ProgrammableMarketing/onerobot.io (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 · 9
  • Big bets · 4
  • Wildcards · 2
  • How we can help
Generated Nov 10, 2026
  1. Programmable
  2. Playbooks
  3. One Robot
One Robot logo

growth playbook · Nov 10, 2026

onerobot.io (opens in a new tab)

One Robot growth playbook

Diagnosis

One Robot is a two-person YC W26 team, backed by Accel, with a sharp thesis: robotics can't industrialize without an evaluation layer, and real-robot testing is the bottleneck. The site does not yet sell that thesis. The homepage has one rollout video, a mailto link and no number showing that the world model predicts real-world results, while Runway publishes a 0.95 sim-to-real correlation and NVIDIA offers open world models.

This playbook starts by proving fidelity in public (a sim-to-real scorecard for the folding task), turning the mailto into a scoped pilot request, leading with the evaluation-layer message, and fixing link previews and findability. Next it packages a fixed-fee eval sprint, makes evaluation a habit with a regression report for every checkpoint, and publishes failure galleries and technical write-ups that also help hiring, alongside partner and lab channels.

The big bets are a public folding leaderboard, a priced fix-it data product, a calibrated trust score per rollout and owning textile handling as a beachhead.

Growth Score

6/10

Biggest Opportunity

Become the trusted evaluation layer for hard manipulation: publish evidence that the simulator ranks policies the way real robots do, then sell repeatable eval runs and the data that fixes the failures they find.

Biggest Risk

Well-funded players (Runway, 1X, NVIDIA's open Cosmos models) are moving into world-model policy evaluation, and a two-person team has to win trust and hire founding engineers before evaluation becomes a commodity feature.

01

Start here

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

  1. 1

    Put a sim-to-real scorecard for box folding next to the 60-second rollout on the homepage

    Choose the folding task and list the checkpoints you already have from customer or in-house training runs.

    High impact First result in 3 weeks
  2. 2

    Replace the founders' mailto with a five-question pilot request that scopes the first world model

    Write the five questions and the "what happens next" timeline.

    High impact First result in 2 weeks
  3. 3

    Lead the homepage with the evaluation layer: "know where your policy fails before it touches a robot"

    Write three headline options and test them on the next five sales calls.

    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.

  • 7 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 · 3Big bets · 8Fill-ins · 6Reconsider · 5
Effort →

Showing 22 of 22 ideas

This week 7

Quick wins with a clear first action.

  • 01

    A robot-learning lead deciding whether to hand over data needs one answer: does the simulator rank my policies the way my real robot would? One Robot's homepage shows a single 60-second rollout and an email link, with no number, benchmark or result anywhere. Meanwhile Runway's robotics page states a 0.95 correlation between simulated and real scores across eight policies, from 1,450 rollouts. Visitors can watch a rollout that looks real, but nothing on the page shows that its outcomes match the real robot, and fidelity is the claim the whole product depends on.

    High impactMedium effort First result in 3 weeks Co founder & CTO

    First action

    Choose the folding task and list the checkpoints you already have from customer or in-house training runs.

    How to measure it

    Qualified pilot requests from the homepage

    Now
    –
    Target
    8 a month
    By
    60 days

    Steps

    1. 1

      Pick task and checkpoints

      One folding task, 6-10 checkpoints ranging from weak to strong.

      Co-founder & CTO 1 day

    2. 2

      Fix the protocol

      Same starting scenes, success definition and number of trials in sim and on the robot.

      Co-founder & CTO 1 day

    3. 3

      Run the real trials

      At least 10 trials per checkpoint on the physical robot; log time spent.

      Machine Learning Intern 1 week

    4. 4

      Run the simulated trials

      At least 50 rollouts per checkpoint in the world model.

      Co-founder & CTO 2 days

    5. 5

      Compute the card

      Rank correlation, success-rate gap, rollouts, robot-hours replaced.

      Co-founder & CTO 1 day

    6. 6

      Pick honest clips

      One predicted failure and one miss, side by side with real video.

      Co-founder & CEO 1 day

    7. 7

      Publish

      Scorecard under the rollout, plus a one-page method note.

      Co-founder & CEO 1 day

    8. 8

      Use it everywhere

      Outreach emails, the pilot request page and the YC profile.

      Co-founder & CEO ongoing

    First result
    3 weeks
    Cost
    About two weeks of founder and intern time plus robot time; no new spend
    Owner
    Co founder & CTO

    Smallest test

    Run four checkpoints on one task and publish a simple table with the two clips.

    Why it works

    Pick the task already on the site (box or textile folding). Take 6-10 policy checkpoints of different quality, run each on the real robot and in the world model with the same starting scenes, and publish a small scorecard under "See it in action": rank correlation between simulated and real success, the average gap in success rate, number of rollouts, robot-hours the simulated runs replaced, and two side-by-side clips, one where the simulation predicted a real failure and one where it got it wrong. Link a one-page method note, and use the same card in every outreach email and the pilot request (idea 2).

    Technical buyers trust measured numbers over demo videos, and a competitor has already set the expectation that a fidelity number exists. Doing it on deformable, contact-rich folding, which Runway's single-arm tabletop study does not cover, makes the number hard to compare away. It also turns the Founding Robot Learning role's job (validate the world model end to end) into a public asset.

    Ready to use

    ChecklistSim-to-real scorecard template

    Task: box folding (or textile folding)
    Robot and cameras: <arm, grippers, camera count>
    Checkpoints compared: <n>
    Real trials per checkpoint: <n>   Simulated rollouts per checkpoint: <n>
    
    [ ] Rank correlation, simulated vs real success: <value>
    [ ] Mean absolute gap in success rate: <value>
    [ ] Robot-hours the simulated runs replaced: <value>
    [ ] Clip A: failure the simulation predicted, real vs simulated
    [ ] Clip B: case the simulation got wrong, and why
    [ ] Method note: scenes, success definition, date, version
    [ ] Next update date

    Risks and how to handle them

    • The correlation comes out weaker than hoped

      Publish it with the failure analysis anyway, and frame it as version 1 with a dated update plan; an honest number beats none.

    • Customer data cannot be shown

      Use in-house demonstration data for the published card and keep customer runs private.

    Evidence

    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • Competitor researchRunway GWM-Robotics (published February 27, 2026): a world model for evaluating robot policies without hardware; Runway reports a 0.95 Pearson correlation between simulated and real-world progress scores across eight VLA policies on a Franka Panda arm (RoboArena tasks), from 1,450 simulated rollouts and over 16,000 human ratings, and notes it measures ranking, not absolute success rates; teams request access through a form.Source
    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
    • HiringFounding Robot Learning (Ashby, San Francisco on-site, posted August 13, 2026): train manipulation policies (VLAs, diffusion policies, imitation and RL) and run them on real robots, validate the world model end to end and find where sim-to-real transfer fails; "We work with real customer data, real failures, and real deployment pressure." The postings say the company is expanding into policy training.Source
  • 02

    The homepage's only call to action is a mailto link under "Let's work together". A blank email asks the visitor to work out what to say, and the first reply then has to ask the same scoping questions every time: which robot, which task, how much data, in what format. Engineers comparing tools often skip a cold email and keep browsing.

    High impactLow effort First result in 2 weeks Co founder & CEO

    First action

    Write the five questions and the "what happens next" timeline.

    How to measure it

    Qualified pilot requests

    Now
    –
    Target
    2x the current monthly email enquiries
    By
    60 days

    Steps

    1. 1

      Write the questions

      Robot, task, episodes, data format, decision needed.

      Co-founder & CEO 2 hours

    2. 2

      Write what happens next

      Call, data check, first model and report, with timings.

      Co-founder & CEO 2 hours

    3. 3

      Build the form

      A simple form on the homepage; keep the email link beside it.

      Co-founder & CTO half a day

    4. 4

      Score fit

      Flag requests with enough episodes and a supported task.

