Thirdlayer growth playbook

Point both buttons to a booking page with three short questions: team and size, which of the four use cases applies (customer operations, revenue operations, product and engineering, agent evaluation), and whether the deployment must stay inside their boundary. Send a confirmation page with the AutoAgent post and the matching use case to read before the call. Keep hello@ as a small text link for people who prefer email. Point the footer icons to the real LinkedIn and X profiles, or remove them. Tag every booking by source so the team can see which page, post or repository produced each conversation.

Enterprise buyers expect to pick a time. A mailto link breaks in browsers with no mail client, gives no tracking, and makes the next step a blank email to write. Three answers before the call let the founder or FDE arrive with the right example, and the booking data shows which use case pulls the most interest.

[ ] Request Demo -> booking page (30 min)
[ ] Get Started -> same page, or a short form
[ ] Q1: team and size
[ ] Q2: use case (customer ops / revenue ops / product & engineering / agent evaluation)
[ ] Q3: must data stay inside your boundary? (yes / no / not sure)
[ ] Confirmation page with the matching use case and the AutoAgent post
[ ] hello@ kept as a text link
[ ] LinkedIn and X icons go to real profiles
[ ] UTM tags on blog, GitHub and Dex links

• Fewer, less casual enquiries

  Keep the email link visible for people who don't want to book.

• Calls from teams that don't fit

  Use the size and deployment answers to send smaller teams to Dex Teams instead.

• Your websiteHomepage (checked November 10, 2026): eyebrow "Your internal agent factory", H1 "Own the intelligence behind your operations"; both "Get Started" and "Request Demo" are mailto links to [email protected]; the footer LinkedIn and X links point to "#" placeholders; no pricing, testimonial, case study or customer quote anywhere on the page.Source
• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• Dex siteDex site (footer "ThirdLayer, Inc."): "The self-driving workspace for operators"; sign-up through the Chrome extension, with onboarding and private-cloud calls booked through cal.com; testimonials from a PE partner, founders, a Head of BD, a VC and a Series B fintech COO; claims operators at large companies "and 10,000+ more".Source

Publish both posts, or an open version of each with the method, one worked example and a results chart, keeping customer-specific detail out. Offer the full version through the booking page as a reason to talk. Link each post to the matching use case and the demo booking.

Technical buyers judge an agent vendor by its method. Locked posts in the main research section look unfinished, while open ones earn links, shares and search traffic the way AutoAgent did.

• Giving away the method

  Publish the idea and results; keep the implementation and customer data private.

• Your websiteResearch page: three posts; "AutoAgent: Autonomous Harness Engineering" (April 2, 2026) is open, while "Let Agents Build the Company Ontology" and "Agents that improve with every trace" are marked "Password required", although they are the posts about the context layer and learning from traces that the homepage sells.Source
• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source

Add a short section to the README: "Running agents in production? Thirdlayer builds company-specific evals and regression suites from your own traces", with a tagged link to a page for agent-evaluation buyers. Add the same block at the end of the AutoAgent post, and link the repository from the blog and research pages. Keep the hiring line, but link it to /careers.

The repository is the company's largest audience and already self-selects for technical buyers. One tagged link turns that audience into measurable pipeline at no cost.

• Open-source users dislike sales copy

  Keep it to one factual section below the usage docs.

• #1 Replace the two mailto buttons with a booked demo, and fix the dead LinkedIn and X links

• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source
• Your websiteAutoAgent post (April 2, 2026): an open-source system in which a meta-agent improves an agent harness overnight; it reports 96.5% on SpreadsheetBench and 55.1% on TerminalBench, first on both leaderboards; lessons include that trajectories, not just scores, matter and that optimisers overfit; the post ends with an email address, not a link to Dex, the evaluation product or a demo.Source
• Your websiteBlog index: 8 posts, the newest dated August 10, 2026; five are Dex release notes or productivity posts from February to April 2026 (such as "Meet TaskOS" and "The Zeigarnik Effect: Why Unfinished Work Hijacks your Mind"); the blog does not link to the AutoAgent repository.Source

Write one sentence for the company and one for each product, plus a line that connects them (for example: Dex for individual operators; Thirdlayer builds the same system for whole teams with an FDE). Update the YC profile, Chrome listing, Product Hunt page, both site footers and the social bios to match. Add a short "Dex for teams" link from joindex.com to thirdlayer.inc and back.

