Niels Berglund

Interesting Stuff - Week 44, 2026

Week 44 roundup: Claude Code skills nobody writes by hand, Steve Yegge's Counting House for enterprise AI, and why hunger beats tenure for junior engineers.

Interesting Stuff - Week 39, 2026

Week 39 roundup: Claude Opus 5.5 gets cheaper and calmer, an agent teaches itself to out-benchmark Rust libraries, and a model built to decide instead of chat.

Interesting Stuff - Week 38, 2026

Week 38 roundup: Anthropic's 25x CI explosion and how they fixed it, 800,000 lines of agent-written Rust at GitHub, and Cowork folding back into Claude.

Interesting Stuff - Week 37, 2026

Week 37 roundup: why the dark software factory is a lie, OpenAI's numbers on agent-accelerated research, Anthropic's intent.md blueprint, and the cost of automating your own brain.

Interesting Stuff - Week 36, 2026

Week 36 roundup: Pinterest on what it really takes to become an AI team, plus Claude Fable 5.1 and Mythos 5.1 - cheaper cache reads, protein binders, and three breaking API changes.

Interesting Stuff - Week 35, 2026

Week 35 roundup: Claude Academy and AI fluency, Anthropic's AI-native SDLC playbook, Steve Yegge's fences instead of sandboxes, and Andrew Ng's AI Engineering Skills Map.

Interesting Stuff - Week 33, 2026

Week 33 roundup: agent memory you can actually trust, the vanishing middle class of software engineering, OKF token hand-offs, cheaper Claude Code sessions, and Cassandra to PostgreSQL.

Interesting Stuff - Week 27, 2026

Week 27 roundup: is agent 'self-improvement' just string mutation, SkillOpt trains skills like weights, agentic loops explained, and Fable 5 is back.

Interesting Stuff - Week 11, 2026

Week 11 tech roundup: Karpathy's Autoresearch lets AI agents run autonomous ML experiments, Anthropic launches the Claude Marketplace and code review, Joe Njenga explores Claude Code /loop, Andrew Ng's Context Hub tackles agent drift, Microsoft's AgentRx debugs agent failures, and Cash App's monorepo migration story.

Interesting Stuff - Week 10, 2026

Week 10 tech roundup: Netflix battles kernel-level mount lock contention at scale, Confluent explores autonomous data systems for real-time decisioning, Steve Yegge federates a thousand Gas Towns into the Wasteland, and Anthropic's Skill-Creator plugin brings rigorous testing to agent skills.