      Co-founder & CEO 1 hour

    5. 5

      Reply fast

      Answer every request within one working day.

      Co-founder & CEO ongoing

    6. 6

      Review answers

      Monthly: which robots, tasks and formats come up most.

      Co-founder & CTO monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Co founder & CEO

    Smallest test

    Add the five questions as a suggested template in the mailto body this week.

    Why it works

    Keep the email address, but add a short pilot request with five fields: robot and grippers, task, number of recorded episodes, how the data is stored, and the decision they need to make (which checkpoint to ship, where it fails, what data to collect). Show what happens next: a 30-minute call, a data check, and a first world model and failure report in a fixed number of weeks. Route answers to the founders with a simple fit score.

    Structured requests lower the effort for the visitor and let a two-person team spend calls only on teams with the right data. The answers also show which robots and tasks to support next.

    Risks and how to handle them

    • A form feels less personal than founders' email

      Keep the email link next to it and sign replies personally.

    • Teams worry about sharing data details

      Ask only for counts and formats, never files, and say so on the form.

    Evidence

    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
  • 03

    One Robot's hiring pages state its strongest position: "Robotics can't industrialize without an evaluation layer. We're building it," and describe an evaluation platform that shows where a policy will fail and what data would fix it. The homepage instead leads with the category label "World models for robots", which several larger players also use. Visitors see what the technology is before they see the outcome they're buying.

    Medium impactLow effort First result in 2 weeks Co founder & CEO

    First action

    Write three headline options and test them on the next five sales calls.

    How to measure it

    Homepage conversion to pilot request

    Now
    –
    Target
    +30%
    By
    60 days

    Steps

    1. 1

      Draft headlines

      Three outcome-first options.

      Co-founder & CEO 1 hour

    2. 2

      Test in calls

      Ask five prospects which one describes their problem.

      Co-founder & CEO 1 week

    3. 3

      Rewrite the steps

      Rank checkpoints, find failures, fix with data.

      Co-founder & CEO 2 hours

    4. 4

      Ship

      Update the homepage copy.

      Co-founder & CTO 1 hour

    5. 5

      Align profiles

      YC profile, LinkedIn and email signature.

      Co-founder & CEO 1 hour

    6. 6

      Measure

      Pilot requests per homepage visit, before and after.

      Co-founder & CTO monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Co founder & CEO

    Smallest test

    Swap the headline and compare pilot requests for a month.

    Why it works

    Make the outcome the headline (for example, "Know where your policy fails before it touches a robot") and keep "world models for robots" as the line underneath. Rewrite the three steps around the buyer's decisions: rank checkpoints, find failures, get the data that fixes them. Use the same words on the YC profile, LinkedIn and outreach emails.

    Buyers search and talk in problems (eval time, robot time, failures), and an outcome headline sets One Robot apart from general world-model vendors.

    Risks and how to handle them

    • Narrowing to evaluation undersells training data

      Keep "generate training data" as step three on the page.

    Evidence

    • HiringY Combinator jobs page: "Robotics can't industrialize without an evaluation layer. We're building it." It lists four roles, including Founding Machine Learning - Eval Layer, which is not on the Ashby board; founding roles pay $150K-$275K with 0.10%-2.00% equity, and the intern role $6K-$8K a month.Source
    • Your websiteJobs page: One Robot builds task-specific world models and an evaluation platform for robot manipulation policies, to replace trial-and-error policy training with validation that shows where a policy will fail and what data would fix it; it is backed by Accel, YC, several exited founders and engineering leaders at leading AI companies, and is small by design, with everyone an individual contributor. The page text we could read lists no role titles or apply links.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
  • 04

    Most of One Robot's traffic will come from shared links: the YC profile, LinkedIn posts, investor and founder intros, Slack threads between robotics engineers. The page's meta description is only "World models for robots" and there is no social share image, so every shared link shows four words and a blank card.

    Medium impactLow effort First result in 1 week Co founder & CTO

    First action

    Export one striking frame pair from the rollout.

    How to measure it

    Visits from LinkedIn and referrals

    Now
    –
    Target
    +25%
    By
    30 days

    Steps

    1. 1

      Write the description

      Outcome plus focus, under 160 characters.

      Co-founder & CEO 30 minutes

    2. 2

      Make the image

      Real vs simulated frame with the headline.

      Co-founder & CEO 2 hours

    3. 3

      Add the tags

      Open Graph and Twitter card tags in the page head.

      Co-founder & CTO 30 minutes

    4. 4

      Check previews

      LinkedIn post inspector, Slack and iMessage.

      Co-founder & CTO 30 minutes

    5. 5

      Do the jobs page

      Its own description and image for recruiting shares.

      Co-founder & CTO 1 hour

    6. 6

      Reshare

      Post the scorecard (idea 1) with the new preview.

      Co-founder & CEO 1 hour

    First result
    1 week
    Cost
    $0
    Owner
    Co founder & CTO

    Smallest test

    Add the description and image today and reshare the homepage once.

    Why it works

    Write a 150-character description with the outcome and the focus ("Find where your manipulation policy fails, without robot time. Task-specific world models for folding, deformables and assembly."). Add a 1200x630 share image: a real-versus-simulated frame from the folding rollout with the headline. Check the preview in LinkedIn and Slack.

    A visual preview in a feed gets more clicks than a blank card, and a rollout frame shows the product before anyone clicks.

    Risks and how to handle them

    • LinkedIn caches the old preview

      Refresh it with LinkedIn's post inspector.

    Evidence

    • Your websiteSite basics (checked November 10, 2026): the meta description is only "World models for robots"; there is no social share image; /sitemap.xml and /robots.txt both return 404; there is no blog, pricing page or live chat; the homepage links the two founders' personal LinkedIn profiles but not the company LinkedIn page or the launch video.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
  • 05

    /sitemap.xml and /robots.txt both return 404, the site links only the founders' personal LinkedIn profiles, and nothing ties the domain to the company LinkedIn page, the launch video or the YC profile. With no public code, models or papers either, search engines and AI assistants have little to connect "One Robot" (a common phrase) to this company.

    Low impactLow effort First result in 1 month Co founder & CTO

    First action

    Add the sitemap and robots routes to the Next.js app.

    How to measure it

    Impressions for "one robot" queries in Search Console

    Now
    –
    Target
    +50%
    By
    60 days

    Steps

    1. 1

      Add sitemap

      Homepage, jobs and future pages.

      Co-founder & CTO 30 minutes

    2. 2

      Add robots file

      Allow crawling and point to the sitemap.

      Co-founder & CTO 15 minutes

    3. 3

      Add schema

      Organization with logo, founders and sameAs links.

      Co-founder & CTO 1 hour

    4. 4

      Link the footer

      Company LinkedIn, YC profile, launch video.

      Co-founder & CTO 30 minutes

    5. 5

      Submit

      Search Console sitemap submission.

      Co-founder & CTO 15 minutes

    6. 6

      Check monthly

      Branded queries and indexed pages.

      Co-founder & CTO monthly

    First result
    1 month
    Cost
    $0
    Owner
    Co founder & CTO

    Smallest test

    Ship the sitemap and footer links today.

    Why it works

    Generate a sitemap and robots file from the Next.js app, add Organization schema with sameAs links to the company LinkedIn page, the YC profile and the launch video, and link those from the footer. Submit the sitemap in Search Console and check what "one robot world models" returns.

    A generic name needs explicit links to be recognised as one entity, and an hour of work makes the company easier to find for prospects and candidates.

    Risks and how to handle them

    • Little search volume to gain yet

      It is an hour of work, and it matters more as content pages (ideas 8 and 12) are added.