Investors, candidates and buyers compare these surfaces. A consistent story makes the company look bigger and clearer, and helps search engines and AI assistants describe it correctly.

• Confusing current Dex users

  Keep Dex's own pitch and only add the link to teams.

• Y CombinatorY Combinator profile: Winter 2025 batch, San Francisco, describing the company as "the learning and alignment layer for enterprise AI".Source
• Chrome Web StoreChrome Web Store listing for Dex (checked November 10, 2026): 2,000 users, 4.2 stars from 36 ratings, version 3.81 updated October 21, 2026, developer site joindex.com, described as "The AI copilot for your browser".Source
• Dex siteDex site (footer "ThirdLayer, Inc."): "The self-driving workspace for operators"; sign-up through the Chrome extension, with onboarding and private-cloud calls booked through cal.com; testimonials from a PE partner, founders, a Head of BD, a VC and a Series B fintech COO; claims operators at large companies "and 10,000+ more".Source
• Your websiteHomepage (checked November 10, 2026): eyebrow "Your internal agent factory", H1 "Own the intelligence behind your operations"; both "Get Started" and "Request Demo" are mailto links to [email protected]; the footer LinkedIn and X links point to "#" placeholders; no pricing, testimonial, case study or customer quote anywhere on the page.Source

Generate a sitemap of the homepage, research, blog posts and careers pages, and add a robots.txt that points to it. Add a favicon and a 1200x630 share image for the homepage, plus one per research post, showing the headline result (for example AutoAgent's benchmark scores). Submit the sitemap in Search Console.

Research posts travel by sharing. A card with an image and a clear title gets more clicks than a bare link, and a sitemap helps new posts get indexed quickly.

• Locked pages appear in the sitemap

  Leave out password-protected posts until they are opened.

• Your websiteSite basics (checked November 10, 2026): the homepage has no social share image and no favicon; /sitemap.xml and /robots.txt both return 404; the domain was registered in May 2025.Source
• PageSpeedReal visitors (Chrome UX Report, 75th percentile): LCP 766 ms, INP 82 ms, CLS 0, so the homepage passes Core Web Vitals; LCP is 77% slower than in the previous period.Source

State what the 10,000+ counts (for example operators across the extension, iMessage and Slack, or sign-ups to date) and match the wording on the site. Then add an in-product review prompt for users who have completed several automations, and link to the Chrome listing from the Dex onboarding email. Refresh the store screenshots to show automations and agents.

Store ratings and user counts are the first trust check for a browser extension that reads web activity. Clear numbers prevent doubt, and a review prompt timed after a success raises both the count and the average.

• Prompting annoys users

  Ask once, only after a success, and never again after a dismissal.

• Chrome Web StoreChrome Web Store listing for Dex (checked November 10, 2026): 2,000 users, 4.2 stars from 36 ratings, version 3.81 updated October 21, 2026, developer site joindex.com, described as "The AI copilot for your browser".Source
• Dex siteDex site (footer "ThirdLayer, Inc."): "The self-driving workspace for operators"; sign-up through the Chrome extension, with onboarding and private-cloud calls booked through cal.com; testimonials from a PE partner, founders, a Head of BD, a VC and a Series B fintech COO; claims operators at large companies "and 10,000+ more".Source
• Dex siteDex pricing: Core $49/month (3 integrations), Max $99/month ("Most popular", 5 integrations), Ultra $199/month (15 integrations, dedicated account manager), Full Service (custom, "Built and managed like a team of EAs") and Teams & Enterprise ("Flat rate pricing (no seat gates)", shared context, private cloud, "Book a call"); no free tier or trial is mentioned.Source

Add a short top section: the thesis, what the team has shipped (AutoAgent and its benchmarks, Dex), where the team comes from (already on the homepage), how an FDE engagement works, and the YC backing. Link the AutoAgent README's hiring line here instead of to an email address.

Hiring FDEs and researchers is the company's capacity limit. Strong candidates choose on the problem and the team, and the evidence already exists.

• Overselling stage or size

  Stick to shipped facts and public results.

