March 22, 2026
Robots pulled an all-nighter
Show HN: A Markdown file that turns your AI agent into an autonomous researcher
Dev drops single file that turns chatbots into all-night lab assistants — internet freaks out, divides into “shut up and take my bugs” vs “please stop replacing me”
TLDR: A simple text file now lets AI coding helpers run dozens of experiments on your code while you sleep, keeping only the best changes like a tireless lab assistant. The community is hilariously split between “this is my dream intern” and “this is how I get replaced by a .md file,” and that tension is why it matters.
A one-page file called researcher.md just hit Hacker News and instantly split the crowd into two loud camps: the people screaming “I’m going to bed while my robot co‑worker runs 30 experiments,” and the people quietly Googling “early signs my job is being automated.” The file basically tells an AI coding helper how to behave like a hyperactive scientist: it asks what you want to improve, spins up a special branch of your code, tries out idea after idea, keeps what works, throws away what doesn’t, and writes lab notes like a straight‑A student who’s terrified of losing marks.
Fans are hyped, calling it “unit tests for your ideas” and bragging about letting the bot attack slow code, bloated apps, and even messy AI prompts while they sleep. Skeptics are side‑eyeing the whole thing, warning this is how you wake up to a broken app and a very confused manager. One camp is posting memes about “interns who work 24/7 for free,” while the other is imagining a future where your performance review is “why can the markdown file ship more experiments than you?” The hottest micro‑drama: is this genius, or just a fancy way to let an overconfident robot make prettier mistakes, faster?
Key Points
- •The article describes a single Markdown file, `researcher.md`, that turns compatible AI coding agents into autonomous researchers capable of designing and running experiments on code.
- •Once configured, the agent can conduct over 30 experiments autonomously (e.g., overnight), testing hypotheses, discarding failures, and keeping successful changes.
- •An example experiment replaces an O(n²) neighbor search with a KD-tree implementation using `scipy.spatial.KDTree`, reducing p99 latency from 142 ms to 89 ms.
- •The agent operates on git branches, interviewing the user, creating a research branch, committing before each experiment, reverting on failure, forking branches, and logging all actions.
- •Experiment history is stored in an untracked `.lab/` directory that persists through git operations, and the tool is released under the MIT license.