    Evidence

    • Your websiteSite basics (checked November 10, 2026): the meta description is only "World models for robots"; there is no social share image; /sitemap.xml and /robots.txt both return 404; there is no blog, pricing page or live chat; the homepage links the two founders' personal LinkedIn profiles but not the company LinkedIn page or the launch video.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
    • ResearchPublic footprint: no Hugging Face organisation (huggingface.co/onerobot returns 404) and no company GitHub found; no public code, models, datasets or papers from One Robot; no blog or newsletter.Source
  • 06

    For a two-person company, hiring three founding engineers is the growth plan. The onerobot.io jobs page describes the mission and backers, but the page text we could read lists no role titles or apply links; the roles sit on Ashby. The YC jobs page also lists a fourth role, Founding Machine Learning - Eval Layer, which is not on the Ashby board, so candidates see different openings depending on where they look.

    Medium impactLow effort First result in 2 weeks Co founder & CEO

    First action

    Decide the status of the Eval Layer role today.

    How to measure it

    Applicants whose source is the company site

    Now
    –
    Target
    +50%
    By
    60 days

    Steps

    1. 1

      Reconcile roles

      Same list on YC, Ashby and the site.

      Co-founder & CEO 1 hour

    2. 2

      Write summaries

      Two lines per role, with pay and equity ranges.

      Co-founder & CEO 1 hour

    3. 3

      Add apply links

      Direct Ashby links, in the page text.

      Co-founder & CTO 1 hour

    4. 4

      Tag the source

      Source parameter on each link to count site applicants.

      Co-founder & CTO 15 minutes

    5. 5

      Add the open door

      An invitation for strong unsolicited candidates.

      Co-founder & CEO 15 minutes

    6. 6

      Review monthly

      Applicants by source and role.

      Co-founder & CEO monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Co founder & CEO

    Smallest test

    Add the three Ashby links to /jobs today.

    Why it works

    List each open role on /jobs with a two-line summary, the YC salary and equity ranges, and a direct Ashby apply link, written into the page itself rather than only loaded by script. Decide whether the Eval Layer role is open: post it on Ashby or remove it from YC. Add a "Not a fit? Send us a robot failure you can't explain" line for strong unsolicited candidates.

    Candidates who reach the site from a founder's post are the warmest applicants; showing roles, pay and a one-click apply link removes the step where they drop off.

    Risks and how to handle them

    • More unqualified applicants

      Keep the role requirements explicit (real-robot experience, PyTorch).

    Evidence

    • Your websiteJobs page: One Robot builds task-specific world models and an evaluation platform for robot manipulation policies, to replace trial-and-error policy training with validation that shows where a policy will fail and what data would fix it; it is backed by Accel, YC, several exited founders and engineering leaders at leading AI companies, and is small by design, with everyone an individual contributor. The page text we could read lists no role titles or apply links.Source
    • HiringY Combinator jobs page: "Robotics can't industrialize without an evaluation layer. We're building it." It lists four roles, including Founding Machine Learning - Eval Layer, which is not on the Ashby board; founding roles pay $150K-$275K with 0.10%-2.00% equity, and the intern role $6K-$8K a month.Source
    • HiringFounding Machine Learning - World Models (Ashby, San Francisco on-site, posted August 5, 2026): train action-conditioned video and dynamics models, improve long-horizon rollout quality, run multi-GPU training and build a world-model data engine; "We build world models that simulate manipulation scenes faithfully enough to validate" policies.Source
  • 07

    Real Chrome visitors already pass Core Web Vitals (main content at 1.1s for 75% of visits), so this is not urgent. But in Google's lab phone test the main content appears at 10.5s and the page weighs 4.4 MB, with the YouTube embed alone 979 KB over 9 requests and 702 KB of unused JavaScript. Engineers who open a shared link on a phone on slow mobile data get the slowest version of the page.

    Low impactLow effort First result in 1 day Co founder & CTO

    First action

    Swap the YouTube embed for a click-to-load poster.

    How to measure it

    PageSpeed mobile LCP

    Now
    –
    Target
    Under 4s, with field Core Web Vitals still passing
    By
    30 days

    Steps

    1. 1

      Poster the player

      Thumbnail that loads YouTube on click.

      Co-founder & CTO 1 hour

    2. 2

      Cut a loop

      Best 8 seconds, muted MP4 under 1 MB.

      Machine Learning Intern 1 hour

    3. 3

      Trim JavaScript

      Remove unused components from the homepage bundle.

      Co-founder & CTO 2 hours

    4. 4

      Re-test

      PageSpeed mobile and desktop.

      Co-founder & CTO 15 minutes

    5. 5

      Watch field data

      Confirm real-visitor vitals stay passing.

      Co-founder & CTO monthly

    6. 6

      Track plays

      How many visitors click the full video.

      Co-founder & CTO monthly

    First result
    1 day
    Cost
    $0
    Owner
    Co founder & CTO

    Smallest test

    Swap the embed for a poster and re-run PageSpeed.

    Why it works

    Replace the embedded player with a poster image that loads the player on click, and add a short, muted, looping MP4 of the best 8 seconds of the rollout above it. Trim unused JavaScript from the page bundle. Re-run PageSpeed and keep the field numbers passing.

    The rollout is the product demo; a clip that plays immediately shows it faster than a player that has to load first, and the page gets lighter for first-time mobile visitors.

    Risks and how to handle them

    • Fewer full video plays

      The loop shows the key moment; the full video is still one click away.

    Evidence

    • PageSpeedGoogle PageSpeed (November 10, 2026): mobile performance 55, desktop 89. In the lab phone test the main content (the paragraph "Physical robot availability limits how fast teams can iterate on training") appears at 10.5s; the page weighs 4.4 MB, with 2.0 MB of JavaScript of which 702 KB is unused (about 3.1s), and the YouTube embed alone is 979 KB over 9 requests. Real Chrome visitors pass Core Web Vitals: p75 LCP 1.1s (29% faster than before), INP 82 ms, CLS 0.03.Source

This quarter 9

The core plays for the next three months.

  • 08

    One Robot's pitch is finding failures, and its focus is deformable and contact-rich tasks. 1X publishes its world model's known failures (objects changing shape, disappearing, ignoring physics) next to its successes, which reads as honest and expert. One Robot has no public material at all on how policies fail on cloth and boxes, the thing its buyers lose robot hours to.

    Medium impactMedium effort First result in 6 weeks Founding Robot Learning

    First action

    Collect the clearest failure clips from the scorecard runs (idea 1).

    How to measure it

    Visits to the gallery and requests referencing it

    Now
    –
    Target
    1,000 visits and 3 requests a month
    By
    90 days

    Steps

    1. 1

      Collect clips

      From in-house and permitted runs.

      Founding Robot Learning 1 week

    2. 2

      Label failure types

      A short shared vocabulary for manipulation failures.

      Founding Robot Learning 2 days

    3. 3

      Mark predictions

      Did the world model predict each failure?

      Founding Machine Learning - World Models 2 days

    4. 4

      Add own failures

      Where the model drifts, and how it's flagged.

      Founding Machine Learning - World Models 2 days

    5. 5

      Publish

      Gallery page linked from the homepage.

      Co-founder & CTO 1 day

    6. 6

      Share

      One clip a week on LinkedIn and X.

      Co-founder & CEO weekly

    7. 7

      Refresh

      New clips monthly.

      Founding Robot Learning monthly

    First result
    6 weeks
    Cost
    Engineering time
    Owner
    Founding Robot Learning

    Smallest test

    Publish five labelled clips on one page.

    Why it works

    Publish a gallery page of 10-15 short clips, each labelled: the task, how the policy failed (missed grasp, cloth bunching, box flap re-opening), whether the world model predicted it, and what data fixed it. Include a section on the world model's own failure modes and how they are flagged. Add new clips monthly.

    Robot-learning engineers share concrete failure clips with each other, and showing your simulator's limits makes the successes believable.

    Risks and how to handle them

    • Showing model failures scares buyers

      Pair each with how it is detected; buyers already know every simulator has limits.

