ProgrammableMarketing/collinear.ai (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 · 9
  • Big bets · 3
  • Wildcards · 3
  • How we can help
Generated Oct 18, 2026
  1. Programmable
  2. Playbooks
  3. Collinear AI
Collinear AI logo

growth playbook · Oct 18, 2026

collinear.ai (opens in a new tab)

Collinear AI growth playbook

Diagnosis

Collinear is research-led and growing fast: 14 open roles, half in research, opening new tracks in biosafety, robotics, cybersecurity and AI physics.

Its marketing hires (a Growth Marketer for lists, campaigns, events and swag, a Digital Marketing Manager for X and LinkedIn measured on qualified leads, and a Founding Marketing Manager for positioning, case studies and competitor comparisons) say it is building its first real go-to-market.

It has strong assets (CWE-bench, 18K off-the-shelf tasks, founders from Hugging Face and Deloitte) but they convert poorly: benchmarks have no sign-up, samples need a demo, the site names no customers, and only 17 pages are indexed. This playbook focuses those hires on a market of a few hundred named researchers.

Growth Score

6/10

Biggest Opportunity

Treat the benchmarks as the front door and the account list as the market: capture every researcher who reads a result, give them samples in hours, and reach the rest by name at events, by gift and through matched ads.

Biggest Risk

A very small set of frontier-lab buyers means a few lost accounts move the business; without enterprise and self-serve buyers, and with competitors building similar environments, growth depends on a handful of relationships.

01

Start here

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

  1. 1

    Turn "start with five samples" into a sample request that delivers in hours

    Package five ready samples per capability so they can be sent without new work.

    High impact First result in 2 weeks
  2. 2

    Build the account list from the authors of recent post-training, RL and agent papers

    List the labs and teams, then pull authors from their last 12 months of post-training and agent papers.

    High impact First result in 1 week
  3. 3

    Capture the researchers who visit CWE-bench: results alerts and a "test your model" form

    Add the email capture and the evaluation form to cwe-bench.com, connected to the CRM.

    High impact First result in 2 weeks

02

Your idea portfolio

21 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.

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

    2 of these are things Programmable can build and run for you.

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 · 4Big bets · 5Fill-ins · 3Reconsider · 9
Effort →

Showing 21 of 21 ideas

This week 6

Quick wins with a clear first action.

  • 01

    CWE-bench is the most public thing Collinear has made, read by exactly the post-training and security researchers it sells to. The site has no sign-up, and submitting a model means finding a contact.

    High impactLow effort First result in 2 weeks Growth Marketer

    First action

    Add the email capture and the evaluation form to cwe-bench.com, connected to the CRM.

    How to measure it

    Evaluation requests from benchmark sites

    Now
    –
    Target
    10 a month
    By
    6 weeks

    Steps

    1. 1

      Add the forms

      Results alerts and evaluation request, three fields each.

      Growth Marketer 1 day

    2. 2

      Connect the CRM

      Lab, role and model fields mapped.

      Growth Marketer 2 hours

    3. 3

      Same-day follow-up

      Evaluation requests go to a researcher within a day.

      Co-founder and CEO ongoing

    4. 4

      Repeat on FrontierCUA

      The same forms on every benchmark site.

      Growth Marketer 1 day

    5. 5

      Track sources

      Which benchmark and which result page brought each lead.

      Digital Marketing Manager 2 hours

    6. 6

      Report monthly

      Sign-ups, requests and meetings from benchmarks.

      Growth Marketer monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Growth Marketer

    Smallest test

    Add the results-alert form only and count sign-ups for two weeks.

    Why it works

    Add two things to cwe-bench.com: "Email me when results change" (name, work email, lab) and a short "Evaluate your model" form that creates a CRM record with the model and harness. Route both to the team the same day.

    The Growth Marketer is measured on campaigns with a clear tracking setup and follow-up. A benchmark leaderboard without capture is traffic with no follow-up; with it, it becomes the best lead source Collinear owns.

    Risks and how to handle them

    • Researchers avoid forms

      Keep them short and promise only results emails.

    Evidence

    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
  • 02

    Calico asks buyers to "Start with five representative samples", and the homepage promises data "hours away, not months". The path to a sample is a demo booking, which adds days.

    High impactLow effort First result in 2 weeks Founding Marketing Manager

    First action

    Package five ready samples per capability so they can be sent without new work.

    How to measure it

    Sample requests that become paid orders

    Now
    –
    Target
    25%
    By
    end of quarter

    Steps

    1. 1

      Pre-package samples

      Five per capability with environment and verifier.

      MTS - Product 3 days

    2. 2

      Build the form

      Capability, model, harness, work email.

      Growth Marketer 1 day

    3. 3

      Set delivery

      Secure link within hours on weekdays.

      MTS - Product 1 day

    4. 4

      Follow up at 48 hours

      A researcher asks how the samples ran.

      MTS - Research ongoing

    5. 5

      Qualify

      Lab, volume and timeline recorded in the CRM.

      Growth Marketer ongoing

    6. 6

      Measure

      Request to order rate by capability.

      Founding Marketing Manager monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Founding Marketing Manager

    Smallest test

    Add the form to the cyber page only.

    Why it works

    Put a "Get five samples" form on each Calico capability page (cyber, MCP, computer use, DeepSWE, GDPval): pick the capability, the model being trained and the harness, and receive a sample package with tasks, environments and verifiers within hours.

    Fast samples prove the speed claim and let a researcher test on their own model before a sales conversation. Every request is a qualified lead with a stated capability gap.

    Risks and how to handle them

    • Samples leak or are used without buying

      Samples are representative, not held-out; terms on the download.

    Evidence

    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source
  • 03

    The Growth Marketer is asked to "build and maintain accurate account, contact, and audience lists". Collinear sells to a small number of frontier labs and AI teams, and the people who decide what data to buy publish papers and model cards.

