Thoughtworks engineer tests local models for agentic coding on Apple M3 Max and M5 Pro
bibryam · x · 2026-10-05
Thoughtworks Distinguished Engineer Birgitta Böckeler has spent 4 weeks re-evaluating local models for coding, after years of disappointment. Her focus is agentic coding — not just autocomplete — and out-of-the-box usability for developers unwilling to fiddle with specs and tooling.
Key points:
- Test hardware: Apple M3 Max (48GB RAM) and Apple M5 Pro (64GB RAM)
- Many interacting factors make it tedious to find the best setup under resource constraints, and hard to filter signal from noise in online success stories
- A counterintuitive finding: in automated evals, one model clearly performed better on the stronger machine
This is the intro memo; a follow-up will detail her hands-on experience.
Related event: Thoughtworks Engineer Tests Local Models for Agentic Coding(2 posts)→
More from coding & agent
- Jevbox: open-source permission-aware document library with cited AI chat — letandrewcook · 2026-10-05
- Keyline MCP server cuts agent token use 2-6x by replacing HTML screenshots with JSON scenes — yuvalt · 2026-10-05
- Five cheap patterns to make MCP tools tell agents what they didn't check — Honest_Traffic_8613 · 2026-10-05
- Designing a programming language for AI agents, not humans — mark_k · 2026-10-05
- AI agents are about to flood the workforce, and no one's ready: WIRED — Dr_Singularity · 2026-10-05
- New Paper IGP-Bench Teaches AI Agents When to Stop Chasing a Wrong Idea — _sathvikr · 2026-10-05