Aiolum Browse stories
All stories
Multiplayer workspace for AI coding agents; started as an approval API for AI agents · United States

HumanLayer: Revenue, sale and how it was built

Small teamFundedExitB2BDeveloper tools

HumanLayer (Multiplayer workspace for AI coding agents; started as an approval API for AI agents, United States): Acquired by Factory, November 2026. Evidence grade E: no revenue figure was checked. Team: Started by Dex Horthy (founder and CEO on the Y Combinator page); Dex Horthy and Kyle Mistele were the co-founders at the sale; Y Combinator lists a team size of 2 in November 2026; headcount at the sale not published. Launched 29 October 2024; sold 24 months after launch.Added 7 November 2026

Outcome
Acquired by Factory, November 2026
Evidence
E no revenue figure was checked
Launch
29 October 2024
Status
Sold, 24 months after launch
Team
Started by Dex Horthy (founder and CEO on the Y Combinator page); Dex Horthy and Kyle Mistele were the co-founders at the sale; Y Combinator lists a team size of 2 in November 2026; headcount at the sale not published
Outside funding
Y Combinator, Fall 2024 batch; the founders thank investors; amounts not published
Main channel
Founder's posts and audience
No revenue figure

What is not verified

No revenue figure and no deal price were found. We looked for a payment-processor profile, a founder interview with a dated figure, a company filing and the buyer's announcement; Factory's post and Dealroom's news item give no price, and Dealroom calls it "an undisclosed sum". The number of paying customers is not published: the refund statement shows that some existed, and Factory's "hundreds of daily active users" is not an exact count. The Y Combinator page lists a team size of 2, while Factory's post speaks of "Dex, Kyle and team", so the headcount at the sale is not certain. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.

HumanLayer made a shared workspace for coding agents, AI programs such as Claude Code or Codex that read a codebase, plan a change and edit the files themselves. A team groups agent sessions, plan documents and code changes into tasks, and comments on each plan before the agent writes code. Customers use their own model accounts (BYOK, bring your own key, so HumanLayer pays no model bill), and the home page lists a free Starter plan for up to 3 team members and a Pro plan at US$100 per user a month.

Dex Horthy started the company in 2023, as the Y Combinator page shows, and launched it on 29 October 2024 in Y Combinator's Fall 2024 batch as an API for human-in-the-loop approvals, where software stops and asks a person before it runs a risky action. His essays and talks in 2025 on context engineering, the careful choice of what text a model sees, moved the company to coding agents. On 4 November 2026 HumanLayer and Factory, which builds coding agents for enterprises, announced that HumanLayer is joining Factory, with Dex Horthy and Kyle Mistele moving there.

No price for the deal and no revenue figure appear in any source we read, so this is a grade E story. It sold paid plans before the sale: the founders wrote that they will refund paying customers for the last several months, and Factory wrote that the product had hundreds of daily active users.

HumanLayer: Acquired by Factory, November 2026. Multiplayer workspace for AI coding agents; started as an approval API for AI agents, United States, team of 2 to 10, evidence grade E.
The story in one card. Share it with the address of this page.
01

How much did HumanLayer make, and what was it sold for?

Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.