    High impactLow effort First result in 1 week Growth Marketer

    First action

    List the labs and teams, then pull authors from their last 12 months of post-training and agent papers.

    How to measure it

    Target researchers reached

    Now
    –
    Target
    60% of the list
    By
    end of quarter

    Steps

    1. 1

      List the teams

      Frontier labs and agent teams by region.

      Co-founder and CEO 2 hours

    2. 2

      Pull authors

      Post-training, RL, agents and evals papers from the last year.

      Growth Marketer 3 days

    3. 3

      Tag areas

      Cyber, SWE, computer use, knowledge work, behavior.

      Growth Marketer 1 day

    4. 4

      Add to the CRM

      One account per team, contacts by area.

      Growth Marketer 1 day

    5. 5

      Mark warm paths

      Founders' and researchers' existing relationships.

      Co-founder 2 hours

    6. 6

      Refresh monthly

      New papers and job moves.

      Growth Marketer monthly

    First result
    1 week
    Cost
    $0
    Owner
    Growth Marketer

    Smallest test

    Build the list for five labs and check it against what the founders know.

    Build the prospect list. The list compiled from public sources, verified, deduplicated and delivered to your CRM. Usually live within days.

    Book a working session

    Why it works

    Build a list of every lab and team training frontier or agent models, and under each the researchers named on recent post-training, RL, agent and evaluation papers and model cards, with their area. Use it for outreach, events, gifting and ads targeting.

    A list of a few hundred named researchers covers most of the market. Every campaign in this playbook (events, gifts, webinars, ads) works better aimed at those names than at job titles.

    Risks and how to handle them

    • Researchers dislike cold outreach

      Lead with their own research area and a relevant result, never a generic pitch.

    Evidence

    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • Your websiteResearch page: areas in world models, recursive self-improvement, user simulation, long-horizon agents and post-training, with sections for benchmarks, publications and field notes.Source
  • 04

    CWE-bench v1 shows three frontier models tied at 68% and one at 67%. That is a story researchers and AI watchers share, and the CEO's Hugging Face and Salesforce research background gives it weight.

    Medium impactLow effort First result in 1 week Co founder and CEO

    First action

    Write the CWE-bench v1 thread from the current leaderboard.

    How to measure it

    Benchmark alert sign-ups from threads

    Now
    –
    Target
    200
    By
    6 weeks

    Steps

    1. 1

      Pick the angle

      One surprising finding per result.

      MTS - Research 1 hour

    2. 2

      Make the chart

      One clean chart, branded lightly.

      Digital Marketing Manager 1 hour

    3. 3

      Draft the thread

      Five posts, plain language.

      Founding Marketing Manager 1 hour

    4. 4

      CEO posts

      In their own voice, replying for the first hour.

      Co-founder and CEO 1 hour

    5. 5

      Promote

      Boost on X and LinkedIn to the target list.

      Digital Marketing Manager 30 minutes

    6. 6

      Measure

      Visits and sign-ups by thread.

      Digital Marketing Manager weekly

    First result
    1 week
    Cost
    $0 to $500 in promotion per thread
    Owner
    Co founder and CEO

    Smallest test

    Post the CWE-bench v1 thread and measure sign-ups.

    Build the prospect list. The list compiled from public sources, verified, deduplicated and delivered to your CRM. Usually live within days.

    Book a working session

    Why it works

    Within two days of each benchmark update, the CEO posts a five-part thread on X and LinkedIn: the result, one chart, one surprising failure, what it means for training, and a link to the leaderboard with the alerts form.

    The Digital Marketing Manager is asked to run organic social on X and LinkedIn. Research-led posts from a known researcher reach the buyers directly and feed the benchmark sign-up.

    Risks and how to handle them

    • Labs object to how their model is shown

      Report the method, give labs a preview, and correct errors fast.

    Builds on

    Evidence

    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • HiringDigital Marketing Manager (Bengaluru, posted October 13, 2026): paid and organic social on X and LinkedIn, email and landing pages; map funnel passthrough to qualified lead; optimise cost per lead, cost per qualified lead and pipeline contribution.Source
  • 05

    The site says Collinear is trusted by leading frontier labs but names none, and the Founding Marketing Manager is hired to produce case studies. Buyers in a small market ask who else uses it.

    High impactLow effort First result in 3 weeks Founding Marketing Manager

    First action

    Ask the two longest-standing customers what they would allow Collinear to say.

    How to measure it

    Homepage to sample request

    Now
    –
    Target
    +30% relative
    By
    6 weeks

    Steps

    1. 1

      Ask customers

      What can be said: name, quote, numbers or a description.

      Co-founder and CEO 1 week

    2. 2

      Write the proof

      One sentence with a number.

      Founding Marketing Manager 1 hour

    3. 3

      Get approval

      Written sign-off from the customer.

      Co-founder and CEO 1 week

    4. 4

      Place it

      Homepage hero, Calico pages, outreach emails.

      Growth Marketer 2 hours

    5. 5

      Build a one-pager

      The longer story for sales conversations.

      Founding Marketing Manager 2 days

    6. 6

      Repeat each quarter

      One new proof point per quarter.

      Founding Marketing Manager quarterly

    First result
    3 weeks
    Cost
    $0
    Owner
    Founding Marketing Manager

    Smallest test

    Use a described result in ten outreach emails before changing the site.

    Why it works

    Get one customer to agree to a named quote or a described result ("A frontier lab used 3,000 DeepSWE tasks and measured a lift on held-out agent tasks"), with the numbers they allow, and put it on the homepage and every Calico page.

    Labs keep vendors private, so even a described result is rare and credible. It answers the first question a new buyer asks and makes every outreach email stronger.

    Risks and how to handle them

    • No customer will be named

      Use a described result with the lab type, which is still more specific than "trusted by".

    Evidence

    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source
    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 06

    The research page has sections for benchmarks, publications and field notes, but the sitemap lists only 17 URLs and no individual research pages. Researchers searching for a benchmark or a method can't find Collinear's work.

