Developers answer AI-driven layoffs with an open-source AI executive team

An open-source project called Open Executive has reached Hacker News's front page with a pointed framing: a chief executive fired developers to make room for AI, so developers released an open-source AI executive. The repository, published by Sente Labs under an Apache 2.0 license, describes a "virtual executive team" — one advisor persona, pitched as carrying business-school-level knowledge and customized to a specific company.

Eight specialists behind one voice

According to the project's documentation, a single orchestrator model receives each user message, calls the relevant specialists in parallel, and stitches their output into one reply. Eight agents make up the bench: chief strategy, finance, HR, legal, operations, marketing and product officers, plus a board communications director. The multi-agent machinery stays hidden; users only ever converse with a consistent executive voice.

Each specialist grounds its answers through two retrieval layers: a built-in library of MBA-style reference material that ships in the repo and is seeded into an embedded ChromaDB vector store at startup, and the company's own uploaded documents, chunked into a separate collection. Retrieved context is injected into the user's turn rather than the cached system prompt — a design choice the docs say keeps cache hit rates as high as 85 percent after the first few exchanges.

An executive that remembers and nudges

Beyond question-and-answer, the system maintains episodic memory. After every response, a lighter background Claude model extracts decisions, initiatives and advice into SQLite, and the next session opens with a summary of what the executive previously recommended. A scheduler runs background jobs that can proactively surface follow-ups and deadline-sensitive items. The documentation carries an honest operational caveat: the job runner claims work atomically to avoid firing twice, so the API must run as a single instance and should not be scaled horizontally without gating the scheduler first.

Under the hood

The stack is current and unglamorous: a Python 3.11 backend on FastAPI managed with uv, ChromaDB for vectors, SQLite for memory, and a Next.js 15 web UI styled with Tailwind. Anthropic's Claude API supplies the reasoning: claude-sonnet-4-6 handles the orchestrator persona and most specialists, while claude-opus-4-7 with extended thinking covers the strategy, finance, legal and board functions. An onboarding wizard collects the company profile — industry, stage, business model, competitive landscape, priorities, culture — with optional financial details and document upload. The system reaches users through Slack, email, Telegram, Google Chat, Discord, a CLI and the web UI, and ships with Fly.io deployment configs. First boot downloads a roughly 90 MB embedding model to build the local index.

The provocation

Notably, the repository itself reads as a straight product pitch and never mentions any layoff or CEO; the barb is entirely in how the Hacker News post framed it. The implication lands anyway. Executive messaging about AI has largely aimed at headcount below the leadership layer, with the C-suite positioned as the people who deploy automation rather than candidates for replacement. Open Executive inverts that: if an AI system can plausibly absorb analyst and engineering work, the same argument applies to strategy memos, board decks and OKR reviews.

Why it matters

The project works on two levels. As commentary, it forces the AI-and-jobs debate to confront its own logic: automation rhetoric tends to stop at the top floor of the org chart, and this makes that boundary look arbitrary. As software, it is a tidy reference implementation of the pattern now marketed as the AI employee — orchestrated specialist agents, retrieval over private documents, episodic memory, prompt caching and proactive scheduling, assembled from off-the-shelf parts and an API bill. Whether or not anyone hands their company's direction to it, the building blocks are real, openly licensed, and increasingly cheap to copy.