Lab automation with Claude Code & Claude Cowork

Overview

Manual pipetting is a rate limit. Liquid handlers, open-source lab robots, and programmable incubators let a solo scientist run experiments an order of magnitude faster — if they have the software discipline to avoid chaos. This course teaches you to build that discipline using Claude Code as the execution layer and Claude Cowork as the planning and QA layer.

How Claude Code and Claude Cowork fit

• Claude Code connects to your instruments through APIs, OPCUA, serial, or SiLA drivers. It generates protocol scripts, runs them, captures logs, and files results back into your project repo.
• Claude Cowork hosts a lab operations team: a Protocol Planner agent, a Run Monitor agent that watches live run telemetry, and a QA / Audit agent that reviews every run before the next one starts. You remain the human approver for anything high-risk.

Who it's for

• Solo researchers and small labs adding their first Opentrons, Hamilton, or custom robot
• Biotech startups standardizing protocols across multiple sites or benches
• Dev-tools-curious biologists who want to treat wet lab like software

What you'll build

• A protocol library version-controlled in git and executed by Claude Code
• A Claude Cowork workspace with Planner, Monitor, and QA agents sharing run data
• At least one end-to-end automated workflow — for example, PCR setup, plate normalization, or a colony screen

Prerequisites

• At least one programmable instrument (Opentrons OT-2, a thermocycler with an API, or equivalent)
• Comfort writing small Python or TypeScript tools
• Bench experience so you can catch when automation is doing the wrong thing

Tools and setup

1. Inventory controllable instruments, their protocols, and their failure modes
2. Install Claude Code and connect it to each instrument via official SDKs
3. Create a Claude Cowork project with Planner, Monitor, and QA agents

Modules

Module 1: Protocols as code

You will move one manual protocol into a scripted, parameterized, version-controlled form that Claude Code can execute reproducibly.

Module 2: Supervised autonomy with Cowork

You will wire Planner, Monitor, and QA agents into the run loop so proposed protocols are reviewed before execution and outcomes are reviewed before the next run.

Module 3: Scale and harden

You will add calibration routines, error-handling, and human-approval gates for anything irreversible — reagents, organisms, or long-running experiments.

Deliverable

A fully automated workflow that a new team member could rerun from a clean clone of the repo, with a Cowork transcript showing how the agents planned, monitored, and audited each run.

Common mistakes

• Giving agents permission to run irreversible actions without explicit human approval
• Letting the Monitor agent watch logs it cannot actually parse, creating false confidence
• Skipping calibration and drift checks, then debugging ghosts for weeks

Next steps

Integrate lab automation with the CRISPR and structure-prediction courses so your Cowork team can plan the edit, predict the product, run the bench, and analyze the data in one linked project.