    Medium impactLow effort First result in 4-8 weeks Growth Marketer

    First action

    List every research item now on the research page and give each a page.

    How to measure it

    Organic visits to research pages

    Now
    –
    Target
    2,000 a month
    By
    3 months

    Steps

    1. 1

      Inventory

      Every field note, paper and benchmark.

      Growth Marketer 2 hours

    2. 2

      Template

      Title, summary, authors, links, related capability.

      Growth Marketer 1 day

    3. 3

      Publish

      One page per item in Webflow.

      Growth Marketer 2 days

    4. 4

      Sitemap

      Add the collection to the sitemap.

      Growth Marketer 30 minutes

    5. 5

      Link to Calico

      Each page links to the matching task pack.

      Growth Marketer 2 hours

    6. 6

      Monthly field note

      One new note from a researcher each month.

      MTS - Research monthly

    First result
    4 to 8 weeks
    Cost
    $0
    Owner
    Growth Marketer

    Smallest test

    Publish three pages and check they are indexed in two weeks.

    Why it works

    Publish each field note, paper summary and benchmark write-up as its own page with a title, summary, authors and links, add them to the sitemap, and link each to the related Calico capability.

    Research is how Collinear earns trust with researchers. Separate pages are what search engines and AI assistants can find, cite and link to.

    Risks and how to handle them

    • Thin pages

      Only publish items with a real summary and findings.

    Evidence

    • programmable.infoThe sitemap lists 17 URLs: the homepage, research, about, careers and six Calico pages, with no individual research, field-notes or case-study pages. There is a PyPI package, collinear, last released in January 2026, and docs on Mintlify.Source
    • Your websiteResearch page: areas in world models, recursive self-improvement, user simulation, long-horizon agents and post-training, with sections for benchmarks, publications and field notes.Source
    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source

This quarter 9

The core plays for the next three months.

  • 07

    The Growth Marketer runs "select events and community programs" with budgets and on-site execution. Most of Collinear's buyers are in one place at NeurIPS in December.

    High impactMedium effort First result in December Growth Marketer

    First action

    Book a private room near the NeurIPS venue for one evening.

    How to measure it

    Sample requests from dinner guests

    Now
    –
    Target
    8
    By
    30 days after the dinner

    Steps

    1. 1

      Book the room

      Private room, 20 seats, near the venue.

      Growth Marketer 1 week

    2. 2

      Invite from the list

      40 invites by name, from the founders where there is a relationship.

      Co-founder and CEO 2 weeks

    3. 3

      Plan the talk

      15 minutes on frontier failures with new results.

      Co-founder and CEO 1 week

    4. 4

      Prepare follow-up

      Personal email and sample offer per guest.

      Growth Marketer 2 days

    5. 5

      Run the evening

      Logistics, name cards, swag.

      Growth Marketer 1 evening

    6. 6

      Follow up within a week

      Every guest, with their area's samples.

      MTS - Research 1 week

    First result
    December
    Cost
    $5,000 to $10,000
    Owner
    Growth Marketer

    Smallest test

    Host a 10-person dinner in San Francisco first.

    Why it works

    Host one small dinner for 20 researchers from the account list during NeurIPS, with a short talk from the CEO on where frontier agents fail (from CWE-bench and FrontierCUA), and follow up with each guest within a week offering samples.

    In a market of a few hundred buyers, a dinner with 20 of them is worth more than a booth. The follow-up plan is what the job post's success measure asks for.

    Risks and how to handle them

    • Low acceptance during a busy week

      Invite early and pick the least crowded evening.

    Builds on

    Evidence

    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
  • 08

    Collinear is hiring researchers for biosafety, robotics simulation, cybersecurity and AI physics. CWE-bench showed the pattern: a public benchmark makes a new capability visible, and a sample pack makes it buyable.

    High impactMedium effort First result in 1 month MTS Research

    First action

    Pick the track closest to ready and set a launch date.

    How to measure it

    Sample requests per track launch

    Now
    –
    Target
    10
    By
    30 days after each launch

    Steps

    1. 1

      Pick the order

      Launch tracks by readiness and demand.

      Co-founder and CEO 1 hour

    2. 2

      Build the benchmark

      Small held-out set, verifier, leaderboard.

      MTS - Research 4 weeks

    3. 3

      Package samples

      Five tasks per track.

      MTS - Product 1 week

    4. 4

      Write it up

      Method and first results.

      MTS - Research 3 days

    5. 5

      Launch

      Thread, page, outreach to the area's researchers.

      Founding Marketing Manager 1 week

    6. 6

      Follow up

      Sample requests answered within hours.

      Growth Marketer ongoing

    First result
    1 month per track
    Cost
    Research time per track
    Owner
    MTS Research

    Smallest test

    Launch one track with a write-up and samples, without a public leaderboard.

    Why it works

    For each new track, publish a small held-out benchmark with a leaderboard, a short write-up, and a five-sample pack on Calico, launched with a CEO thread and outreach to the researchers on the list who work in that area.

    New tracks need buyers to know they exist. A benchmark gets attention and citations, and the sample pack turns attention into requests.

    Risks and how to handle them

    • Biosafety results could reveal sensitive information

      Publish scores and categories only, never the hazardous tasks.

    Builds on

    Evidence

    • HiringCareers page: 14 open roles (Research 7, Engineering 5, GTM 1, Product 1), including new research tracks in robotics simulation data, biosafety simulations, cybersecurity and AI physics (fluid dynamics, structural mechanics, semiconductor design); the MTS Product role works "directly with researchers and engineers at leading AI labs".Source
    • HiringMTS - Research (Biosafety Simulations): "We are extending our environments into biosafety"; turn the judgment of biosafety professionals into tasks, reference answers and graders, and "Red-team the boundary" between helpful biology and operational uplift. No prior AI experience is required.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 09

    The Growth Marketer manages "company gifting and swag", and Collinear's beanies and tees are "not set up as a shop yet". A gift to a named researcher gets opened; an email often doesn't.

