February 26, 2026
Rise of the Codebots
Show HN: Agent Swarm – Multi-agent self-learning teams (OSS)
“My CPU Is a Learning Computer”: Hackers React to Self‑Teaching AI Coders Running in a Swarm
TLDR: Agent Swarm is an open‑source “team” of AI coders that learns from past work to write better code over time, stirring both excitement and unease. Commenters joke about Terminator, worry about slowly self‑improving bots, and debate whether indie tools like this can survive against big corporate AI platforms.
Agent Swarm just dropped on Hacker News promising a self‑learning team of robot coders that live in little Docker containers and talk to Slack, GitHub, and email… and the comments instantly turned into a mix of sci‑fi hype and quiet existential dread.
The creator shows up cheerfully announcing they rebuilt everything so the bot army can now “compound” its own knowledge over time — basically, the more it ships code, the smarter it gets. One user immediately goes full Terminator, deadpanning: “my CPU is a neural net processor, a learning computer”, and everyone knows exactly what movie they’re quoting. Another dev confesses they keep “subconsciously rejecting” this whole idea, like their brain is trying to slam the door on the thought of machines slowly learning the entire codebase.
Others are more practical: one commenter calls the memory system “what separates useful automation from novelty”, comparing the agents’ shared vs personal memory to how real teams work. But there’s also a spicy side‑thread from a developer who says, essentially, love the idea, but I’m just going to let Claude build my robot coworkers for me, linking to rival tools and hinting this might be re‑inventing the wheel. Between the Skynet jokes, the low‑key fear, and the "why build this when big AI companies will" debate, Agent Swarm has officially become the latest flashpoint in the “are we automating ourselves out of a job?” saga.
Key Points
- •Agent Swarm is an open-source multi-agent orchestration system for AI coding assistants like Claude Code, Codex, and Gemini CLI.
- •A lead agent receives tasks via Slack, GitHub, Email, or API, breaks them down, and delegates to Docker-based worker agents.
- •Features include task lifecycle management, Docker isolation, service discovery, scheduled tasks, a dashboard, and compounding memory with persistent identities.
- •Setup options include full Docker Compose, local API with Docker workers using bun, and using Claude Code directly as the lead agent; requires API_KEY and a Claude Code OAuth token.
- •Architecture centers on an MCP API server with SQLite; workers run full dev environments (git, Node.js, Python), track progress, and deliver outputs such as PRs and issue updates.