TIL that brew has an auto update command. Goodbye cronjobs to try to keep up with Claude releases. :D
# Install the tap
brew tap homebrew/autoupdate
# Start auto-updating every 12 hours (43200 seconds), including casks and cleanup
brew autoupdate start 43200 --upgrade --cleanup --greedy
Posts by Miguel David
These certainly feel like LLM-generated responses. Not the point I was making in sharing the article.
Managers kill productivity by onboarding contractors like employees. While employees have months to ramp up, contractors lose time for each week of poor onboarding. Set access and concrete tasks on day one.
leaddev.com/communicatio...
The "Claude Code Engineer" orchestrates AI code instead of writing it, working in parallel git worktrees. The bottleneck shifts from fast-typing to defining work well. QA outranks coding; Platform Eng becomes toolsmiths.
rendiment.io/engineering/...
AI is Killing B2B SaaS. Unless they become Systems of Record platforms that can be easily built upon reliably and leverage their data moat.
nmn.gl/blog/ai-kill...
AI fatigue is real and nobody talks about it
siddhantkhare.com/writing/ai-f...
Tagging @buffer.com correctly now. :)
Buffer swag: t-shirt, cup, notebook, pens, stickers and a postcard.
Day 2 at @buffer and I got this nice swag. Way to feel welcomed! 🥰
TIL that for historical reasons Gare du Midi in Brussels has nothing to do with “middle” or “center” but with Midi in France which is … in the South. Now I have to sed s/Midi/Sud/ .
Want to hear what’s going on in many EM’s minds right now with the wave of AI? Read this honest and balanced experience: mediations.candost.blog/p/mediations...
Start today. The perfect conditions will never arrive.
We are fast approaching a reality where human developers and AI agents work together around the same things. Workflows must be adapted to ensure that collaboration adds value to the humans instead of creating friction and noise.
leaddev.com/ai/designing...
Card. It doesn’t run out of battery or connection issues
For the past few months I've evolved a personal harness for Claude Code. But the models keep advancing so my harness keeps evolving and I didn't want to write something out of date the next day. Well, Claude is happy to write AND maintain for me. So here is my WIP-always ephemeral blog post about my
Original link joel.is/posts/176694...
"If you only have goals for one dimension or aspect of life, when you're having a hard time in that area you can feel crushed. If you have multiple dimensions of life you're focusing on with clear goals and progress, you can set yourself up to always be winning in some area." - Joel Gascoigne
Spot on, Klement! 🚀
"just write code" roles largely disappear, and long-term value comes from delivering business outcomes, not lines of code.
individual functions, with hand-written code increasingly restricted in high‑risk domains.
- Engineering teams shrink into minimum viable engineering teams, where fewer, more senior, systems‑thinking engineers focus on organization, communication, and strategy; entry-level ...
- As AI surpasses humans at both writing and reviewing code, humans move into a "task verifier / systems designer" role: defining work, validating outcomes via artifacts (tests, screenshots, videos), managing models, costs, security, and debugging workflows instead of ...
humanwhocodes.com/blog/2026/01...
- AI is rapidly shifting software engineers from coders to orchestrators of autonomous agents, with IDEs and workflows reorganized around managing multiple coding agents rather than editing code directly, likely becoming mainstream by ~2028.
...
- Every "serious" swarm quietly reintroduces hierarchy and deterministic shells — planners, judges, schemas, type systems, tests, and deterministic reducers impose coherence; the LLM swarm is just a noisy candidate generator, not the source of architecture.
- Architecture needs global invariants; swarms are structurally local and forgetful — no shared decision register, no persistent memory of constraints, no authority to reject incompatible work, so duplication and contradictions are the statistical default.
...
buff.ly/4yctcWF
tl;dr
- Swarms can ship big, working systems that are structurally junk drawers — duplicate libs, inconsistent patterns, no coherent architectural choices, just whatever happened to compile and pass tests.
...
Any senior engineer will tell you that less code is better than more, but until now that was hard to prove. With AI being paid by the token and context windows being a limiter, compaction is now a quantifiable strategy. aicoding.leaflet.pub/3may5niwoyk2n