    Medium impactMedium effort First result in 4 weeks Growth Marketer

    First action

    Pick the first 25 recipients by area and order the boxes.

    How to measure it

    Gift codes redeemed

    Now
    –
    Target
    30%
    By
    6 weeks after sending

    Steps

    1. 1

      Pick recipients

      100 researchers by area, with addresses where known.

      Growth Marketer 1 week

    2. 2

      Design the box

      Swag plus a card with link and code.

      Growth Marketer 1 week

    3. 3

      Match samples

      Each code unlocks five tasks in the recipient's area.

      MTS - Product 2 days

    4. 4

      Write notes

      A short personal note per recipient.

      MTS - Research 2 days

    5. 5

      Send in waves

      25 a week.

      Growth Marketer 4 weeks

    6. 6

      Follow up

      One email two weeks after delivery.

      Growth Marketer ongoing

    First result
    4 weeks
    Cost
    $80 to $150 per box
    Owner
    Growth Marketer

    Smallest test

    Send 10 boxes and measure redemptions.

    Why it works

    Send 100 researchers from the account list a box with Collinear swag and a card with a link and code to run five tasks from their area against their own model, with a personal note from a Collinear researcher.

    It combines the two things that work with researchers: something personal and something they can run. The code ties each box to a sample request.

    Risks and how to handle them

    • Labs restrict gifts

      Keep boxes low value and send to home addresses only when offered.

    Builds on

    Evidence

    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 10

    Buyers compare RL environment and data vendors, and directories already list Collinear next to AfterQuery, Bespoke Labs, Huzzle Labs and Fleet AI. The Founding Marketing Manager is hired to write competitor comparisons.

    Medium impactLow effort First result in 1 month Founding Marketing Manager

    First action

    Write down the last five competitors buyers mentioned on calls.

    How to measure it

    Win rate against named competitors

    Now
    –
    Target
    +15 points
    By
    2 quarters

    Steps

    1. 1

      Collect mentions

      Competitors named in calls.

      Co-founder and CEO 1 hour

    2. 2

      Research each

      Public offers, pricing signals, benchmarks.

      Founding Marketing Manager 1 week

    3. 3

      Write cards

      One page each, fair and specific.

      Founding Marketing Manager 1 week

    4. 4

      Review with research

      Check technical claims.

      MTS - Research 2 hours

    5. 5

      Share

      In the CRM and before every call.

      Founding Marketing Manager 1 hour

    6. 6

      Update quarterly

      New vendors and changes.

      Founding Marketing Manager quarterly

    First result
    1 month
    Cost
    $0
    Owner
    Founding Marketing Manager

    Smallest test

    Write cards for the two most-mentioned vendors.

    Why it works

    One page per alternative: what they offer, where Collinear is stronger (off-the-shelf volume, verified rewards, held-out evals, enterprise runtimes), where they are, and the questions to ask. Share them with everyone who talks to buyers.

    In a young category, buyers form views fast. Prepared answers make every call consistent and shorter.

    Risks and how to handle them

    • Cards go stale

      Date each card and review quarterly.

    Evidence

    • Competitor researchAn RL vendor directory ranks Collinear #12 and describes it as now positioned "mainly as a supplier of training data, environments and evals to frontier labs"; related vendors named are AfterQuery, Bespoke Labs, Huzzle Labs and Fleet AI.Source
    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 11

    The Digital Marketing Manager is asked to optimise cost per qualified lead and pipeline contribution on X and LinkedIn. With a few hundred buyers, broad targeting wastes most of the budget.

    Medium impactMedium effort First result in 4 weeks Digital Marketing Manager

    First action

    Upload the first 300 contacts as a matched audience.

    How to measure it

    Qualified meetings from paid social

    Now
    –
    Target
    10
    By
    8 weeks

    Steps

    1. 1

      Build audiences

      Matched audiences from the list, by area.

      Digital Marketing Manager 1 day

    2. 2

      Write three offers

      Sample pack, benchmark results, field note.

      Founding Marketing Manager 2 days

    3. 3

      Set tracking

      UTM to form to CRM to meeting.

      Digital Marketing Manager 1 day

    4. 4

      Launch small

      Fixed weekly budget per ad set.

      Digital Marketing Manager 1 hour

    5. 5

      Report weekly

      Cost per lead, per qualified lead and per meeting.

      Digital Marketing Manager weekly

    6. 6

      Cut and scale

      Keep the offers that produce meetings.

      Digital Marketing Manager weekly

    First result
    4 weeks
    Cost
    $3,000 to $6,000 over eight weeks
    Owner
    Digital Marketing Manager

    Smallest test

    One ad set for one area for four weeks.

    Why it works

    Upload the account list as matched audiences on LinkedIn and X, run three ad sets by area (cyber, agents and computer use, SWE) offering the area's sample pack or benchmark results, and judge them only on qualified meetings.

    Matched audiences keep spend on the buyers, and offers from benchmarks and samples are useful to researchers rather than generic brand ads.

    Risks and how to handle them

    • Audiences too small to deliver

      Expand to lookalikes at the same companies.

    Builds on

    Evidence

    • HiringDigital Marketing Manager (Bengaluru, posted October 13, 2026): paid and organic social on X and LinkedIn, email and landing pages; map funnel passthrough to qualified lead; optimise cost per lead, cost per qualified lead and pipeline contribution.Source
    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 12

    The Founding Marketing Manager role says Collinear serves "leading AI labs and Fortune 500 enterprises", and Calico covers 70+ enterprise runtimes (customer service, IT service management, HR, finance). Enterprises deploying agents need to know where they will fail.

    High impactMedium effort First result in 1 quarter Founding Marketing Manager

    First action

    Interview three enterprise teams deploying customer service or IT agents about how they test them.

    How to measure it

    Enterprise agent-readiness engagements

    Now
    –
    Target
    3
    By
    2 quarters

    Steps

    1. 1

      Interview buyers

      Three enterprise agent teams.

      Co-founder 2 weeks

    2. 2

      Define the scope

      Fixed tasks, report and price.

      MTS - Product 1 week

    3. 3

      Write the page

      For enterprise AI and IT leaders.

      Founding Marketing Manager 3 days

    4. 4

      Sample report

      An example readiness report on a public agent.

      MTS - Research 2 weeks

    5. 5

      Outreach

      Through the co-founder's consulting network.

      Co-founder 4 weeks

    6. 6

      Case study

      First engagement, with permission.

      Founding Marketing Manager 2 weeks

    First result
    1 quarter
    Cost
    Research and product time
    Owner
    Founding Marketing Manager

    Smallest test

    Run one readiness report for a friendly enterprise team.

    Why it works

    Package a fixed-scope "agent readiness" engagement for enterprises: run their agent through simulations of their own tools (ServiceNow, Salesforce, Jira), report failures by task, and deliver training data to fix them. Write one page and one case study.

    It opens a much larger buyer pool than frontier labs, with clear budgets for agent rollouts, using environments Collinear has already built.

    Risks and how to handle them

    • Distracts from labs

      Fixed scope only, delivered from existing environments.

    Evidence

    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
  • 13

    The Founding Marketing Manager runs webinar campaigns, and Collinear's research focus is finding where models fail. A monthly session on one failure area is useful to researchers and gives the team a reason to invite the list.

    Medium impactMedium effort First result in 1 month Founding Marketing Manager

    First action

    Pick the first topic and researcher and set a date four weeks out.

    How to measure it

    Target-account registrants per webinar

    Now
    –
    Target
    40
    By
    third session

    Steps

    1. 1

      Pick topics

      Three months of failure areas.

      MTS - Research 1 hour

    2. 2

      Set up registration

      Form into the CRM.

      Growth Marketer 2 hours

    3. 3

      Invite the list

      Personal invites by area.

      Growth Marketer 2 hours a month

    4. 4

      Present

      30 minutes plus questions.

      MTS - Research 1 hour a month

    5. 5

      Clip it

      Three clips for X and LinkedIn.

      Digital Marketing Manager 2 hours a month

    6. 6

      Follow up

      Samples offered to attendees.

      Growth Marketer 1 day a month

    First result
    1 month
    Cost
    Under $100 a month for tooling
    Owner
    Founding Marketing Manager

    Smallest test

    Run one session for the list.

    Why it works

    Each month, a researcher presents 30 minutes on one failure area (long-horizon tool use, simulated users, cyber patching) with examples from Collinear benchmarks, then takes questions. Record it and cut clips.

    Researchers attend talks that teach them something. Each session builds the contact list and produces content for social and email.

    Risks and how to handle them

    • Low attendance

      Record and send to all registrants; judge on views.

    Builds on

    Evidence

    • HiringFounding Marketing Manager: positioning and messaging; "case studies and co-branded success stories"; one-pagers, decks, blog posts and competitor comparisons; paid, social, email, webinar and conference campaigns; serves "leading AI labs and Fortune 500 enterprises".Source
    • Your websiteResearch page: areas in world models, recursive self-improvement, user simulation, long-horizon agents and post-training, with sections for benchmarks, publications and field notes.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
  • 14

    Collinear is opening a San Francisco office and hiring researchers every week. A regular meetup on RL environments and evals gives researchers a reason to visit and gives recruiting and sales the same audience.

    Medium impactMedium effort First result in 2 months Growth Marketer

    First action

    Set the first date for the month after the office opens.

    How to measure it

    Target-account attendees per meetup

    Now
    –
    Target
    15
    By
    third meetup

    Steps

    1. 1

      Set a date

      Same evening each month.

      Growth Marketer 30 minutes

    2. 2

      Book speakers

      One Collinear, one guest.

      Co-founder and CEO 2 weeks

    3. 3

      Invite

      List contacts plus local communities.

      Growth Marketer 1 week

    4. 4

      Run it

      Logistics, food, swag.

      Growth Marketer 1 evening

    5. 5

      Record talks

      Post them as research pages and clips.

      Digital Marketing Manager 1 day

    6. 6

      Follow up

      Attendees into the CRM with notes.

      Growth Marketer 1 day

    First result
    2 months
    Cost
    $1,500 to $3,000 a month
    Owner
    Growth Marketer

    Smallest test

    One meetup, then decide on monthly.

    Why it works

    Host a monthly evening with two short talks (one from Collinear, one from a guest researcher), drinks and swag, capped at 60 people, invited from the account list and local AI communities.

    San Francisco has the densest concentration of the buyers and of the researchers Collinear wants to hire, and a meetup with a fixed schedule builds a community around the office.

    Risks and how to handle them

    • Draws job seekers more than buyers

      Invite buyers personally; recruiting benefits either way.

    Evidence

    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • HiringGrowth Marketer: campaigns across email, outbound, social and community; account and contact lists; landing pages, forms, sequences and CRM fields; "select events and community programs"; company gifting and swag; success is "Every campaign and event has a clear audience, operating plan, tracking setup, and follow-up process." Early-career, in office in Sunnyvale or San Francisco.Source
    • HiringCareers page: 14 open roles (Research 7, Engineering 5, GTM 1, Product 1), including new research tracks in robotics simulation data, biosafety simulations, cybersecurity and AI physics (fluid dynamics, structural mechanics, semiconductor design); the MTS Product role works "directly with researchers and engineers at leading AI labs".Source
  • 15

    Collinear has a PyPI package (last released January 2026) and Mintlify docs, but the buying path is a demo booking. Researchers prefer to try things in code before a call.

    Medium impactMedium effort First result in 6 weeks MTS Product

    First action

    Write the five-line quickstart that should work, then build to it.

    How to measure it

    Accounts running sample tasks

    Now
    –
    Target
    40
    By
    end of quarter

    Steps

    1. 1

      Write the quickstart

      Install, authenticate, pull, run.

      MTS - Product 1 day

    2. 2

      Update the package

      Sample pulls and harness adapters.

      MTS - Product 3 weeks

    3. 3

      Issue free keys

      Work email only, five tasks per area.

      MTS - Product 1 week

    4. 4

      Docs page

      The quickstart on Mintlify.

      MTS - Product 2 days

    5. 5

      Notify sales

      Runs create a CRM signal.

      Growth Marketer 1 day

    6. 6

      Announce

      Thread and email to the list.

      Founding Marketing Manager 1 day

    First result
    6 weeks
    Cost
    About a month of engineering
    Owner
    MTS Product

    Smallest test

    Send the quickstart to five friendly researchers.

    Why it works

    Update the package so a researcher with a free key can pull five sample tasks in their area and run them in their harness with a few lines of code, with the docs quickstart showing exactly that.

    It shortens the "evaluate the samples" step to minutes and shows the team which accounts actually ran tasks, a strong buying signal.

    Risks and how to handle them

    • Samples get scraped

      Keys limited by work email and rate.

    Builds on

    Evidence

    • programmable.infoThe sitemap lists 17 URLs: the homepage, research, about, careers and six Calico pages, with no individual research, field-notes or case-study pages. There is a PyPI package, collinear, last released in January 2026, and docs on Mintlify.Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
    • HiringCareers page: 14 open roles (Research 7, Engineering 5, GTM 1, Product 1), including new research tracks in robotics simulation data, biosafety simulations, cybersecurity and AI physics (fluid dynamics, structural mechanics, semiconductor design); the MTS Product role works "directly with researchers and engineers at leading AI labs".Source

Big bets 3

Larger plays that can change the trajectory.

  • 16

    Collinear runs CWE-bench, FrontierCUA, MCP-Atlas and OSWorld-Verified on separate sites. Together they are a map of where frontier agents fail, which no single benchmark shows.

    High impactHigh effort First result in 1 quarter MTS Research

    First action

    Agree a common results format across the four benchmarks.

    How to measure it

    Evaluation requests from the hub

    Now
    –
    Target
    25 a quarter
    By
    2 quarters

    Steps

    1. 1

      Common format

      Shared fields across benchmarks.

      MTS - Research 2 weeks

    2. 2

      Build the hub

      Leaderboards, methods, alerts.

      MTS - Product 4 weeks

    3. 3

      First report

      Cross-benchmark findings.

      MTS - Research 3 weeks

    4. 4

      Brief press

      AI reporters under embargo.

      Co-founder and CEO 1 week

    5. 5

      Send to the list

      Report plus sample offer.

      Growth Marketer 1 day

    6. 6

      Repeat quarterly

      Fixed release dates.

      MTS - Research quarterly

    First result
    1 quarter
    Cost
    About two months of research and product time
    Owner
    MTS Research

    Smallest test

    Publish one cross-benchmark blog post.

    Why it works

    Bring every benchmark into one hub with consistent leaderboards, methods and alerts, and publish a quarterly report across all of them, pitched to AI press and sent to the account list.

    A quarterly cross-benchmark report becomes a reference people cite, and the hub keeps every benchmark's audience in one place with one sign-up.

    Risks and how to handle them

    • Results disputed by labs

      Publish methods and let labs comment before release.

    Builds on

    Evidence

    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteResearch page: areas in world models, recursive self-improvement, user simulation, long-horizon agents and post-training, with sections for benchmarks, publications and field notes.Source
  • 17

    Calico already sells "single capabilities or pre-built packs" off the shelf, but only through a conversation. Many teams training agents (startups, open-model teams, universities) are too small for a sales process but would buy a pack.

    High impactHigh effort First result in 2 quarters MTS Product

    First action

    Set prices for three standard pack sizes in one capability.

    How to measure it

    Self-serve pack buyers

    Now
    –
    Target
    20 a month
    By
    2 quarters after launch

    Steps

    1. 1

      Set packs and prices

      Three sizes per capability.

      Co-founder and CEO 2 weeks

    2. 2

      Licence terms

      Standard terms for training use.

      Co-founder 2 weeks

    3. 3

      Checkout

      Card payment and delivery through the package.

      MTS - Product 4 weeks

    4. 4

      Store page

      Packs, samples, terms.

      Founding Marketing Manager 1 week

    5. 5

      Launch

      To the list and research communities.

      Founding Marketing Manager 1 week

    6. 6

      Upsell

      Large buyers move to custom deals.

      Co-founder and CEO ongoing

    First result
    2 quarters
    Cost
    About two months of engineering
    Owner
    MTS Product

    Smallest test

    Sell one pack by invoice at a published price.

    Why it works

    Let teams buy standard packs (for example 500 cyber or MCP tasks with environments and verifiers) by card, with fixed prices and licence terms, while large labs keep custom volume deals.

    It opens a long tail of buyers at no sales cost, and smaller buyers grow into larger ones.

    Risks and how to handle them

    • Undercuts large deals

      Standard packs only; held-out and custom data stay in negotiated deals.

    Builds on

    Evidence

    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
    • Competitor researchAn RL vendor directory ranks Collinear #12 and describes it as now positioned "mainly as a supplier of training data, environments and evals to frontier labs"; related vendors named are AfterQuery, Bespoke Labs, Huzzle Labs and Fleet AI.Source
    • programmable.infoThe sitemap lists 17 URLs: the homepage, research, about, careers and six Calico pages, with no individual research, field-notes or case-study pages. There is a PyPI package, collinear, last released in January 2026, and docs on Mintlify.Source
  • 18

    New tracks need domain experts: the biosafety role needs no AI experience and asks for a PhD, and AI physics needs simulation specialists. Hiring each expert full-time limits how fast new domains launch.

    Medium impactHigh effort First result in 1 quarter MTS Research

    First action

    Write the contributor page for one domain and share it with three expert communities.