A two-person Y Combinator company that sold per-seat plans for its coding-agent workspace and was acquired by Factory; no revenue figure or deal price is public
FactValueGradeHow we knowSources
Status, 4 November 2026acquired by Factory; price not disclosedBFactory's post: "As part of this acquisition, we’re excited to bring some of the most beloved features of HumanLayer to Factory." The founders' post: "Today, HumanLayer is joining Factory". Dealroom's news item says Factory acquired HumanLayer "for an undisclosed sum".[2, 1, 3]
Paying customers before the saleyes; recent payments to be refundedCFounders' post of 4 November 2026: "We’ll also be refunding paying customers for the last several months." The open-source post of 23 October 2026: "If you're a paid or free user of cloud-hosted HumanLayer, nothing changes for now".[1, 7]
Prices on 7 November 2026Starter free for up to 3 team members; Pro US$100 per user a month; Enterprise customCHome page pricing section. The FAQ: "Yes - the Starter plan is free for teams of three or less."[4]
Usershundreds of daily active usersBFactory's post: "Dex and Kyle have created an impressive product and a devoted developer community with hundreds of daily active users". No exact count is given.[2]
Company start2023CThe Y Combinator company page gives 2023 as the year founded, San Francisco as the location and Fall 2024 as the batch.[5]
Public launch29 October 2024CLaunch YC post by Dexter Horthy, created 29 October 2024: "I’m Dex, and I’m building HumanLayer." The Internet Archive's first capture of www.humanlayer.dev is from 13 August 2024.[5, 6]
Founders and teamDex Horthy and Kyle Mistele; team size 2 on the Y Combinator pageCY Combinator lists Dexter Horthy as Founder/CEO and a team size of 2. The sale posts are signed "Dex & Kyle", and Dealroom writes: "Co-founders Dex Horthy and Kyle Mistele are joining Factory". Factory's post speaks of "Dex, Kyle and team", so others may have worked there; the headcount at the sale is not published.[5, 1, 3, 2]
FundingY Combinator, Fall 2024; other investors not named with amountsCThe founders thank "investors" and "the whole team at YC". No round or amount appears in the sources we read.[5, 1]
First product, November 2024API and SDK for human approval of AI agent actionsCInternet Archive copy of 8 November 2024: "HumanLayer is an API and SDK that enables AI Agents to contact humans for feedback, input, and approvals." Approvals went through Slack, email and Discord.[6]
Open-source releasecore code under the Apache 2.0 license, 23 October 2026C"So today we're open sourcing HumanLayer core". The post says it is the same code HumanLayer Cloud runs.[7]
Supported agents and modelsClaude Code, Codex, Copilot, FireworksCHome page FAQ. The daemon in the open-source code runs Claude Code and Codex sessions.[4, 7]
Time from the public launch to the sale24 monthscalcCalculated: 29 October 2024 to 4 November 2026 is 24.2 months.[5, 1]

How to read this table: a grade B value is a statement by the buyer, Factory, on its own site. A grade C value is a statement by HumanLayer on its own pages or in an Internet Archive copy of them, by the Y Combinator directory, or by a news summary. A "calc" value was computed on this page. No revenue amount exists in the sources.

HumanLayer prices from its home page, read 7 November 2026, three days after the sale was announced (grade C)
PlanPer month, billed yearly (US$)Per month, billed monthly (US$)Limit
Starter0up to 3 team members; up to 200 sessions a month
Pro100 per userunlimited tasks and sessions; bring your own model keys; remote daemons
EnterprisecustomSSO/SAML, audit logs, volume pricing, on-premises or private cloud

The page does not say whether Pro is billed monthly or yearly, so the yearly column is empty. The page says customers plug in their own Claude, Codex and other subscriptions or API keys and pay the model provider separately. An Internet Archive copy of 8 November 2024 shows the earlier approval product with a free Starter plan of 100 operations a month, a Premium plan listed at US$500 a month with 2,000 operations, and a custom Enterprise plan.

02

When did HumanLayer launch, and how did it grow?

From an approval API for AI agents to a coding-agent workspace bought by Factory, in 24 months after the public launch. Each entry carries its grade.

  1. 2023CDex Horthy starts the company. Before HumanLayer he was building an AI agent that would clean up unused database tables, and he added human approval steps to it. [5]
  2. 29 October 2024CLaunch YC post in the Fall 2024 batch: an API with Python and TypeScript clients that lets software ask humans for approval in Slack, SMS, email and other channels. [5, 6]
  3. April to August 2025CDex Horthy publishes "12-Factor Agents" in April 2025, and a Y Combinator talk on context engineering for coding agents follows on 25 August 2025; the company's resource page says HumanLayer then reached #1 on GitHub Trending. [8, 9]
  4. 17 June 2026CThe HumanLayer launch video presents the coding environment built around human review of agent work. [8]
  5. 23 October 2026CHumanLayer releases the core of its product as open source under Apache 2.0; the hosted service stays unchanged. [7]
  6. 4 November 2026BHumanLayer joins Factory. The platform stays up for at least 30 days so users can export data, move to Factory or run the open-source version, and paying customers get refunds. [2, 1, 3]
03

Why did HumanLayer work?

