AI Weekly #001

Welcome to the first issue. I'll keep this short and signal-dense — no fluff, just things worth your attention as a developer building real things.

What's Moving

Reasoning models are getting cheaper fast. The gap between frontier models and what you can run locally is closing. Worth watching if you're building anything with LLMs — your infra costs in 12 months will look very different.

Agentic frameworks are still a mess. Everyone is shipping agents. Few work reliably in production. The tooling is improving but we're still early. Don't over-invest in any one framework yet.

Context windows keep growing. 1M+ token windows are becoming standard. This changes the architecture of a lot of applications — you need RAG less often, but prompt cost management matters more.

What I'm Using

• Claude for long-context reasoning and code review
• Cursor for day-to-day coding — still the best IDE integration
• Local models (via Ollama) for quick iteration where latency or privacy matters

One Thing Worth Reading

The framing of "AI as a junior developer" is wrong. Better framing: AI as a senior developer who needs precise requirements and gets bored with ambiguous tasks. Write prompts like you're writing a good ticket.

More next week.