    How to measure it

    Active expert contributors

    Now
    –
    Target
    50
    By
    2 quarters

    Steps

    1. 1

      Define the work

      Task, answer and grader standards per domain.

      MTS - Research 2 weeks

    2. 2

      Set pay

      Per task and per review.

      Co-founder 1 week

    3. 3

      Vetting

      Credential checks and a trial task.

      MTS - Research 2 weeks

    4. 4

      Public page

      For specialists, by domain.

      Founding Marketing Manager 1 week

    5. 5

      Recruit

      Professional societies and university networks.

      Growth Marketer 4 weeks

    6. 6

      Quality loop

      Peer review between experts.

      MTS - Research ongoing

    First result
    1 quarter
    Cost
    Contributor pay per task
    Owner
    MTS Research

    Smallest test

    Recruit five biosafety experts on contract.

    Why it works

    Create a contributor program where vetted experts (biosafety, cyber, physics, finance) write tasks, reference answers and graders on contract, with a public page explaining the work and pay.

    It lets Collinear launch domains faster, and the expert network becomes a supply advantage that is hard for others to copy.

    Risks and how to handle them

    • Quality varies across contributors

      Peer review and calibration against staff experts.

    Evidence

    • HiringMTS - Research (Biosafety Simulations): "We are extending our environments into biosafety"; turn the judgment of biosafety professionals into tasks, reference answers and graders, and "Red-team the boundary" between helpful biology and operational uplift. No prior AI experience is required.Source
    • HiringCareers page: 14 open roles (Research 7, Engineering 5, GTM 1, Product 1), including new research tracks in robotics simulation data, biosafety simulations, cybersecurity and AI physics (fluid dynamics, structural mechanics, semiconductor design); the MTS Product role works "directly with researchers and engineers at leading AI labs".Source

Wildcards 3

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

  • 19

    Collinear is research-led and its buyers read papers. University labs can't afford task packs but publish constantly.

    Medium impactMedium effort First result in 2 quarters Co founder and CEO

    First action

    List ten university labs working on RL environments and agents.

    How to measure it

    Papers using Collinear data

    Now
    –
    Target
    5
    By
    3 quarters

    Steps

    1. 1

      Pick labs

      Ten labs with active agent research.

      MTS - Research 1 week

    2. 2

      Set terms

      Research use, citation request.

      Co-founder 1 week

    3. 3

      Offer

      Personal note from the CEO.

      Co-founder and CEO 1 week

    4. 4

      Support

      A researcher contact for each lab.

      MTS - Research ongoing

    5. 5

      Feature papers

      Research page entries.

      Growth Marketer as published

    6. 6

      Review yearly

      Renew labs that publish.

      Co-founder and CEO yearly

    First result
    2 quarters
    Cost
    Free licences
    Owner
    Co founder and CEO

    Smallest test

    Offer licences to three labs.

    Why it works

    Offer ten university labs a free research licence for one capability, with a request to cite Collinear and share results, and feature the papers on the research pages.

    Citations build credibility with the exact researchers who buy, and university students become future lab hires who already know the tools.

    Risks and how to handle them

    • Data reaches competitors through papers

      Licence non-held-out tasks only.

    Evidence

    • Your websiteResearch page: areas in world models, recursive self-improvement, user simulation, long-horizon agents and post-training, with sections for benchmarks, publications and field notes.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
    • Your websiteCalico: 18K tasks off the shelf with runnable environments and verified rewards; MCP and computer use 5,000 tasks each across 70+ enterprise runtimes; cyber 4,000; DeepSWE 3,000; GDPval 1,000; buyers "Start with five representative samples".Source
  • 20

    Collinear calls itself "the 911 for data" and promises speed. Model launches are the moments AI watchers want numbers, and CWE-bench already covers the major labs' models.

    Medium impactMedium effort First result in Next model launch MTS Research

    First action

    Set up the harness to run a new model with one command.

    How to measure it

    Model launches scored within 48 hours

    Now
    –
    Target
    80%
    By
    2 quarters

    Steps

    1. 1

      Ready harness

      One-command runs for new models.

      MTS - Research 1 week

    2. 2

      Watch releases

      Alerts for new model launches.

      Growth Marketer ongoing

    3. 3

      Run and check

      Run, verify, write one line of insight.

      MTS - Research 1 day per model

    4. 4

      Post

      CEO thread with chart.

      Co-founder and CEO 1 hour

    5. 5

      Update the leaderboard

      Same day.

      MTS - Research 1 hour

    6. 6

      Tell the lab

      Send results to the model's team.

      Co-founder and CEO 15 minutes

    First result
    Next model launch
    Cost
    Inference costs per run
    Owner
    MTS Research

    Smallest test

    Score the next major release and measure the thread's reach.

    Why it works

    Keep a ready harness so each new frontier model is run on CWE-bench as soon as it is available, and post the result with the CEO thread format within 48 hours.

    Fast, credible numbers on launch day get picked up by people covering the release, and it shows the speed Collinear sells.

    Risks and how to handle them

    • Rushed results contain errors

      Same verifier as always; mark results provisional for a week.

    Builds on

    Evidence

    • Your websiteHomepage: "When frontier intelligence can't wait, there's Collinear"; off-the-shelf tasks for cyber, SWE, computer use, knowledge work and behavior; "The 911 for data"; 32B+ tokens of agent training data generated; benchmarks including CWE-Bench, FrontierCUA, MCP-Atlas and OSWorld-Verified; tool surfaces from GUIs and APIs to MCP, GitHub, Jira, ServiceNow and Salesforce.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".Source
  • 21

    Collinear is building biosafety simulations that red-team "the boundary" between helpful biology and operational uplift, and runs a cyber benchmark. Government AI safety bodies test frontier models for these risks.

    Medium impactMedium effort First result in 3 quarters Co founder

    First action

    Write a two-page brief on the biosafety and cyber evaluation tracks.