Four causes, in the order they mattered. Each one is a reusable lesson.

01

Writing and talks as the sales channel

The founders wrote that their writing and speaking "has earned millions of views and helped us to land HumanLayer in the hands of engineers". The resource page lists essays, talks and videos with hundreds of thousands of views each, and every page of the site points readers to the product.

02

Free for small teams, paid per seat

The home page lists a free Starter plan for teams of up to three and a Pro plan at US$100 per user a month. Customers bring their own model subscriptions, so the company does not pay for the model calls of its users.

03

Following the users to coding agents

The company started with approvals for any AI agent in 2024, then moved to workflows and tools for coding agents in 2025, when its writing on that subject drew the most attention. The 2026 product added team features: shared tasks, comments on plans and remote agents.

How it was built

These details come from the open-source announcement of 23 October 2026, the home page and the 2024 archive copy. They are grade C. [7, 4, 6, 5]

  • A desktop app in Electron, with a Tauri build, on top of a web app that draws the user interface.
  • A daemon in the humanlayer command-line tool that runs Claude Code and Codex sessions on the customer's own machines.
  • A cloud API and an Electric proxy that syncs the Postgres database out to every client, so the same tasks show on desktop, web and phone.
  • Integrations with GitHub, Linear, Jira and Slack, all released under the Apache 2.0 license.
  • SOC 2 Type II compliance, monitored with Vanta, as the home page states.
  • The 2024 product: Python and TypeScript clients with a decorator that blocks a function call until a human approves it, and replies through Slack, email and Discord.

Tools named in the sources

Each one opens its page in the Tools tab, with every story that uses it.

04

Who tried this before HumanLayer, and why did they fail?

Three companies in our database of failed startups sold tools that would make programmers faster and failed. The values below are from that database and are not independently verified.

Died 2014 to 2024Burned US$21M

Kite DevTool

The promise

An AI code completion engine that suggested code inside the programmer's editor, years before GitHub Copilot.

Why it died

It could not turn free users into paying ones, and GitHub Copilot had wider distribution and stronger models.

Competition
Died 2012 to 2016Burned US$8M

Nitrous.io

The promise

A development environment in the cloud, so programmers could code from anywhere without setting up their own machine.

Why it died

Amazon bought its competitor Cloud9 and gave it away for free.

Competition
Died 2020 to 2024Burned US$10M

CodeSee

The promise

Interactive maps of a codebase that help programmers understand large, complex code faster.

Why it died

It could not show engineering leaders a clear return, and free alternatives covered the need.

No Market Need

The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.

What changed

Kite, Nitrous.io and CodeSee each sold programmers a better tool, and each lost to free alternatives or to a large platform that bundled the same feature. HumanLayer faced the same risk: the companies that make coding agents can add team features to their own products.

Two things differed. HumanLayer did not build its own model; it ran on top of Claude Code and Codex with the customer's own keys, so its costs stayed small while the agents improved. And its audience came first, from free essays and talks that engineers read and watched. It also ended inside a larger company instead of closing, although its hosted service is being folded into Factory.

05

Can one person copy HumanLayer now?

The audience-first method, in part. The venture-backed sale, no.

One person can copy the method: write useful, specific guides for a new kind of tool, publish free workflows, and sell a product to the people who read them. HumanLayer also had Y Combinator, a second founder and a buyer in the same market, which a solo founder cannot plan for.

Where it is still open: Team features around coding agents that the agent makers do not offer yet: review of agent plans, shared agent history, and rules for regulated companies.

