# ai-employees ships business roles as folders of scheduled routines

> ai-employees is a MIT licensed npm package that installs eight business roles, 60 scheduled routines in total, onto the coding agent you already use. It suits a small team automating one role at a time, and it hands the agent control of your browser.

**markfulton/ai-employees** — Open source AI Employees. 8 scheduled business roles, 60 routines, on Claude Code and 10 other harnesses. They drive your browser the way you do and improve every run. You own the files.

- Repository: https://github.com/markfulton/ai-employees
- Stars: 765 · Forks: 221
- Language: JavaScript
- License: MIT
- Published: 2026-10-21 · Updated: 2026-10-21 · Language: en
- Canonical page: https://hysenlabs.com/projects/markfulton-ai-employees

## An AI Employee is a folder of routines, not a model

The unit of this project is deliberately boring. The README defines an AI Employee as a folder of scheduled routines that covers one business role, and the repository layout backs that up with an employees directory alongside skills, shared, installer, docs and assets. Nothing in the tree suggests a model, a fine-tune or a hosted inference service. The package description makes the same point in a different register: the AI agent you already use, running on your own machine, with the claim that every run makes the next one better. The consequence for evaluation is straightforward. You are not deciding whether a vendor's model is good enough; you are deciding whether a set of routines, prompts and schedules produces work you would otherwise do yourself.

## Eight roles and sixty routines, with the table cut off mid-list

The role table is the centre of the README, and it lists for each employee what it owns and how many routines it carries. The GTM Engineer owns positioning, the launch board, outbound drafts, directory and press forms and a weekly scoreboard across 8 routines. The SEO/AEO Employee handles keyword research, one article a weekday, publishing, indexing and rank review across 8. The Web Dev Employee covers site health, error triage, small changes on a branch and dependency review across 8. The Social Media Employee drafts posts per platform with a veto window across 7 routines, the Ad Manager Employee handles account reads and creative sets with money moving only on approval across 7, and the Sales Employee works prospect sweeps and first touches across 7. The table in the README is truncated at that point, so the remaining roles have to be read from the repository directly.

## Approval gates are the design, not a feature list

Two entries in that table describe a control that matters more than the rest. The Social Media Employee drafts replies and posts for you but holds a veto window, so the output exists for review rather than going live unattended. The Ad Manager Employee moves money only when you approve, and it produces a weekly change list rather than acting silently. Read together with the GTM Engineer entry, which produces outbound drafts and press forms, the pattern is that the agent prepares and the human disposes. That is a defensible default for anything that touches a public account or a budget, and it is also the practical limit of the whole system, since every routine that does not need a human in the loop is the one that will need you to decide how much you trust it.

## Thirteen harnesses, one filesystem contract

The harness strip names the agents the routines are expected to run on: Claude Code, OpenClaw, Hermes, OpenCode, Grok Bot, Codex, Antigravity, Pi, Cline, Qwen Code, DeepSeek, Muse and Dots. That is the widest claim in the project and the one the repository layout supports least visibly, since the tree contains a .claude-plugin directory but no equivalent for the other twelve. The mechanism implied is that a routine is a set of files and scheduled instructions that an agent can be pointed at, rather than a plugin API that only one harness implements. The README also states that the employees drive your browser and your PC the way you do, which is why browser automation sits inside the routine definitions rather than behind a service API.

## The npm package is an installer with a hire verb

Distribution runs through npm under the package name ai-employees, with the binary mapped to installer/cli.mjs and the published files limited to the installer, employees, skills, docs, README, LICENSE and TRADEMARKS. The package version is 1.9.0, the type field is module and the engines requirement is Node 18 or later. The install and hire actions live in the scripts section of the manifest.

```json
  "scripts": {
    "test": "node .github/scripts/selftests.mjs && node .github/scripts/no-dashes.mjs && node .github/scripts/aeo-check.mjs && node .github/scripts/work-cycle-check.mjs && node .github/scripts/upgrade-check.mjs && node .github/scripts/release-check.mjs",
    "hire": "node installer/cli.mjs hire"
  },
```

The hire entry point is the CLI itself, so the role you want is picked and installed through that command. The test script is worth noting for a different reason: it runs six separate checks, including one for dashes and one for AEO behaviour, which suggests the project treats its own output format as something to test rather than assume.

## The repository is unusually explicit about what it is not

Several details in the README set this project apart from the usual agent demo. The phrase in the repository description is that every routine and schedule is in the repo and nothing is held back, and the owner claim repeats it. The project also ships SECURITY.md, CONTRIBUTING.md, CREDITS.md, TRADEMARKS.md and AGENTS.md at the top level, which is the paperwork set you would expect from something asking to be taken seriously as a dependency rather than a demo. The MIT licence covers the code, and the trademark file exists because the employees have names and identities that the project wants to keep distinct from the routines themselves. The one thing the repository does not have is a GitHub release history, so the version in package.json is the reference point.

## The real cost is the machine you hand over

The limitations are not hidden but they are not in a limitations section either. The employees run on your own machine and drive your browser and your PC, which means a routine that publishes, files press forms or sends drafts has the same account access you do, and a routine that goes wrong goes wrong on the machine where your work happens. The improvement claim, that every run makes the next one better, also depends on state that lives outside the repository. The author runs the business behind the project every weekday and gives a start date of August 27, 2026, which dates the claim to a matter of weeks rather than a track record. The last push was on 2026-10-17, so the code is moving, and with 60 routines across eight roles there is a lot of prompt surface to review before trusting it with a publishing account.

## Conclusion

Adopt ai-employees if you already run a coding agent on your own machine and want one business role, such as SEO or social posting, handled by scheduled routines you can read and edit, because the files stay yours and nothing is held back. Skip it if you need a managed, multi-tenant product with audit trails and role-based access, since this repository is a set of prompt files and a local installer rather than a service. Verify first that the agent you use is on the supported list of thirteen harnesses, that the role you are installing matches the directory you actually want, and that the npm package version matches the routines you read in the repository before you let anything run on a schedule.

## FAQ

### What is ai-employees?

It is an MIT licensed npm package that installs eight business roles as folders of scheduled routines, 60 routines in total, onto the AI agent you already use on your own machine.

### Which agents does ai-employees work with?

The README names thirteen harnesses, including Claude Code, OpenCode, Codex, Cline, Qwen Code, DeepSeek, Hermes, Antigravity, Grok Bot, Pi, Muse and Dots.

### Do the AI Employees act on their own?

Not for the outward facing work. The Social Media Employee holds a veto window for posts and replies, and the Ad Manager Employee moves money only when you approve, producing a weekly change list instead.

### How do I install an AI Employee?

The npm package exposes the installer as the ai-employees binary, and the package scripts include a hire entry point that runs the installer CLI. The published package requires Node 18 or later.

## Sources

- [Issues](https://github.com/markfulton/ai-employees/issues)
- [License: MIT](https://github.com/markfulton/ai-employees/blob/main/LICENSE)
- [markfulton/ai-employees on GitHub](https://github.com/markfulton/ai-employees)
- [README](https://github.com/markfulton/ai-employees/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/markfulton-ai-employees