    How to measure it

    Government evaluation pilots

    Now
    –
    Target
    1
    By
    3 quarters

    Steps

    1. 1

      Write the brief

      Capabilities, methods, safeguards.

      MTS - Research 1 week

    2. 2

      Find contacts

      Evaluation teams at each institute.

      Co-founder 2 weeks

    3. 3

      Present

      A short technical briefing.

      Co-founder and CEO 1 meeting each

    4. 4

      Propose a pilot

      A scoped evaluation project.

      Co-founder 2 weeks

    5. 5

      Handle procurement

      Registrations and requirements.

      Co-founder as needed

    6. 6

      Publish what is allowed

      With permission, a summary.

      Founding Marketing Manager after delivery

    First result
    3 quarters
    Cost
    Time and procurement setup
    Owner
    Co founder

    Smallest test

    One briefing with one institute.

    Why it works

    Prepare a short brief on Collinear's biosafety and cyber evaluation capabilities and approach the national AI safety and security institutes in the US and UK, using the co-founder's experience advising on AI data.

    It opens a buyer with budget and credibility, and work for safety institutes strengthens trust with labs.

    Risks and how to handle them

    • Long government timelines

      Keep it a side track until a pilot is real.

    Builds on

    Evidence

    • HiringMTS - Research (Biosafety Simulations): "We are extending our environments into biosafety"; turn the judgment of biosafety professionals into tasks, reference answers and graders, and "Red-team the boundary" between helpful biology and operational uplift. No prior AI experience is required.Source
    • Search resultsCWE-bench v1 (September 25): 120 held-out tasks across 73 CWEs; Grok 4.7, Gemini 4 Argon and GPT-6 Astra tie at 68%, Claude Opus 5.5 at 67%; models are submitted by contacting Collinear; the site has no newsletter or sign-up.Source
    • Your websiteAbout: founded by Nazneen Rajani (CEO, previously led post-training research at Hugging Face and Salesforce Research) and Soumyadeep Bakshi (previously led Deloitte's AI practice); "trusted by leading frontier labs" with no labs named; opening a San Francisco office next to the Sunnyvale HQ; swag is "not set up as a shop yet".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.

Have it built

ProgrammableAgency

2 of the 21 ideas are things Programmable builds and runs for its customers: pages, outreach, prospect lists, booking flows and monitors. Pick one, or send the whole list.

Prospect lists2The list compiled from public sources, verified, deduplicated and delivered to your CRM.

A 30-minute session with the engineer who would build it, not a sales call. You leave with a plan whether or not you hire us.

Book a working session

Usually live within days.

Tools for the rest

ProgrammableTalk

Fill the events in this plan

These ideas are events. Talk gets registrants for them and tells you who's coming and what their company does before the day.

Plan an event (opens programmable.talk 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.

  • Curio Digital growth strategyA B2B web agency that designs, builds and manages conversion-focused websites, mostly on Webflow, and sells them as one system (GrowthOS)…23 growth ideas
  • Pineo growth strategyA London HubSpot consultancy, Platinum partner for both HubSpot and Zapier. It offers fixed-price HubSpot onboarding, implementation…23 growth ideas
  • Kikoff growth strategyA consumer fintech whose app helps people build credit: a Kikoff credit account that reports to all three bureaus, rent and bill…21 growth ideas
  • Brimstone growth strategyAn industrial start-up whose patented Rock Refinery co-produces portland cement, supplementary cementitious materials and smelter grade…20 growth ideas
  • Incode growth strategyAn AI identity verification platform: document verification, biometric liveness and deepfake defense, government record matching, KYC and…21 growth ideas
  • Softr growth strategyA no-code AI app builder for client portals and internal tools, with its own database, workflows, permissions, AI agents and an MCP.21 growth ideas
  • Home Landing growth strategyAn online shop for designer tea towels and table linen: printed and woven cotton and linen from ten European makers such as Le Jacquard…24 growth ideas
  • 3Cloud growth strategyA Microsoft Azure consultancy for data and AI, app innovation and cloud platform work, with its own AI-assisted delivery method (MILO) and…22 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 18, 2026, and cites them throughout. Collinear AI did not commission, review or endorse it, and Programmable is not affiliated with Collinear AI. 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 Collinear AI runs its business.

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

30/60/90 roadmap

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

  1. Days 1 to 308 ideas

    Capture and convert: benchmark sign-ups, fast samples, the account list, CEO threads, one proof point and indexed research pages; book NeurIPS and plan the first new-track launch.

    Programmable can ship 2 of these in days.

  2. Days 31 to 608 ideas

    Reach the list: gift boxes, competitor comparisons, matched paid social, the enterprise offer, webinars, the SF meetup, the self-serve package and launch-day model scores.

  3. Days 61 to 905 ideas

    University licences and government evaluators, then commit to the benchmark hub, the online store and the expert network on the first 60 days of evidence.

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 tasks where frontier models fail. The most cited problem: static evals that miss real agent failures.

  • Runnable environments with verified rewards
  • Slow, custom data projects
  • In their words: “hill climb”

Market

Competitive landscape

4 competitors compared: AfterQuery, Bespoke Labs, Fleet AI and Huzzle Labs.

  • Gap to exploit: Off-the-shelf, verified tasks delivered in hours
  • AfterQuery: Positioning overlaps with Collinear
  • Bespoke Labs: Less enterprise runtime coverage

Hiring signals

Collinear has 14 open roles on its careers page, half in research, plus marketing roles on Wellfound and Built In, and together they show a research-led company building its first go-to-market.

  • Growth Marketer: Lists, email, outbound, landing pages, events, community and swag, each with tracking and follow-up. Ideas 1, 3, 6, 7, 9, 14.
  • Digital Marketing Manager: Paid and organic X and LinkedIn measured on cost per qualified lead and pipeline. Ideas 4, 11.
  • Founding Marketing Manager: Positioning, case studies, competitor comparisons and webinars for labs and Fortune 500 enterprises. Ideas 2, 5, 10, 12…

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