Factory announced on November 4, 2026 that HumanLayer is joining it. HumanLayer is the multiplayer coding-agent IDE and cloud from Dex Horthy and Kyle Mistele. Factory says it will open source HumanLayer and invite all HumanLayer users over. HumanLayer's core code was already open sourced on October 23 under the Apache 2.0 license, which matters for anyone worried about a product disappearing after an acquisition. Deal terms were not disclosed in either company's post. This post covers what each side says, what changes for users, and what to check.
TL;DR
| Question | Answer, per the companies |
|---|---|
| What happened? | HumanLayer joins Factory; founders Dex Horthy and Kyle Mistele join the team |
| Price and terms? | Not disclosed in the posts we read |
| Is the code open? | HumanLayer core is open source under Apache 2.0 since October 23, 2026 |
| Do hosted users lose access? | Factory says the platform stays up during a transition with export, migration or self-hosting options |
| What does Factory get? | Multiplayer tasks, workstreams and collaboration on agent-written plans, specs and design docs, promised in the next few weeks |
| What is unverified? | The transition length, deal value and any feature dates |
What Factory says
In its announcement, Factory says it admired Horthy and Mistele and describes them as power users of coding agents since early 2025. It credits them with early ideas such as agent swarms and with publishing the Research/Plan/Implement framework in August 2025, which it calls one of the first viral "skill packs." Factory says HumanLayer has often been "6-9 months ahead of the zeitgeist" on best practices for working with coding agents, especially for larger teams with hundreds or thousands of engineers. It also says HumanLayer has "hundreds of daily active users."
On what happens next, Factory says it will open source HumanLayer and invite all of its users to Factory. The current HumanLayer platform stays up through a transition period, during which Factory will help users do any or all of three things: export their data, migrate to Factory, or self-host the open-source code. In the next few weeks, the HumanLayer team is to bring features such as multiplayer tasks and workstreams, and in-app collaboration on agent-written plans, product specs and design docs, into Factory.
Factory describes itself as bringing autonomous software engineering to the enterprise, with Droids that plan, build, validate and improve work inside a company's infrastructure and approved models. It was founded in 2023 by Matan Grinberg and Eno Reyes.
What HumanLayer says
HumanLayer's own blog lists the joining post, dated November 4, 2026, under the tags software factories and announcements. Its earlier post, Open sourcing HumanLayer from October 23, argues that software factories have to be open and composable, because teams build their own from compute, harnesses and review flows they already have, and "the last thing great builders want is a black box sitting in the middle of their SDLC."
The open-source release is large. By the post's list, it contains the desktop app (Electron, with a Tauri build) and the web app behind it, the daemon that runs Claude Code and Codex sessions on your machines and ships in the humanlayer CLI, the cloud API and an Electric proxy that syncs Postgres to clients, the sources for the humanlayer-pi and humanlayer-opencode plugins, the RPI skills and a Linear CLI, and the data model for orgs, teams, projects, tasks, sessions, plans, comments, diffs, pull requests and GitHub, Linear, Jira and Slack integrations. Everything is under Apache 2.0, and the post says a self-built stack sends no telemetry unless you configure keys. A docker compose file is provided for running the whole stack.
That timeline is worth noticing. HumanLayer open sourced its core 12 days before announcing the move to Factory. For users, the order is reassuring: the code you depend on is already outside a single company's control, whatever happens to the hosted product.
What HumanLayer's writing covered
Factory leans on HumanLayer's reputation for practical advice, so it is worth seeing what the team actually published. The HumanLayer blog index lists several posts from 2026 that show the themes:
- Skill Issue: Harness Engineering for Coding Agents (Kyle, March 12, 2026), a 23-minute read on the setup around an agent, not the model itself.
- Getting Claude to Actually Read Your CLAUDE.md (Dex, March 17), which describes using conditional XML blocks so the model pays attention to the right instructions at the right time.
- Long-Context Isn't the Answer (Kyle, March 23), whose summary says the team tried Opus with a 1M-token context for coding agents and switched back, because quality of context mattered more than window size.