• HiringCareers page: seven openings, all in San Francisco: Platform Engineer, Forward Deployed Engineer, Product Engineer, Software Engineer (Intern), Research Scientist, Research Engineer and a General Application; the only company text is "We keep an eye out for exceptional people regardless of what is posted."Source
• HiringForward Deployed Engineer: deploys the platform "into live customer environments", builds ETL pipelines and connectors for company-specific data, owns work from scoping through production and turns customer patterns into product features; the post's thesis: "AI models are generic, but people's work and processes are specific."Source
• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source

For each use case (customer operations, revenue operations, product and engineering, agent evaluation), write one short story with an approved customer: the operation before, the system built on Dex, the eval results and the outcome (tickets resolved without escalation, hours saved, regressions caught). Where a customer can't be named, publish it anonymised by industry and size. Place each story under its use case on the homepage.

A logo shows that someone tried the product; a number shows that it worked. Buyers comparing Thirdlayer with Glean or Dust need a result they can repeat to their boss.

• Customers decline to be named

  Offer anonymised stories with industry and size.

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• Competitor researchRivals' own homepages: Glean leads with named outcomes (Zillow "1.5+ hours saved weekly per user" and "3.4k+ agents"; TIME "3 weeks to go-live") and a "Get a demo" page; Dust shows "3,000+ global organizations", "300,000+ agents deployed", a free trial and case studies with numbers; Distyl calls itself "an AI-native operations company" with "50+" enterprise deployments.See the research Source
• Your websiteHomepage (checked November 10, 2026): eyebrow "Your internal agent factory", H1 "Own the intelligence behind your operations"; both "Get Started" and "Request Demo" are mailto links to [email protected]; the footer LinkedIn and X links point to "#" placeholders; no pricing, testimonial, case study or customer quote anywhere on the page.Source

Package a four-week pilot with a fixed price: Thirdlayer replays the customer's traces, builds a company-specific eval suite, runs a harness optimisation pass and delivers a report on pass rates, failure clusters and regressions. The pilot ends with a proposal to keep the suite running on Dex. Market it to AutoAgent's GitHub audience and on the agent-evaluation use-case page.

A small, measurable first deal is easier to sign than an open-ended transformation. It also creates the eval data and trust that lead to the larger FDE engagement.

• Pilots stay pilots

  Agree the conversion criteria and price before the pilot starts.

• #3 Point AutoAgent's 4.6k-star README at the evaluation product, not only at a hiring email

• Your websiteAutoAgent post (April 2, 2026): an open-source system in which a meta-agent improves an agent harness overnight; it reports 96.5% on SpreadsheetBench and 55.1% on TerminalBench, first on both leaderboards; lessons include that trajectories, not just scores, matter and that optimisers overfit; the post ends with an email address, not a link to Dex, the evaluation product or a demo.Source
• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• HiringPlatform Engineer: owns the context capture and learning pipeline with customers, and builds "evaluation and reliability tooling" that decides when agents act on their own and when they check in.Source

Group Dex accounts by work email domain each week. When a domain reaches three paying users, or one Ultra user, reach out with a team proposal: shared context across their tools, one flat price that replaces the individual subscriptions, and an FDE-built first system. Add a "Using Dex with your team?" prompt inside the product that leads to the same offer.

Bottom-up usage is the cheapest enterprise pipeline: the buyer already trusts the product, and the team plan solves a problem individual plans can't (shared context).

• Users feel watched

  Use only paying-account domains and reach out as an offer, not a notice.

• #4 Say one thing everywhere: four surfaces describe Thirdlayer four different ways

• Dex siteDex pricing: Core $49/month (3 integrations), Max $99/month ("Most popular", 5 integrations), Ultra $199/month (15 integrations, dedicated account manager), Full Service (custom, "Built and managed like a team of EAs") and Teams & Enterprise ("Flat rate pricing (no seat gates)", shared context, private cloud, "Book a call"); no free tier or trial is mentioned.Source
• Dex siteDex site (footer "ThirdLayer, Inc."): "The self-driving workspace for operators"; sign-up through the Chrome extension, with onboarding and private-cloud calls booked through cal.com; testimonials from a PE partner, founders, a Head of BD, a VC and a Series B fintech COO; claims operators at large companies "and 10,000+ more".Source
• HiringProduct Engineer: ships features across "the browser extension, web surfaces, and platform", defines how users interact with agents, and builds integrations and data pipelines.Source

Add an "Engagement" section or page: week 1 map the operation, weeks 2 to 4 build and evaluate, week 5 deploy with approvals, then ongoing improvement. List what the customer receives (agents, workflows, eval suite, dashboard), what the customer provides (data access, a process owner), and a "from" price or price range.