  1. 1Pick a new tool that many engineers are learning at once, such as coding agents.
  2. 2Publish free, specific guides and workflows for it, and keep them updated as the tools change.
  3. 3Build a product that packages the workflow you teach, on top of the tools your readers already pay for.
  4. 4Offer a free plan for small teams and charge per seat for larger ones.

What could go wrong for HumanLayer?

The risks are the mirror image of the causes above.

What is not verified

No revenue figure and no deal price were found. We looked for a payment-processor profile, a founder interview with a dated figure, a company filing and the buyer's announcement; Factory's post and Dealroom's news item give no price, and Dealroom calls it "an undisclosed sum". The number of paying customers is not published: the refund statement shows that some existed, and Factory's "hundreds of daily active users" is not an exact count. The Y Combinator page lists a team size of 2, while Factory's post speaks of "Dex, Kyle and team", so the headcount at the sale is not certain.

Trust signals on the home page

The home page shows "Trusted by engineers at" above nine company logos: Upstart, Casco, Osmosis, Ambral, Nautilus, SQDS, Weave, Ristto and Roadrunner. No source we read names these companies as customers or links a case study for them. The founders' post says the product reached engineers "from small startups to publicly traded enterprises" without naming them.

The hosted service is in transition

The founders wrote that the current platform "will remain up and running during a transition period of at least 30 days" and that its features are moving into Factory. On 7 November 2026 the home page still offers the Pro plan. Users can export their data, move to Factory or run the Apache 2.0 open-source version themselves.

Competition and model providers

On this site, BridgeMind sells a desktop workspace for coding agents and Omnara and Pushary sell control and approval tools for agents; Dealroom reports that on the same day Meta acquired Conductor, which coordinates several coding agents. HumanLayer depends on other companies' agents: its home page lists Claude Code, Codex, Copilot and Fireworks, and its daemon runs Claude Code and Codex sessions, so changes by Anthropic or OpenAI to those tools or their terms affect the product.

Sources

The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.

  1. C HumanLayer blog, 4 November 2026: HumanLayer is joining Factory (Dex & Kyle)
    Read in full (downloaded 7 November 2026). Source for the sale, the refunds, the transition period, the audience and the thanks to investors and YC.
  2. B Factory news, 4 November 2026: HumanLayer is joining Factory (Factory Team)
    Read in full on 7 November 2026; factory.ai redirects to factory.com. The buyer's own statement of the acquisition, the founders' names and the daily active users.
  3. C Dealroom News, 4 November 2026: Factory buys HumanLayer to add team coordination for coding agents
    Read in full on 7 November 2026 (robots.txt allows ClaudeBot). A news summary. Source for "undisclosed sum", the co-founder names and the Conductor deal on the same day.
  4. C HumanLayer home page with pricing, read 7 November 2026
    Read in full; /pricing redirects to the home page. Source for the product, the plans and prices, the supported agents, SOC 2 and the logo row.
  5. C Y Combinator company page: HumanLayer, with the Launch YC post of 29 October 2024
    Read the company data embedded in the page on 7 November 2026: year founded, team size, batch, location, founder and the launch post.
  6. C Internet Archive copy of the HumanLayer home page, 8 November 2024
    Read in full. Source for the first product and its 2024 prices. The archive's capture index lists the first capture on 13 August 2024.
  7. C HumanLayer blog, 23 October 2026: Open sourcing HumanLayer (Dex)
    Read in full. Source for the open-source release, the license, the parts of the system and the paid users.
  8. C HumanLayer blog, 4 September 2026: HumanLayer - All Resources (Dex)
    Read the dated list of essays, talks and videos. Source for the 2025 essays and talks, the GitHub Trending note and the June 2026 launch video.
  9. C HumanLayer blog index, read 7 November 2026
    Read in full. Dates and authors of the posts by Dex Horthy and Kyle Mistele.
  10. C Database of failed startups, local export
    Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Run this company? Send corrections or a verified figure to aiolum@livyden.com

Back to all stories