- Context Forking to Save Time, Tokens and Trouble (Kyle, May 15), about building up good context once and reusing it several times.
- show-me (Dex, August 12), a small skill that makes agents explain software with diagrams, trees and diffs.
- A chronology of HumanLayer's writing and speaking (Dex, September 4), a 15-minute resource page.
The common thread is context engineering: deciding what an agent sees, when, and in what form. That is the practical skill behind a multiplayer tool. When several engineers review an agent's plan, the plan artifact and the diffs are the shared context, and the tool's job is to keep them in one place. HumanLayer's site describes tasks that group agent sessions, artifacts and worktrees in one workspace the whole team can edit.
These are the team's own claims about its work, and the blog index is a summary, not a review. We have not tested HumanLayer's product against Factory's, and nobody outside the two companies has published a comparison yet. The value for readers is the ideas, which stay public in the posts and in the open-source repository regardless of what happens to the company.
Why this matters
The "software factory" idea is a pitch that teams of coding agents, managed through shared tooling, can produce software at the scale of a whole engineering organization. We explained the term in what a software factory is and the SMB promise. Factory is one of the companies pushing it for enterprises, and HumanLayer was one of the tools that focused on the human side: multiple people reviewing agent plans and diffs together.
Two trends meet in this deal. The first is consolidation among coding-agent tools. Cognition, for example, just released a cheaper model built on an open base, as we covered in SWE-2 Flash on Inkling-Small, and bigger vendors keep absorbing small teams. The second is the push for openness at the tooling layer, visible in other open projects such as Worktrunk for running parallel agents with git worktrees and Agent AutoLoop for background agent runs. HumanLayer's writing on skills also shaped practice, as in our look at a coding-agent skill for visual output from its founders, and the broader discipline of loop engineering with coding agents.
What is not known
- Deal value and structure. Neither post gives a price.
- Length of the transition. Factory says "a transition period" without a number in the text we read.
- Which HumanLayer features reach Factory, and when. Factory names examples and says "in the next few weeks," which is a plan.
- Future of the hosted cloud. HumanLayer said on October 23 that hosted users saw no change for now. The new post points to a transition. Expect the hosted product to wind down at some point, but no date is given.
- Governance of the open-source repo. Factory says it will open source HumanLayer, while HumanLayer says the core is already open. How the project will be maintained under Factory is not spelled out.
What to do if you use HumanLayer
- Export your data now, even though the platform stays up. A transition is when settings and history are easiest to lose.
- Try the self-hosted route in a test environment using the docker compose file, so you know the fallback works.
- Check which integrations you depend on. The open-source release lists GitHub, Linear, Jira and Slack.
- Read the license. Apache 2.0 allows commercial use and modification, with the usual notices and patent terms.
- If you want Factory, wait for the promised multiplayer features before moving a whole team, and test them on one project first.
How acquisitions of small agent tools usually go
Small developer-tool teams are often absorbed for their people and ideas more than their revenue. Factory's wording fits that pattern: it talks about aligned missions and bringing HumanLayer's best features into its own product, and it counts HumanLayer's users in the hundreds of daily active users, not in the millions. For the acquirer, the gains are talent, a developer community and a body of public writing. For the acquired team, the gains are distribution and enterprise reach. For users, the risk is a forced move to a product with a different price and different priorities. An open-source core, as here, reduces that risk but does not remove it, because running a self-hosted stack takes real effort.
What to do if you evaluate agent tooling
Ask vendors three questions: is the core open under a permissive license, what is the export path, and what is the plan if the company is acquired? HumanLayer's answer to the first was yes before the acquisition. That makes this an easier case than a closed tool bought by a larger rival.
Related reading
- What is a software factory
- Cognition SWE-2 Flash
- Worktrunk for parallel agents
- Agent AutoLoop
- Loop engineering with coding agents
Details come from Factory's and HumanLayer's own posts as read on November 4, 2026 and may change. Follow @explainx_ai for updates.