A visible process and a price anchor qualify buyers before the call and shorten the sales cycle. Buyers who book have already accepted the shape of the deal.

• Price scares off some buyers

  Show a range and what changes the price.

• #1 Replace the two mailto buttons with a booked demo, and fix the dead LinkedIn and X links

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• Dex siteDex pricing: Core $49/month (3 integrations), Max $99/month ("Most popular", 5 integrations), Ultra $199/month (15 integrations, dedicated account manager), Full Service (custom, "Built and managed like a team of EAs") and Teams & Enterprise ("Flat rate pricing (no seat gates)", shared context, private cloud, "Book a call"); no free tier or trial is mentioned.Source
• HiringForward Deployed Engineer: deploys the platform "into live customer environments", builds ETL pipelines and connectors for company-specific data, owns work from scoping through production and turns customer patterns into product features; the post's thesis: "AI models are generic, but people's work and processes are specific."Source

Publish a security page with a deployment diagram for each mode (cloud, private cloud, on-prem), the data flow from each connector, how approvals and role-based access work, retention and deletion, sub-processors, and the status of any audit or certification. Offer a pre-filled security questionnaire on request through the booking page.

Security review is often the longest step for an agent that touches company data. Answering the common questions up front shortens deals and lets the buyer's champion argue internally.

• Claims that outrun the product

  Publish only what is in place today, with dates.

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• HiringPlatform Engineer: owns the context capture and learning pipeline with customers, and builds "evaluation and reliability tooling" that decides when agents act on their own and when they check in.Source
• Your websiteHomepage (checked November 10, 2026): eyebrow "Your internal agent factory", H1 "Own the intelligence behind your operations"; both "Get Started" and "Request Demo" are mailto links to [email protected]; the footer LinkedIn and X links point to "#" placeholders; no pricing, testimonial, case study or customer quote anywhere on the page.Source

After every deployment, the FDE saves a starter system: the connector set, the workflow map, a starter eval rubric and a first-week checklist, with customer data removed. Group starters by the four homepage use cases. New customers in the same use case begin from the starter, and the best ones appear (described generically) on the use-case pages.

Reuse shortens time to first production system, which is the number buyers care about and the one that limits how many customers a small team can serve.

• Starters leak customer detail

  Review by a second engineer before saving.

• HiringForward Deployed Engineer: deploys the platform "into live customer environments", builds ETL pipelines and connectors for company-specific data, owns work from scoping through production and turns customer patterns into product features; the post's thesis: "AI models are generic, but people's work and processes are specific."Source
• Dex siteDex use case library: 32 use cases in eight categories (CRM, Operations, Inbox, Recruiting, GTM & Sales, Support, Research, Knowledge), each tagged with a role such as Account Executive, Support Lead or Chief of Staff; no customer stories or numbers.Source

Build four pages (customer operations, revenue operations, product and engineering, agent evaluation). Each covers the problem, the system Thirdlayer builds, the connectors involved, the eval approach, a sample approval flow like the homepage CRM example, the outcome story when available, and the booking link pre-set to that use case.

Function-specific pages match how buyers search and how they pass links internally, and they give each outcome story a home.

• Thin pages without proof

  Launch with the sample flows and add stories as they are approved.

• #5 Add a sitemap, robots.txt, favicon and social share image to thirdlayer.inc

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• Dex siteDex use case library: 32 use cases in eight categories (CRM, Operations, Inbox, Recruiting, GTM & Sales, Support, Research, Knowledge), each tagged with a role such as Account Executive, Support Lead or Chief of Staff; no customer stories or numbers.Source
• Your websiteSite basics (checked November 10, 2026): the homepage has no social share image and no favicon; /sitemap.xml and /robots.txt both return 404; the domain was registered in May 2025.Source

Run a monthly 45-minute live session: a short talk on one lesson (for example, catching overfitting in harness optimisation), then live review of two attendee harnesses or eval suites, volunteered in advance. Record it, post clips, and follow up with attendees from companies that run production agents.