    Builds on

    Evidence

    • Competitor research1X World Model: a learned simulator trained on thousands of hours of humanoid video and action data to evaluate policies, including deformable items like laundry and a full t-shirt fold; 1X publishes the model's known failures (objects changing shape or disappearing, physics violations), released over 100 hours of video under Apache 2.0, and runs public challenges with a $10,000 prize, with an Evaluation Challenge listed as coming soon.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • HiringFounding Robot Learning (Ashby, San Francisco on-site, posted August 13, 2026): train manipulation policies (VLAs, diffusion policies, imitation and RL) and run them on real robots, validate the world model end to end and find where sim-to-real transfer fails; "We work with real customer data, real failures, and real deployment pressure." The postings say the company is expanding into policy training.Source
  • 09

    There is no pricing or offer anywhere on the site or in the launch post; every engagement starts as an open-ended conversation. For robotics teams, an unscoped pilot with a startup is hard to approve, while a small fixed-fee engagement with a defined deliverable can often be signed by the team lead.

    High impactMedium effort First result in 1 month Co founder & CEO

    First action

    Write the one-page sprint description and a fee to test on the next three calls.

    How to measure it

    Paid eval sprints

    Now
    –
    Target
    3 in the first quarter
    By
    90 days

    Steps

    1. 1

      Define the deliverable

      Model, evaluations, ranking, failures, data advice.

      Co-founder & CTO 1 day

    2. 2

      Set data requirements

      Minimum episodes, cameras and formats.

      Co-founder & CTO 1 day

    3. 3

      Price it

      Fixed fee, credited toward a contract.

      Co-founder & CEO 1 day

    4. 4

      Write the page

      Offer, timeline, sample report.

      Co-founder & CEO 2 days

    5. 5

      Test on calls

      Offer it to the next five qualified teams.

      Co-founder & CEO 1 month

    6. 6

      Template the report

      So each sprint takes less time.

      Founding Robot Learning 2 weeks

    7. 7

      Review

      Delivery hours and conversion to contracts.

      Co-founder & CEO monthly

    First result
    1 month
    Cost
    Founder time
    Owner
    Co founder & CEO

    Smallest test

    Offer the sprint on the next three calls and record the reaction to the price.

    Why it works

    Define one offer: send N recorded episodes for one task; within two weeks receive a task-specific world model, simulated evaluations of up to five policy checkpoints, a ranked comparison, the top failure modes with clips, and a recommendation for what data to collect. Set a fixed fee credited against a longer contract, and describe it on the site with the scorecard (idea 1) as proof.

    A fixed price and deliverable shortens approval, gives the team repeatable delivery to improve, and starts revenue before the platform is self-serve.

    Risks and how to handle them

    • Sprints turn into custom consulting

      Keep one supported task family per sprint and a fixed checkpoint limit.

    • Underpricing

      Track delivery hours per sprint and reprice after three.

    Builds on

    Evidence

    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • Your websiteSite basics (checked November 10, 2026): the meta description is only "World models for robots"; there is no social share image; /sitemap.xml and /robots.txt both return 404; there is no blog, pricing page or live chat; the homepage links the two founders' personal LinkedIn profiles but not the company LinkedIn page or the launch video.Source
    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
  • 10

    Step one on the homepage is "train a world model on your robot's data", and the World Models role is hired to build a world-model data engine. In practice the first delay in a pilot is usually getting a customer's recordings (cameras, actions, timestamps) into the training pipeline. Every day spent converting data is a day before the customer sees value.

    High impactMedium effort First result in 6 weeks Founding Machine Learning Wor…

    First action

    List the data formats from the last five conversations.

    How to measure it

    Time to first customer rollout

    Now
    –
    Target
    Under 5 working days
    By
    90 days

    Steps

    1. 1

      List formats

      From past conversations and pilot requests.

      Co-founder & CTO 1 day

    2. 2

      Write importers

      The top two or three formats.

      Founding Machine Learning - World Models 3 weeks

    3. 3

      Build the checker

      A script customers run before uploading.

      Machine Learning Intern 1 week

    4. 4

      Publish requirements

      Supported formats and minimum data on the site.

      Co-founder & CEO 1 day

    5. 5

      Time each pilot

      Data received to first rollout.

      Co-founder & CTO ongoing

    6. 6

      Add formats

      As new customers need them.

      Founding Machine Learning - World Models quarterly

    First result
    6 weeks
    Cost
    Engineering time
    Owner
    Founding Machine Learning World Models

    Smallest test

    Release the checker script for one format.

    Why it works

    Write importers for the two or three episode formats that pilot requests (idea 2) show most often, plus a validation script customers can run themselves that reports missing cameras, action dimensions and frame drops before they upload. Publish the supported formats and the checker on the site.

    Time to first useful rollout is the activation moment; removing the data-conversion step shortens every pilot and lets a small team run more of them.

    Risks and how to handle them

    • Format sprawl

      Support only formats requested by paying teams.

    Builds on

    Evidence

    • HiringFounding Machine Learning - World Models (Ashby, San Francisco on-site, posted August 5, 2026): train action-conditioned video and dynamics models, improve long-horizon rollout quality, run multi-GPU training and build a world-model data engine; "We build world models that simulate manipulation scenes faithfully enough to validate" policies.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
  • 11

    The founders found that evaluating a policy took longer than training it, and the intern post calls policy training "vibes-based today". A one-off evaluation solves that once; a team that trains a new checkpoint every few days needs it every time. YC's jobs page frames the company as the evaluation layer robotics needs to industrialize, which implies something that runs on every release, like tests in software.

    High impactHigh effort First result in 1 quarter Founding Robot Learning

    First action

    Define a fixed scenario set for one pilot customer and run two checkpoints through it by hand.

    How to measure it

    Active regression customers

    Now
    –
    Target
    3 teams running at least weekly
    By
    2 quarters

    Steps

    1. 1

      Fix scenario sets

      Per customer task, versioned.

      Founding Robot Learning 1 week

    2. 2

      Manual run

      Two checkpoints for one customer, by hand.

      Founding Robot Learning 1 week

    3. 3

      Design the report

      Success, new failures, fixed failures, clips.

      Co-founder & CTO 3 days

    4. 4

      Automate submission

      Upload or API call from their pipeline.

      Founding Machine Learning - World Models 1 month

    5. 5

      Price it

      Monthly per scenario set.

      Co-founder & CEO 1 week

    6. 6

      Track usage

      Runs per customer per week.

      Co-founder & CTO weekly

    First result
    1 quarter
    Cost
    Engineering time and compute
    Owner
    Founding Robot Learning

    Smallest test

    Send one pilot customer a manual regression report for their next checkpoint.

    Why it works

    Give each customer a simple way to submit a checkpoint (an upload or an API call from their training pipeline) and get back, within hours, a report comparing it with the last accepted checkpoint on a fixed scenario set: success rate, new failures, fixed failures and clips. Price it per month by scenario set.

    Recurring checks turn a pilot into a habit and a subscription, and each run adds data that makes the customer's world model better.

    Risks and how to handle them

    • Compute costs per run

      Cap rollouts per report and price by scenario set.

    • Customers' tasks change

      Version scenario sets and retrain on new data quarterly.

    Builds on

    Evidence

    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
    • HiringMachine Learning Intern (Ashby, San Francisco, posted July 24, 2026): "Training end-to-end policies for robots is vibes-based today." The intern contributes to training runs, runs real-robot evaluations, builds tooling and tests sim-to-real transfer; "real problems with real robots, not toy datasets".Source
    • HiringY Combinator jobs page: "Robotics can't industrialize without an evaluation layer. We're building it." It lists four roles, including Founding Machine Learning - Eval Layer, which is not on the Ashby board; founding roles pay $150K-$275K with 0.10%-2.00% equity, and the intern role $6K-$8K a month.Source
  • 12

    The job posts name hard, interesting problems: long-horizon rollout quality, action-conditioned video models, where sim-to-real transfer fails. None of this appears publicly: there is no blog, paper, code or model from One Robot. The people One Robot wants to hire and sell to (robot-learning engineers) judge teams by exactly this kind of technical writing.

