Labmate
Labmate is an agent-native data science harness.
Core trick
A managed agent runs each hypothesis as an experiment in an isolated Modal sandbox and records every run in an evidence ledger.
Description
Labmate is an agent-native data science harness. A human gives the business judgment — a dataset, a target, a metric, and constraints — and a real Anthropic Managed Agent (Opus 4.8) does the science: it profiles the data, writes a data contract, proposes hypothesis-driven experiments, runs them for real in Modal sandboxes, critiques its own results for leakage and test-set tuning, folds in plain-language human feedback, and produces a reproducible report with provenance. Built in one day, entirely with Claude Code.
README excerpt
# Labmate — an agent-native data science harness **Live:** https://amazingvince.com · **Repo:** https://github.com/amazingvince/Labmate > Give Claude a dataset, a target, a business objective, and a rubric. A real > **Anthropic Managed Agent (Opus 4.8)** profiles the data, proposes hypotheses, > runs controlled experiments in sandboxes, critiques its own results, asks for human > guidance at decision points, and produces a reproducible report — and **every > hypothesis, run, critique, and decision lands in an agent-native evidence ledger > you can steer live.** 🏗️ **Built in one day at the Claude Build Day (June 13, 2026), entirely with Claude Code (Opus 4.8).** Everything in this repo was created during the event — see [`submission/SESSIONS.md`](submission/SESSIONS.md) for the session-by-session log. --- ## The problem Data scientists spend most of their time on experiment *mechanics*, not science: profiling data, checking for leakage, writing training scripts, comparing runs, documenting decisions, and turning vague business feedback into modeling work. **AutoML optimizes a model once the problem is already framed** — it doesn't help frame the target, define the metric, encode leakage rules, or manage the collaborative loop. | AutoML does | **Labmate does** | | --- | --- | | Optimizes for a fixed target/metric | Helps *define* target, metric, leakage rules, business utility | | Treats experiments as parameter sweeps | Treats experiments as **hypothesis-driven evidenc