Practical help builds trust with the exact engineers who later champion an eval or agent platform, at almost no cost.

• Draws hobbyists, not buyers

  Prioritise live reviews for production teams.

• #3 Point AutoAgent's 4.6k-star README at the evaluation product, not only at a hiring email

• Your websiteAutoAgent post (April 2, 2026): an open-source system in which a meta-agent improves an agent harness overnight; it reports 96.5% on SpreadsheetBench and 55.1% on TerminalBench, first on both leaderboards; lessons include that trajectories, not just scores, matter and that optimisers overfit; the post ends with an email address, not a link to Dex, the evaluation product or a demo.Source
• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source
• HiringResearch Scientist: post-trains models on tasks and environments built from production data, treats "evals as a research problem in their own right" and builds pipelines that "turn messy production traces into training signal".Source

Write a guide to building in-house versus with Thirdlayer, then three fair comparison pages: who each option suits, how agents get built, how they are evaluated and improved, deployment options and the kind of help provided. Date each page and cite rivals' own pages.

Comparison searches are late-stage and high-intent, and the pages help the internal champion explain the choice.

• Outdated claims about rivals

  Date each page and cite sources.

• #14 Give each of the four use cases its own page on thirdlayer.inc

• Competitor researchRivals' own homepages: Glean leads with named outcomes (Zillow "1.5+ hours saved weekly per user" and "3.4k+ agents"; TIME "3 weeks to go-live") and a "Get a demo" page; Dust shows "3,000+ global organizations", "300,000+ agents deployed", a free trial and case studies with numbers; Distyl calls itself "an AI-native operations company" with "50+" enterprise deployments.See the research Source
• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source

Offer Dex users who invest in or advise companies a simple introduction programme: a one-paragraph blurb and a booking link they can forward, with a reward such as free months of Dex or a fee per signed team. Start with the users quoted in testimonials.

Investors and operators make warm introductions all the time. A ready-made blurb makes it easy, and a warm intro converts far better than outbound.

• Referrers feel used

  Ask only users who already praised Dex, and keep it optional.

• #1 Replace the two mailto buttons with a booked demo, and fix the dead LinkedIn and X links

• Dex siteDex site (footer "ThirdLayer, Inc."): "The self-driving workspace for operators"; sign-up through the Chrome extension, with onboarding and private-cloud calls booked through cal.com; testimonials from a PE partner, founders, a Head of BD, a VC and a Series B fintech COO; claims operators at large companies "and 10,000+ more".Source
• Product HuntDex launched on Product Hunt on December 11, 2025 as "Turn tabs into automated workflows and tasks": 234 upvotes, fifth product of the day.Source

Send an automatic weekly email to each customer's sponsor: pass rate and change, new regression tests, failures fixed and where they came from, approvals handled, and estimated time saved. Add one line from the FDE on what changes next week. Use the same numbers in quarterly reviews and renewals.

Sponsors renew what they can see improving. A weekly number keeps the value visible between meetings and sets up expansion to the next operation.

• Bad weeks look bad

  Show trend lines and what was learned from new failures.

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• HiringPlatform Engineer: owns the context capture and learning pipeline with customers, and builds "evaluation and reliability tooling" that decides when agents act on their own and when they check in.Source

Open-source a starter kit for building a company eval: a schema for turning traces and human decisions into test cases, rubric templates for support and revenue operations, a regression-suite runner, and a guide. Thirdlayer's paid product becomes the managed, continuously updated version, with FDEs and private deployment.

An open standard makes Thirdlayer the reference point in the category it sells, and every team that adopts the kit is a qualified lead for the managed version.

• Competitors adopt it

  Keep the managed service, private deployment and trace learning in the product.

• #9 Sell a fixed-scope eval pilot to teams already running agents in production

• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source
• HiringResearch Scientist: post-trains models on tasks and environments built from production data, treats "evals as a research problem in their own right" and builds pipelines that "turn messy production traces into training signal".Source
• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source

Certify two or three small AI implementation firms on Dex using the starter systems. Thirdlayer keeps the platform, evals and the first architecture review; partners handle connectors and rollout under Thirdlayer's checklist. Start with one partner on one use case and compare outcomes with in-house deployments.

