A post titled "CEO fired developers to make room for AI. Developers create open source AI CEO" hit the front page of Hacker News on August 27, 2026, and by the time I checked it had 670 points and 438 comments. The open source AI CEO in that headline isn't a stunt or a meme — it's a real, working project called OpenExecutive, and you can run it yourself.
Short answer: The story behind "CEO fired developers, developers built an AI CEO" points to OpenExecutive, a free, Apache-2.0 open source AI CEO tool from Sente Labs that runs eight specialist Claude agents behind one executive persona. It's self-hostable, costs nothing beyond your own Anthropic API usage, and as of this week has 986 GitHub stars and 56 forks.
In my test, I cloned the repo and worked through the setup, the architecture docs, and the eval harness — the parts you can verify without wiring up a paid API key. Here's what the project actually does, what it costs to run, and where the "AI replaces the CEO" framing oversells it.
What you'll need
To run OpenExecutive yourself you'll need Python 3.11 or newer, Node 22 or newer, and an Anthropic API key — the project uses claude-sonnet-4-6 for most of its eight specialist agents and claude-opus-4-7 for the harder reasoning roles (strategy, finance, legal, board comms), so you'll need billing enabled on your Anthropic account. Nothing here requires Claude Pro or Claude Max; this is a separate API-key setup, distinct from a claude.ai subscription. You'll also want uv (Python's fast package manager) and about 15 minutes for the first run, since the initial install pulls in ChromaDB and sentence-transformers and downloads a small local embedding model. If you'd rather not manage Python and Node dependencies on your own machine, a container is the cleaner path — see the sandboxing section below.
Step-by-step: running OpenExecutive
1. Clone the repository
“bash git clone https://github.com/SenteLabsAI/OpenExecutive.git cd OpenExecutive “
The repo is Apache-2.0 licensed, so you're free to self-host, modify, and even fork it commercially with attribution.
2. Set your Anthropic API key
“bash cp .env.example .env “
Open .env and add ANTHROPIC_API_KEY=sk-ant-.... When I read through it, the .env.example file turned out to be the most useful part of the whole repo. It documents 30-plus environment variables, including per-role model overrides. You can swap claude-opus-4-7 for a cheaper model on roles where you don't need the extra reasoning depth.
3. Start the stack
“bash make dev “
This boots the FastAPI backend on port 8000 and the Next.js UI on port 3000 in one command. The README is upfront about the first run being slow — a few minutes while uv sync installs ML dependencies and the app downloads a ~90MB embedding model — and fast on every run after that.
4. Complete the onboarding wizard
The first visit to localhost:3000 walks you through a company profile: industry, stage, business model, competitive landscape, and strategic priorities. Every specialist agent pulls from this profile, so a thin answer here produces generic advice later — treat it like you're briefing a new hire, not filling out a form.
5. Upload documents (optional but worth it)
“bash openexecutive upload deck.pdf model.xlsx strategy.md “
Pitch decks, financial models, and strategy docs get chunked into a separate ChromaDB collection so the Executive can reference your actual numbers instead of generic frameworks.
6. Connect it to where you already work
OpenExecutive ships integrations for Slack, email, Telegram, Discord, and Google Chat, plus a CLI (openexecutive chat). Each one runs inside the same FastAPI process and shares the same SQLite memory store, so a decision the Executive made over Slack on Monday shows up when you ask about it over email on Friday.
Example prompts you can copy
Once it's running, these get you past the generic "give me a strategy" answers:
- "Based on my company profile, what are the top three risks to our runway in the next two quarters?"
- "Draft a board update on our Q3 numbers in the tone I'd actually send — no filler."
- "Pull up what you told me last month about the pricing change and tell me if it still holds given the new competitor."
- "I'm about to make my first engineering hire. Walk me through the compensation bands and equity math for our stage."
- "Review this contract clause and flag anything a non-lawyer founder would miss."
The multi-agent routing is invisible by design — you never pick "ask the CFO" versus "ask the General Counsel"; the orchestrator decides which specialist (or specialists) to consult and returns one voice.
Common mistakes to avoid
The obvious one: treating this as a literal replacement for a human CEO because of the headline. It isn't. It's a research-and-drafting tool organized around C-suite roles, and the project's own maintainers frame it as an advisor, not a decision-maker with signing authority. Second, skipping the onboarding wizard or answering it in one line. The whole value of the multi-agent design is company-specific context. A thin profile gets you the same boilerplate as a plain prompt. Third, running every role on claude-opus-4-7 by default. It's the pricier model, and the project's own config only routes the harder reasoning roles there. Override the rest to a cheaper model unless you're seeing worse answers. Fourth, forgetting this needs its own billed API key. It does not use a claude.ai login, so a Claude Pro or Max subscription won't cover it.
