@AleFranz

Principal Software Engineer ¦ .NET lover

 London, UK
alessio.franceschelli.me/
Joined March 2009
  • Tweets2.5K
  • Following1.1K
  • Followers286
189Media
You drop an AI agent into your real codebase and it produces garbage. Wrong style, subtle bugs, side effects everywhere. The problem isn't the model. Your codebase isn't ready for AI. Think of it like handing off a biathlon relay with no map of the course 🥇🧵👇
1
86
Running AI coding tools on Windows can feel like playing on hard mode, especially when your PC is also your gaming rig 🎮 OpenCode 2 works well out of the box, but three quick settings make it run much cleaner. 🧵👇
1
18
more replies
Running local models? Tune compaction in opencode.​jsonc: "compaction": { "auto": true, "keep": { "tokens": 65536 } } The default keeps ~15k recent tokens. With a big local context window, 64k keeps recent work intact without premature amnesia.
1
10
Full guide, with the trade-offs and how to undo each setting: alessio.franceschelli.me/pos…

Suggested OpenCode settings for Windows

Three practical settings for running OpenCode 2 on Windows: fix in-app updates, stop the background service when you're done, and tune context compaction for local models.

alessio.franceschelli.me
7
If you followed my guide for using the Windows ssh-agent in WSL, there’s a small improvement worth making. Add unlink-early to the socat command so a stale socket does not stop the bridge restarting after WSL shuts down. Updated guide here: alessio.franceschelli.me/pos…

Use the Windows ssh-agent in WSL without breaking auto-poweroff

Expose the Windows ssh-agent to WSL while keeping WSL's automatic shutdown behaviour intact.

alessio.franceschelli.me
18
AI works best where it can test itself. Software can do thousands of automatic loops: make a mistake, run tests, fix it, repeat. In work that needs an expert to review every attempt, where does the big saving actually come from?
1
32
Can AI work outside of software development? alessio.franceschelli.me/pos…

Can AI work outside of software development?

AI works well in software because it can test and correct itself automatically. In jobs that still need expert review at every step, the promised saving gets much smaller.

alessio.franceschelli.me
16
A wizard is never late. Nor is he early. Unlike my read replica, which is currently behind.
26
A few days ago I published Burla. This follow-up is about the part I care more about: the design bets behind it. Toy examples are where mocking libraries look tidy. Real migrations are where they tell the truth. Thread below. 🧵👇
1
27
more replies
I do not think the AI angle is marketing fluff. Inconsistent APIs tax humans and LLMs in the same way: more context spent on framework quirks, less energy left for the code under test.
1
10
Full write-up here: alessio.franceschelli.me/pos…

Why Burla is built for real migrations

Burla was designed around the ugliest part of mocking libraries: moving real test suites without turning the diff into soup.

alessio.franceschelli.me
7
I finally published Burla, a .NET mocking library I have been rewriting for months while pressure-testing models, local setups, and agentic coding workflows. I needed a real project, not one more opinion about AI. Built with LLMs in the loop, for the LLM era. Thread below. 🧵👇
1
40
more replies
Here are the Docs: alefranz.github.io/Burla/
1
5
Repo, NuGet, and the broader story are in the blog post: alessio.franceschelli.me/pos…

Burla is finally public

Burla is a lightweight .NET mocking library I built while pressure-testing models, local setups, and agentic workflows. Now it is public.

alessio.franceschelli.me
6
Load more