Partners add delivery capacity without headcount and bring their own client relationships, while the platform and eval data stay with Thirdlayer.

• Quality drops

  Thirdlayer signs off architecture and evals before go-live.

• #13 Turn each deployment into a reusable starter system, so the second customer in a use case starts at week two

• HiringForward Deployed Engineer: deploys the platform "into live customer environments", builds ETL pipelines and connectors for company-specific data, owns work from scoping through production and turns customer patterns into product features; the post's thesis: "AI models are generic, but people's work and processes are specific."Source
• Competitor researchRivals' own homepages: Glean leads with named outcomes (Zillow "1.5+ hours saved weekly per user" and "3.4k+ agents"; TIME "3 weeks to go-live") and a "Get a demo" page; Dust shows "3,000+ global organizations", "300,000+ agents deployed", a free trial and case studies with numbers; Distyl calls itself "an AI-native operations company" with "50+" enterprise deployments.See the research Source
• HiringCareers page: seven openings, all in San Francisco: Platform Engineer, Forward Deployed Engineer, Product Engineer, Software Engineer (Intern), Research Scientist, Research Engineer and a General Application; the only company text is "We keep an eye out for exceptional people regardless of what is posted."Source

Offer the first system with part of the fee tied to an agreed target, for example a share of L1-L2 tickets resolved without escalation at a set pass rate, or CRM fields kept current, measured by the eval suite both sides agree in week one. A base fee covers delivery; a success fee is paid when the target is met for a month.

It lowers the buyer's risk at the hardest step, the first system, and turns Thirdlayer's eval strength into a commercial edge rivals can't easily copy.

• Disputes about measurement

  Agree the eval suite and baseline in writing in week one.

• #18 Send each customer's sponsor a weekly "what your agents learned" report

• Your websiteHomepage: a logo strip headed "Deploying with teams at the frontier" with 15 logos (among them Vercel, Cursor, Stripe, Ramp, Figma, Notion, Shopify, GitHub and Databricks); four use cases (customer operations, revenue operations, product and engineering, and agent evaluation with "replayed environments, rubrics, and regression suites"); "A dedicated AI FDE rethinks the operation with your team and launches the first production system on Dex"; security claims: deploy inside your boundary, deterministic controls, trace every participant.Source
• Dex siteDex pricing: Core $49/month (3 integrations), Max $99/month ("Most popular", 5 integrations), Ultra $199/month (15 integrations, dedicated account manager), Full Service (custom, "Built and managed like a team of EAs") and Teams & Enterprise ("Flat rate pricing (no seat gates)", shared context, private cloud, "Book a call"); no free tier or trial is mentioned.Source
• Competitor researchRivals' own homepages: Glean leads with named outcomes (Zillow "1.5+ hours saved weekly per user" and "3.4k+ agents"; TIME "3 weeks to go-live") and a "Get a demo" page; Dust shows "3,000+ global organizations", "300,000+ agents deployed", a free trial and case studies with numbers; Distyl calls itself "an AI-native operations company" with "50+" enterprise deployments.See the research Source

Host a simple leaderboard page on thirdlayer.inc where AutoAgent users submit optimised harnesses and scores on a few open tasks. Feature the best submission each month in a short write-up and invite its author to office hours. Use submissions to spot strong engineering candidates and companies building agents seriously.

A leaderboard gives an open-source project a reason to come back, earns links and posts, and surfaces both hires and buyers.

• Gamed scores

  Re-run every featured submission and use held-out tasks.

• #15 Hold monthly harness-engineering office hours for teams shipping agents

• Your websiteAutoAgent post (April 2, 2026): an open-source system in which a meta-agent improves an agent harness overnight; it reports 96.5% on SpreadsheetBench and 55.1% on TerminalBench, first on both leaderboards; lessons include that trajectories, not just scores, matter and that optimisers overfit; the post ends with an email address, not a link to Dex, the evaluation product or a demo.Source
• GitHubAutoAgent repository: 4.6k stars, 493 forks, MIT licence, on a founder's personal account; the README logo links to thirdlayer.inc and says "We're hiring engineers. If this work interests you, reach out to [email protected]."; it does not mention Dex or Thirdlayer's evaluation product.Source