers built on a 157-billion-record corpus of…22 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. Armature did not commission, review or endorse it, and Programmable is not affiliated with Armature. 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 Armature runs its business.

Something wrong, or do you represent Armature? 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 307 ideas

    Capture the research traffic: vendor blocks on library pages, add-my-tool on leaderboards, a research email, the vendor chapters as a lead magnet, a first anonymised result, rival pages and agent-readable pages of Armature's own.

  2. Days 31 to 609 ideas

    Automated outbound, a fixed-price audit, usability activation and a paid retention tier, agent.reviews claims, Armature's own pick rate, a method page, customer reporting and homepage speed.

  3. Days 61 to 906 ideas

    Commit to the bets: a self-serve pick-rate monitor, a quarterly index, investor portfolio scans, release-day alerts, an enterprise-persona report and an open dataset.

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 hard data on what coding agents actually install, with every trace published. The most cited problem: being mentioned by agents but not picked.

  • Explanations of why an agent picked a rival
  • Agents disagreeing with each other, so one fix does not work everywhere
  • In their words: “get picked”

Market

Competitive landscape

4 competitors compared: Gauge, Amplifying.ai, Profound and Peec AI.

  • Gap to exploit: Vendor-level answers to "why did the agent pick my rival?"
  • Gauge: Not run as a done-with-you service with shipped fixes
  • Amplifying.ai: No public leaderboards on the scale of Armature's

Hiring signals

Armature, a team of three, is hiring its whole founding team in Paris at once, and each post names a growth gap.

  • Founding GTM: Research brings customers but no repeatable way to find the next ones. Ideas 1, 3, 4, 7.
  • Founding Forward Deployed Engineer (FDE): Owns customer results and automates delivery. Ideas 5, 15, 21.
  • Founding Engineer, Applied AI Research: Runs the leaderboards and decides what to study next. Ideas 2, 18, 22.

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 first (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)
  • 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