    Medium impactMedium effort First result in 1 month Founding Machine Learning Wor…

    First action

    Turn the 60-second rollout into the first post: how it was made and where it breaks.

    How to measure it

    Post-sourced applicants and requests

    Now
    –
    Target
    5 a month combined
    By
    1 quarter

    Steps

    1. 1

      Pick topics

      Three posts from current engineering work.

      Co-founder & CTO 1 day

    2. 2

      Agree what to share

      Results and methods, not customer data.

      Co-founder & CEO 1 day

    3. 3

      Write post one

      The 60-second rollout: method and limits.

      Founding Machine Learning - World Models 1 week

    4. 4

      Add a blog section

      With RSS and the sitemap from idea 5.

      Co-founder & CTO 1 day

    5. 5

      Share

      Founders' LinkedIn, X and robotics communities.

      Co-founder & CEO per post

    6. 6

      Keep the cadence

      One post a month.

      Founding Machine Learning - World Models monthly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Founding Machine Learning World Models

    Smallest test

    Publish the first post and count applicants who mention it.

    Why it works

    Publish one technical post a month on onerobot.io, each built around a result: how rollout quality degrades over 60 seconds and what fixed it, how well the model responds to different actions from the same frame, a sim-to-real gap case study. End each post with the open roles and the pilot offer.

    One post serves recruiting, sales credibility and search at once, which matters for a team too small to do each separately.

    Risks and how to handle them

    • Giving away know-how

      Share results and lessons, keep training recipes private.

    Builds on

    Evidence

    • HiringFounding Machine Learning - World Models (Ashby, San Francisco on-site, posted August 5, 2026): train action-conditioned video and dynamics models, improve long-horizon rollout quality, run multi-GPU training and build a world-model data engine; "We build world models that simulate manipulation scenes faithfully enough to validate" policies.Source
    • ResearchPublic footprint: no Hugging Face organisation (huggingface.co/onerobot returns 404) and no company GitHub found; no public code, models, datasets or papers from One Robot; no blog or newsletter.Source
    • Your websiteJobs page: One Robot builds task-specific world models and an evaluation platform for robot manipulation policies, to replace trial-and-error policy training with validation that shows where a policy will fail and what data would fix it; it is backed by Accel, YC, several exited founders and engineering leaders at leading AI companies, and is small by design, with everyone an individual contributor. The page text we could read lists no role titles or apply links.Source
  • 13

    One Robot's own intern post describes the pain it solves: running real-robot evaluations, testing sim-to-real transfer, training that is "vibes-based". Other robotics companies describe the same work in their own job posts. A company hiring people to run robot evaluations or collect demonstrations for folding, packing or assembly is telling the world it has the problem One Robot sells against, right now.

    High impactLow effort First result in 2 weeks Co founder & CEO

    First action

    Find ten current posts that mention manipulation and robot evaluation.

    How to measure it

    Calls booked from job-signal outreach

    Now
    –
    Target
    4 a month
    By
    60 days

    Steps

    1. 1

      Define signals

      Keywords: manipulation, VLA, teleoperation, robot evaluation, folding, assembly.

      Co-founder & CEO 1 hour

    2. 2

      Weekly search

      Public job boards and company careers pages.

      Machine Learning Intern 2 hours a week

    3. 3

      Find the person

      Head of robot learning or the hiring manager.

      Co-founder & CEO weekly

    4. 4

      Write specifically

      One line on their task, the scorecard, the sprint.

      Co-founder & CEO weekly

    5. 5

      Track

      Signals, replies, calls, sprints.

      Co-founder & CEO weekly

    6. 6

      Refine

      Drop signals that never reply.

      Co-founder & CEO monthly

    First result
    2 weeks
    Cost
    Founder time
    Owner
    Co founder & CEO

    Smallest test

    Send ten emails this week and count replies.

    Why it works

    Each week, search public job boards for roles that mention manipulation policies, VLAs, teleoperation data collection or robot evaluation. Write to the hiring manager or head of robot learning with one specific line about the task their post describes, the scorecard (idea 1), and the sprint offer (idea 9). Keep a simple list of signals, replies and outcomes.

    Job posts are a public, current signal of budget and pain, and a founder writing about the reader's own task gets far more replies than a generic sequence.

    Risks and how to handle them

    • Feels intrusive

      Reference only what the public post says and offer something useful (the scorecard).

    Builds on

    Evidence

    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
    • HiringMachine Learning Intern (Ashby, San Francisco, posted July 24, 2026): "Training end-to-end policies for robots is vibes-based today." The intern contributes to training runs, runs real-robot evaluations, builds tooling and tests sim-to-real transfer; "real problems with real robots, not toy datasets".Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
  • 14

    Every customer starts from a specific arm, gripper and camera setup, and world models are trained on that robot's data. Runway's published evaluation, for instance, covers a single arm. Hardware makers and teleoperation-rig sellers want their customers to get working policies quickly, and a pre-trained world model for their platform is a reason to buy their hardware.

    Medium impactMedium effort First result in 1 quarter Co founder & CEO

    First action

    List the arms and teleoperation kits that appear most in pilot requests.

    How to measure it

    Pilot requests from partner channels

    Now
    –
    Target
    3 a quarter
    By
    2 quarters

    Steps

    1. 1

      Pick a platform

      Most common arm or kit in requests.

      Co-founder & CEO 1 week

    2. 2

      Approach the maker

      Offer a co-branded starter world model.

      Co-founder & CEO 2 weeks

    3. 3

      Collect base data

      Demonstrations on a standard task.

      Founding Robot Learning 3 weeks

    4. 4

      Train the base model

      Platform-specific starter.

      Founding Machine Learning - World Models 3 weeks

    5. 5

      Launch together

      Docs page, newsletter mention, webinar.

      Co-founder & CEO 1 week

    6. 6

      Track

      Leads and sprints by partner.

      Co-founder & CEO monthly

    First result
    1 quarter
    Cost
    Engineering time and possibly a loaned arm
    Owner
    Co founder & CEO

    Smallest test

    Pitch one hardware maker on a co-authored post using the scorecard data.

    Why it works

    Pick one widely used research arm or teleoperation kit. Train a base world model on it for a standard folding or pick-and-place task, and offer it with the hardware maker as a starter: buyers get a ready simulator and a discounted eval sprint. The maker features it in their docs and newsletter; One Robot gets a steady stream of teams already on a supported platform.

    It turns a partner's customer base into warm leads, and a base model per platform cuts the data each new customer needs.

    Risks and how to handle them

    • The partner wants exclusivity

      Agree a time-limited co-marketing window only.

    Builds on

    Evidence

    • Competitor researchRunway GWM-Robotics (published February 27, 2026): a world model for evaluating robot policies without hardware; Runway reports a 0.95 Pearson correlation between simulated and real-world progress scores across eight VLA policies on a Franka Panda arm (RoboArena tasks), from 1,450 simulated rollouts and over 16,000 human ratings, and notes it measures ranking, not absolute success rates; teams request access through a form.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
  • 15

    One Robot has no papers, public models or datasets, while 1X builds standing in the research community with an open dataset and prize challenges. Academic labs publish policy comparisons constantly and are short of robot time; their students go on to join the industrial teams One Robot sells to.

    Medium impactMedium effort First result in 1 quarter Co founder & CTO

    First action

    List five labs publishing on deformable manipulation and email one.

    How to measure it

    External publications using One Robot

    Now
    –
    Target
    2
    By
    2 quarters

    Steps

    1. 1

      List labs

      Deformable and contact-rich manipulation groups.

