Hello Bioinformatics lovers,

Tommy here. Let's talk about Good Bioinformatics workflow and it is very similar to a good workflow for Claude Code.

Anthropic’s recommended workflow for Claude Code is four steps: Explore, Plan, Code, Commit.

Every bioinformatician I know who is good at their job has been doing this for years. Just without the AI part.

Here is why it maps so cleanly.

Explore.

Claude Code: read the relevant files before touching anything. Understand what exists. Map the dependencies.

Bioinformatics: look at the data before you analyze it. Plot the distributions. Check the metadata. Count the NAs. Ask the wet-lab person what they actually did. Read the existing pipeline.

Most bad analyses come from skipping this step. Most bad PRs do too.

Plan.

Claude Code: define success criteria up front. What does done look like? What are the edge cases? What tests need to pass?

Bioinformatics: define the hypothesis. What is the contrast? What is the null? What confounders need to be modeled? What would falsify this?

If you cannot write the plan in a paragraph before you run anything, the analysis is not ready. Neither is the code. Same failure mode, different field.

pro tip: always use /plan command for any bioinformatics analysis with Claude Code.

Code.

Claude Code: implement. This is the easy part if the first two steps were done well.

Bioinformatics: run the analysis. Also the easy part if the first two steps were done well.

It is wild how much time people spend here trying to compensate for skipping the first two. It does not work. You cannot debug your way out of a missing hypothesis.

Commit.

Claude Code: run subagent reviewers before pushing. /ultrareview catches the obvious bugs so humans can focus on the architectural calls.

Bioinformatics: peer review before publication. Read your own results out loud. Have someone who was not in the room look at the figures and tell you what story they tell. Sanity check against the biology.

Both fields have the same problem: the person closest to the work is the worst person to catch the obvious mistakes. You need a separate reviewer with separate attention.

The reason this workflow generalizes is that it is not really about coding or bioinformatics. It is about how careful work gets done. Understand the context. State your goal in writing. Do the work. Have someone check it.

AI agents make this loop faster. They do not change the structure. The bioinformaticians on my team who are best at using Claude Code are not the ones with the most ML background.

They are the ones who already did experimental design well. They had the workflow. The AI just made it cheaper to execute.

If you are a junior bioinformatician trying to figure out what skill compounds the hardest in an AI-assisted world: it is not learning more tools. It is learning to do these four steps deliberately.

The tools change every six months. The discipline does not.

What does your version of “Explore” look like before you start an analysis? Reply and tell me — I read every response, and the good ones often become future newsletters.

Happy Learning!

Tommy aka crazyhottommy

PS:

If you want to learn Bioinformatics, there are four ways that I can help:

1. My free YouTube Chatomics channel, make sure you subscribe to it.
2. I have many resources collected on my github here.
3. I have been writing blog posts for over 10 years https://divingintogeneticsandgenomics.com/​
4. Lastly, I post daily on Linkedin and you may find gems like this one https://www.linkedin.com/feed/update/urn:li:activity:7462136205346963456/​

Stay awesome!

PPS: