Jacob Tomlinson
#coding-agents
Tag
-
Giving Agents Minimal Tools
Agents are powerful because they can reason their way towards a goal using tools to interact with the world. Some tools gather new information, other tools perform actions. The most compelling agentic workflows are tasks where you have a well defined end goal, but the intermediate steps are unclear.
-
Documenting your software libraries for agents
This post is a companion to a talk I gave at PyData London 2026. You can watch the recording here and see the slides here.I’ve spent a lot of my software engineering career working on open source tools, libraries and frameworks. These are chunks of software that other software engineers use to build their software. I make things for makers, and I like that. A large part of working on these projects is telling people about the library, and explaining to them how to use it. Open Source Software Engineers spend a big chunk of their time on grassroots marketing, getting other engineers to use their code.
-
Guiding your contributor's agents to better behaviours
Many open source maintainers have noticed an uptick in low-effort AI generated PRs recently, myself included. The most frustrating of these is when someone prompts their agent to
"Fix <url to issue> and make a PR with the changes". Reviewing these PRs can be time consuming because diffs can be large and the contributors rarely respond to review feedback, they just prompt and move on.