      Co-founder & CTO 1 day

    2. 2

      Write the offer

      Free runs, data rights, acknowledgement.

      Co-founder & CEO 1 day

    3. 3

      Sign up three

      Email principal investigators directly.

      Co-founder & CTO 1 month

    4. 4

      Run evaluations

      On their checkpoints and tasks.

      Founding Robot Learning 1 month

    5. 5

      Collect results

      Sim vs real numbers for the scorecard.

      Founding Robot Learning ongoing

    6. 6

      Recruit

      Invite strong students to apply for internships.

      Co-founder & CEO ongoing

    First result
    1 quarter
    Cost
    Compute and engineering time
    Owner
    Co founder & CTO

    Smallest test

    Run one lab's checkpoints and ask them to post the comparison.

    Why it works

    Offer three labs a fixed number of free simulated evaluation runs on their folding or manipulation policies, in exchange for reporting One Robot's simulated results next to their real-robot results in their papers or project pages. Collect permission to reuse the sim-versus-real data in the scorecard.

    Every paper that reports One Robot's results next to real-robot numbers is independent evidence of fidelity, and the students become future customers and hires.

    Risks and how to handle them

    • Academic timelines are slow

      Ask for a project-page result first, the paper later.

    Evidence

    • ResearchPublic footprint: no Hugging Face organisation (huggingface.co/onerobot returns 404) and no company GitHub found; no public code, models, datasets or papers from One Robot; no blog or newsletter.Source
    • Competitor research1X World Model: a learned simulator trained on thousands of hours of humanoid video and action data to evaluate policies, including deformable items like laundry and a full t-shirt fold; 1X publishes the model's known failures (objects changing shape or disappearing, physics violations), released over 100 hours of video under Apache 2.0, and runs public challenges with a $10,000 prize, with an Evaluation Challenge listed as coming soon.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
  • 16

    The job posts say the team works with "real customer data, real failures, and real deployment pressure", so there are customers. But the homepage and the launch post name no customer and report no result. For a two-person startup selling into engineering teams, one credible customer story is often the difference between "interesting" and "let's try it".

    High impactLow effort First result in 1 month Co founder & CEO

    First action

    Ask one customer this week whether they'd agree to a named or anonymised story.

    How to measure it

    Calls that become sprints

    Now
    –
    Target
    +25%
    By
    90 days

    Steps

    1. 1

      Ask

      Named or anonymised, with approval rights.

      Co-founder & CEO 1 day

    2. 2

      Gather numbers

      Robot-hours, failures found, time saved.

      Founding Robot Learning 1 week

    3. 3

      Interview

      30 minutes with their robotics lead.

      Co-founder & CEO 1 hour

    4. 4

      Write

      One page, one chart, one quote.

      Co-founder & CEO 2 days

    5. 5

      Approve

      Customer sign-off.

      Co-founder & CEO 1 week

    6. 6

      Publish and use

      Homepage, outreach and the sprint page.

      Co-founder & CEO 1 day

    First result
    1 month
    Cost
    $0
    Owner
    Co founder & CEO

    Smallest test

    Publish a three-number anonymised result box.

    Why it works

    Ask the happiest current customer for a short case study, named or anonymised by industry: the task, robot-hours per evaluation before and after, failures found in simulation that later showed up (or were avoided) on the real robot, and a quote from their robotics lead. Put it on the homepage under the scorecard.

    Peer proof answers the question technical buyers ask first (has anyone like us used this?), and anonymised stories are acceptable to most robotics teams.

    Risks and how to handle them

    • The customer declines

      Offer anonymised by industry and robot type.

    Evidence

    • HiringFounding Robot Learning (Ashby, San Francisco on-site, posted August 13, 2026): train manipulation policies (VLAs, diffusion policies, imitation and RL) and run them on real robots, validate the world model end to end and find where sim-to-real transfer fails; "We work with real customer data, real failures, and real deployment pressure." The postings say the company is expanding into policy training.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source

Big bets 4

Larger plays that can change the trajectory.

  • 17

    1X lists an Evaluation Challenge as "coming soon" and runs prize challenges on its own robot data, and Runway's study uses an existing benchmark on one tabletop arm. Nobody yet offers a standing, public benchmark for deformable folding scored in a world model. That is One Robot's focus task, and a leaderboard would bring every team working on it to One Robot's door.

    High impactHigh effort First result in 2 quarters Co founder & CTO

    First action

    Decide the scenario set and how submissions will run safely.

    How to measure it

    Unique teams on the leaderboard

    Now
    –
    Target
    15
    By
    2 quarters

    Steps

    1. 1

      Define the task

      Folding scenarios, success metric, rules.

      Co-founder & CTO 2 weeks

    2. 2

      Build submission

      Sandboxed policy interface.

      Founding Machine Learning - World Models 1 month

    3. 3

      Seed it

      Score open baseline policies first.

      Founding Robot Learning 2 weeks

    4. 4

      Real-robot checks

      Verify the top entries quarterly.

      Founding Robot Learning quarterly

    5. 5

      Launch

      With labs from idea 15 and partners from idea 14.

      Co-founder & CEO 1 week

    6. 6

      Convert

      Offer submitters the eval sprint on their own task.

      Co-founder & CEO ongoing

    First result
    2 quarters
    Cost
    Compute, robot time and a small prize
    Owner
    Co founder & CTO

    Smallest test

    Publish a leaderboard of five open baseline policies and invite submissions.

    Why it works

    Release a fixed folding scenario set and a submission path: teams submit a policy (or run it through a provided interface), One Robot scores it in the world model and verifies the top entries on a real robot each quarter. Publish rankings, real-robot checks and the correlation between them.

    A leaderboard is a self-renewing source of leads, credibility and data, and real-robot verification of the top entries is a fidelity proof competitors would have to match.

    Risks and how to handle them

    • Low participation

      Seed with lab partners and a modest prize.

    • Teams overfit the benchmark

      Hold out scenarios and rotate them.

    Builds on

    Evidence

    • Competitor research1X World Model: a learned simulator trained on thousands of hours of humanoid video and action data to evaluate policies, including deformable items like laundry and a full t-shirt fold; 1X publishes the model's known failures (objects changing shape or disappearing, physics violations), released over 100 hours of video under Apache 2.0, and runs public challenges with a $10,000 prize, with an Evaluation Challenge listed as coming soon.Source
    • Competitor researchRunway GWM-Robotics (published February 27, 2026): a world model for evaluating robot policies without hardware; Runway reports a 0.95 Pearson correlation between simulated and real-world progress scores across eight VLA policies on a Franka Panda arm (RoboArena tasks), from 1,450 simulated rollouts and over 16,000 human ratings, and notes it measures ranking, not absolute success rates; teams request access through a form.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
  • 18

    Step three on the homepage is "generate training data": synthetic trajectories that target the failures found. The jobs page says the platform shows what data would fix a failure, and the postings say the company is expanding into policy training. Evaluation tells a team what is wrong; the data that fixes it is what they would otherwise pay for in teleoperation hours, which makes it the larger budget.

    High impactHigh effort First result in 2 quarters Co founder & CEO

    First action

    For one pilot, generate trajectories for its top failure and measure the change.

    How to measure it

    Share of revenue from data packs

    Now
    –
    Target
    30%
    By
    3 quarters

    Steps

    1. 1

      Prove it once

      Targeted trajectories for one pilot's top failure.

      Founding Robot Learning 1 month

    2. 2

      Measure the gain

      Before and after, in sim and on the robot.

      Founding Robot Learning 2 weeks

    3. 3

      Package

      Pack size, format and turnaround.

      Co-founder & CTO 1 week

    4. 4

      Price

      Against teleoperation hours replaced.

      Co-founder & CEO 1 week

    5. 5

      Offer after every eval

      Built into the failure report.

      Co-founder & CEO ongoing

    6. 6

      Automate generation

      Part of the world-model data engine.

      Founding Machine Learning - World Models 1 quarter

    First result
    2 quarters
    Cost
    Compute and engineering time
    Owner
    Co founder & CEO

    Smallest test

    One free pack for one pilot, with a real-robot before-and-after.

    Why it works

    After each evaluation, offer a priced pack of synthetic trajectories aimed at the top failure modes, with a before-and-after evaluation on the same scenarios to show the improvement. Price per pack, or as a share of the teleoperation hours it replaces, and report the gain in the regression report (idea 11).

    Tying payment to a measured fix makes the value obvious and moves One Robot from an evaluation line item into the data budget.

    Risks and how to handle them

    • Synthetic data doesn't improve the real robot

      Only sell packs on tasks where a real-robot gain has been shown.

    Builds on

    Evidence

    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
    • Your websiteJobs page: One Robot builds task-specific world models and an evaluation platform for robot manipulation policies, to replace trial-and-error policy training with validation that shows where a policy will fail and what data would fix it; it is backed by Accel, YC, several exited founders and engineering leaders at leading AI companies, and is small by design, with everyone an individual contributor. The page text we could read lists no role titles or apply links.Source
    • HiringFounding Robot Learning (Ashby, San Francisco on-site, posted August 13, 2026): train manipulation policies (VLAs, diffusion policies, imitation and RL) and run them on real robots, validate the world model end to end and find where sim-to-real transfer fails; "We work with real customer data, real failures, and real deployment pressure." The postings say the company is expanding into policy training.Source
  • 19

    Runway's published result explicitly measures ranking of policies, not absolute success rates, and 1X lists ways its world model breaks (objects changing shape, disappearing, ignoring physics). A buyer's real fear is not that the simulator is imperfect, but that they can't tell when it is wrong. A score that says "trust this rollout" or "check this one on the robot" answers that fear directly.

    High impactHigh effort First result in 2 quarters Founding Machine Learning Wor…

    First action

    Compare the scorecard runs (idea 1) where sim and real disagreed, and look for signals that predicted it.

    How to measure it

    Agreement with the real robot for high-trust rollouts

    Now
    –
    Target
    90% or better
    By
    2 quarters

    Steps

    1. 1

      Collect disagreements

      Where sim and real differed.

      Founding Robot Learning 2 weeks

    2. 2

      Find signals

      Drift from training data, physics checks.

      Founding Machine Learning - World Models 1 month

    3. 3

      Build the score

      Per rollout, from the signals.

      Founding Machine Learning - World Models 1 month

    4. 4

      Calibrate

      Against held-out real-robot trials.

      Founding Robot Learning 1 month

    5. 5

      Ship in reports

      With a real-robot check list for low-trust cases.

      Co-founder & CTO 2 weeks

    6. 6

      Publish calibration

      Add it to the scorecard.

      Co-founder & CEO 1 day

    First result
    2 quarters
    Cost
    Research time and robot time
    Owner
    Founding Machine Learning World Models

    Smallest test

    Mark the five least trustworthy rollouts in the next report and check them on the robot.

    Why it works

    For each rollout, estimate how far it drifts from the training data and flag physical inconsistencies, then calibrate the score against real-robot outcomes so that, for example, rollouts marked high-trust match reality in 90% of cases. Show it in every report and route low-trust cases to a short real-robot check list.

    It turns the simulator's limits into a feature: teams spend scarce robot time only on the cases the model is unsure about, which is a clear, measurable saving.

    Risks and how to handle them

    • The score is poorly calibrated at first

      Ship it as "beta" with published calibration and update it quarterly.

    Builds on

    Evidence

    • Competitor researchRunway GWM-Robotics (published February 27, 2026): a world model for evaluating robot policies without hardware; Runway reports a 0.95 Pearson correlation between simulated and real-world progress scores across eight VLA policies on a Franka Panda arm (RoboArena tasks), from 1,450 simulated rollouts and over 16,000 human ratings, and notes it measures ranking, not absolute success rates; teams request access through a form.Source
    • Competitor research1X World Model: a learned simulator trained on thousands of hours of humanoid video and action data to evaluate policies, including deformable items like laundry and a full t-shirt fold; 1X publishes the model's known failures (objects changing shape or disappearing, physics violations), released over 100 hours of video under Apache 2.0, and runs public challenges with a $10,000 prize, with an Evaluation Challenge listed as coming soon.Source
  • 20

    YC lists One Robot's initial focus as textile and box folding, and the homepage targets deformable objects. Cloth is one of the hardest things for physics simulators to model and one of the most visible tasks in robotics demos; 1X shows t-shirt folding too. Open, general-purpose world models are becoming freely available and will be good enough for many rigid tasks, so a narrow, deep lead on cloth is easier to defend.

    High impactHigh effort First result in 2 quarters Co founder & CEO

    First action

    Count how many inbound requests and outreach replies are textile tasks.

    How to measure it

    Paying teams on textile tasks

    Now
    –
    Target
    5
    By
    3 quarters

    Steps

    1. 1

      Size demand

      Textile share of requests and replies.

      Co-founder & CEO 1 month

    2. 2

      Build the base

      Varied fabrics, sizes and grasps.

      Founding Machine Learning - World Models 1 quarter

    3. 3

      Build the library

      Standard textile scenarios.

      Founding Robot Learning 1 month

    4. 4

      Publish the scorecard

      Cloth-specific fidelity.

      Co-founder & CTO 2 weeks

    5. 5

      Sell into the vertical

      Garment, laundry and soft-goods automation teams.

      Co-founder & CEO ongoing

    6. 6

      Decide the next category

      Boxes or assembly, based on data.

      Co-founder & CEO 3 quarters

    First result
    2 quarters
    Cost
    Data collection, compute and engineering time
    Owner
    Co founder & CEO

    Smallest test

    Pitch a textile-only offer to five garment or laundry automation teams.

    Why it works

    Build a reusable base world model for textiles across fabric types, sizes and grasps, so each new garment or laundry customer needs far less data. Package it with a textile scenario library, a published fidelity scorecard for cloth, and a sales motion aimed at companies automating garment, laundry and soft-goods handling.

    Winning one hard category earns references, a data advantage and a reputation that carries into the next category, rather than competing broadly with general-purpose models.

    Risks and how to handle them

    • The textile market is too small early on

      Keep box folding as a parallel task and decide on data at six months.

    Builds on

    Evidence

    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".Source
    • Competitor research1X World Model: a learned simulator trained on thousands of hours of humanoid video and action data to evaluate policies, including deformable items like laundry and a full t-shirt fold; 1X publishes the model's known failures (objects changing shape or disappearing, physics violations), released over 100 hours of video under Apache 2.0, and runs public challenges with a $10,000 prize, with an Evaluation Challenge listed as coming soon.Source
    • Competitor researchNVIDIA Cosmos: open world foundation models (Cosmos 3 in 64B, 16B and 4B sizes) that developers post-train on their own embodiment, camera and sensor data, under the OpenMDW license, positioned to generate synthetic training data and "evaluate outcomes in a closed loop".Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source

Wildcards 2

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

  • 21

    The launch post's core complaint is that existing simulators are either not photorealistic or not physics-realistic, and the homepage promises environments "realistic to see and realistic to interact with". A single rollout video asks visitors to take that on trust. Letting people try to tell the difference themselves makes the point in seconds, and people share a test they nearly failed.

    Medium impactLow effort First result in 2 weeks Co founder & CEO

    First action

    Pick 10 matched real and simulated clip pairs from the scorecard runs.

    How to measure it

    Test completions

    Now
    –
    Target
    2,000
    By
    60 days

    Steps

    1. 1

      Pick pairs

      10 matched clips, same starting frame.

      Machine Learning Intern 2 days

    2. 2

      Build the page

      Pick-the-real-one, score, reveal.

      Co-founder & CTO 2 days

    3. 3

      Write the reveal

      Where physics differed and why.

      Founding Robot Learning 1 day

    4. 4

      Launch

      Founders' LinkedIn and X, robotics communities.

      Co-founder & CEO 1 day

    5. 5

      Publish the number

      How often people could tell, on the homepage.

      Co-founder & CEO after 2 weeks

    6. 6

      Refresh

      New pairs as the model improves.

      Machine Learning Intern quarterly

    First result
    2 weeks
    Cost
    $0
    Owner
    Co founder & CEO

    Smallest test

    Post three pairs as a poll on LinkedIn.

    Why it works

    Build a small page with 10 pairs of short clips from the same starting frame, one real and one generated, and ask visitors to pick the real one. Show their score against everyone else's, then reveal which were simulated and where the physics differed. End with the scorecard and the pilot request.

    It is interactive, shareable and on-message, and visitors' guessing accuracy becomes a public photorealism number.

    Risks and how to handle them

    • Seen as a gimmick by engineers

      Pair it with the physics reveal and the scorecard, so the test is backed by measured results.

    Builds on

    Evidence

    • Y CombinatorLaunch post: the founders found that evaluating a new policy took longer than training it, because of scene setup, trials and resets, and that existing simulators are either not photorealistic or not physics-realistic. The ask is to connect with teams training robots for real-world manipulation, by email. No customers or results are named.Source
    • Your websiteHomepage (checked November 10, 2026): H1 "World models for robots"; "We build simulation environments that are realistic to see and realistic to interact with"; three steps: train a world model on your robot's data, find failures by rolling out your policy, and generate synthetic trajectories that target those failures; "Built for hard manipulation": contact-rich manipulation, deformable objects and precise assembly; "See it in action" shows one 60-second autoregressive rollout; the only call to action is "Let's work together" with a mailto link to the founders' address; no customer logos, results, benchmarks, papers, blog, pricing or form; one "Backed by Y Combinator" badge.Source
  • 22

    All three open roles are on-site in San Francisco and need real-robot experience, and the company's thesis is that robotics needs an evaluation layer. The engineers One Robot wants to hire and the robot-learning leads it wants to sell to are largely the same small group, many of them in the Bay Area.

    Medium impactMedium effort First result in 1 month Co founder & CEO

    First action

    Invite 20 robot-learning engineers to the first evening.

    How to measure it

    Pilots or hires sourced from events

    Now
    –
    Target
    2
    By
    2 quarters

    Steps

    1. 1

      Set the format

      One failure, one live run, discussion.

      Co-founder & CEO 1 day

    2. 2

      Find a space

      Office, YC or a partner's lab.

      Co-founder & CEO 1 week

    3. 3

      Invite

      Founders' networks, YC, past applicants.

      Co-founder & CEO 2 weeks

    4. 4

      Prepare the run

      The presenting team's task in the world model.

      Founding Robot Learning 1 week

    5. 5

      Write it up

      A short post after each evening.

      Co-founder & CTO 1 day

    6. 6

      Follow up

      Candidates to Ashby, teams to the sprint offer.

      Co-founder & CEO 1 week

    First result
    1 month
    Cost
    A few hundred dollars per event
    Owner
    Co founder & CEO

    Smallest test

    One evening with 15 people.

    Why it works

    Host a small monthly evening: one team presents a failure they couldn't explain, One Robot runs it in the world model live, and the group discusses. Keep it to 30-40 robot-learning practitioners, invite through the founders' networks and YC, and publish a short write-up after each one.

    A room of the right 30 people every month produces hires, pilots and content, and costs little more than pizza.

    Risks and how to handle them

    • Presenters can't share data

      Use in-house or public tasks when needed.

    Evidence

    • HiringFounding Machine Learning - World Models (Ashby, San Francisco on-site, posted August 5, 2026): train action-conditioned video and dynamics models, improve long-horizon rollout quality, run multi-GPU training and build a world-model data engine; "We build world models that simulate manipulation scenes faithfully enough to validate" policies.Source
    • HiringY Combinator jobs page: "Robotics can't industrialize without an evaluation layer. We're building it." It lists four roles, including Founding Machine Learning - Eval Layer, which is not on the Ashby board; founding roles pay $150K-$275K with 0.10%-2.00% equity, and the intern role $6K-$8K a month.Source
    • Y CombinatorY Combinator profile: Winter 2026 batch, founded 2025, team of 2, San Francisco; world-model simulation for training and evaluating robot action policies (VLAs), with an initial focus on textile and box folding; the CEO previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL and Google, and the CTO spent five years at Tesla on factory automation and the Dojo team; the profile links a company LinkedIn page and a launch video, "One Robot Launch 2026".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.

  • Canopy Connect growth strategyInsurance data infrastructure: consumers share verified policy details, declaration pages and claims history from their carrier in…21 growth ideas
  • Plaud growth strategyAI note-taking hardware (Plaud Note, Note Pro, NotePin, NotePin S and the coming Plaud One) and the Plaud Intelligence software…23 growth ideas
  • Molagri growth strategyEngineers species-specific biopesticides from insect virus-like particles, built so the targeting element or the payload can be swapped…23 growth ideas
  • PST.AG growth strategyGlobal trade data for 160+ countries: customs tariffs and duty rates, free trade agreements, sanctions party lists and export control…22 growth ideas
  • Launch To Profit Media growth strategyA Shopify conversion rate optimization (CRO) agency: store audits, conversion-led redesigns, revenue optimization and A/B testing for…22 growth ideas
  • Osynth growth strategyCreates and publishes AI-generated short-form videos for brands on TikTok, Instagram Reels and YouTube Shorts: studies the niche, creates…21 growth ideas
  • Nessie growth strategyA context layer for AI-native teams: desktop apps for Mac, Windows and Linux, plus a CLI, daemon and MCP server. Together they ingest AI…23 growth ideas
  • Arize AI growth strategyAn AI and agent engineering platform for observability and evaluation: Arize AX (managed, with Signal, Alyx and ADB) and Phoenix (open…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. One Robot did not commission, review or endorse it, and Programmable is not affiliated with One Robot. 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 One Robot runs its business.

Something wrong, or do you represent One Robot? 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

    Prove fidelity and fix the front door: the sim-to-real scorecard, a scoped pilot request, the evaluation-layer headline, link previews, findability, the jobs page and a lighter demo embed.

  2. Days 31 to 607 ideas

    Package and publish: the fixed-fee eval sprint, data import with no reformatting, the failure gallery, technical write-ups, intent-led outreach, the first customer result and the real-or-simulated test.

  3. Days 61 to 908 ideas

    Make evaluation a habit and open channels: per-checkpoint regression reports, hardware and lab partners, the evaluation night; commit to the big bets (leaderboard, fix-it data packs, trust score, textile beachhead).

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 testing policies without tying up physical robots. The most cited problem: evaluating a new policy takes longer than training it.

  • Finding failure cases before deployment
  • Scene setup, trials and resets on real robots
  • In their words: “robot time”

Market

Competitive landscape

4 competitors compared: Runway (GWM-Robotics), 1X World Model, NVIDIA Cosmos and Traditional physics simulators.

  • Gap to exploit: Published fidelity numbers on deformable, contact-rich tasks
  • Runway (GWM-Robotics): Measures ranking, not absolute success
  • 1X World Model: Built around 1X's own robot

Hiring signals

One Robot is a two-person team hiring its first engineers, all on-site in San Francisco, and the roles map onto the product: a world-model engineer to improve long-horizon rollouts and build the data engine (ideas 10…

  • Founding Machine Learning - World Models: Long-horizon rollouts and a world-model data engine. Ideas 10, 12, 19.
  • Founding Robot Learning: Validate the world model on real robots; expand into policy training. Ideas 1, 8, 16, 18.
  • Machine Learning Intern: Real-robot evaluations; policy training is "vibes-based today". Ideas 11, 13.

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