OpenExecutive vs. the alternatives
| Option | Cost | Setup | Best for |
|---|---|---|---|
| OpenExecutive (self-hosted) | Free software + your Anthropic API usage | ~15 min, needs Python/Node/Docker comfort | Founders who want a persistent, document-aware advisor across Slack/email/Discord |
| Claude with a Project and custom instructions | Included in Claude Pro/Max or pay-per-token API | Minutes | One-off strategy or finance questions, no setup tolerance |
| ChatGPT custom GPT | Included in ChatGPT Plus, or API pay-per-token | Minutes | Teams already standardized on ChatGPT over Claude |
| A fractional CFO or startup advisor | $2,000–$8,000+/month, typical market rates | Weeks to find and onboard | Decisions with real legal or financial liability |
The row that actually matters: OpenExecutive's cost is just token usage, not a subscription. The default setup uses claude-sonnet-4-6 at $3 per million input tokens and $15 per million output tokens for most roles, and claude-opus-4-7 at $5/$25 per million tokens for the four deep-reasoning roles. A typical week of asking questions and getting drafted documents back runs a few dollars, not a few hundred, per Anthropic’s published API pricing (checked August 27, 2026).
Tools that make this easier
If you don't want Python and Node dependency management on your own laptop, run OpenExecutive inside a disposable container instead of bare metal. My guide to disposable, isolated sandboxes for AI agents covers exactly this pattern, and it also keeps an agent with bash and file-write access away from your real filesystem. If you're deciding whether Claude or ChatGPT is the better backbone for this kind of always-on assistant, my Claude vs. ChatGPT comparison breaks down where each one actually wins. And if you're new to Claude's ecosystem generally — API keys versus the consumer app, Projects, artifacts — start with how to use Claude AI before you touch OpenExecutive's .env file. For the coding side of the same workflow, my best AI tool for code roundup is the natural next stop.
My take
The headline is doing more work than the project is. Nobody's board fired a real executive team and replaced them with this. The actual story is a group of developers building a genuinely useful, well-documented open source tool as commentary on the "AI replaces workers" narrative, and it's good enough that it picked up 986 stars in under three months. The architecture is sound: routing eight specialist personas behind one voice, with prompt caching and episodic memory, is a smarter design than a single mega-prompt pretending to be an executive.
Where I'd push back on the framing making the rounds this week: young people already distrust AI CEOs as a class, and this project reads more like a joke aimed at that distrust than a genuine threat to anyone's job. The jobs data backs a more nuanced picture than "AI CEO replaces staff," too — see AI and entry-level hiring, what’s really happening to jobs, and why the AI jobs apocalypse probably isn’t imminent. If you run a small company and want a document-aware advisor that remembers what it told you last month, OpenExecutive is worth an afternoon. If you're expecting it to run your company, it won't — and neither will any tool that puts "CEO" in its name.
Frequently Asked Questions
Is OpenExecutive free to use?
The software itself is free and open source under Apache-2.0. What you pay is Anthropic API usage — a few dollars per week for typical use, based on Anthropic's published per-token pricing for the models the project defaults to.
How long does it take to set up?
About 15 minutes for the first run, mostly spent waiting on dependency installs and a one-time embedding-model download. Every run after that starts in seconds, and the onboarding wizard that follows takes another 5–10 minutes if you answer it properly.
Did a real CEO actually fire developers to build this?
No confirmed company or CEO is named in the Hacker News post, the GitHub repo, or Sente Labs' own site — the framing is a pointed, satirical response to the broader "replace developers with AI" narrative, not a documented incident at a named company.
What's the easiest way to try the idea without self-hosting?
Open a Claude or ChatGPT chat, paste in your company profile, and ask it to act as a specific executive role (CFO, CSO, etc.) for one question at a time. You lose the persistent memory and document RAG that OpenExecutive provides, but it costs nothing to try in five minutes.
Does this replace a real CEO or executive team?
No. It's a drafting and research tool organized around executive roles, with no legal signing authority, no fiduciary duty, and no accountability — every mistake it makes is yours to catch before you act on it.