July 26, 2026
Stuff I read on the internet — blogs, tweets, threads, newsletters — with a short note on why it stuck. Pick a month to browse, or hit Draw one for a random pick from that month.
lean write-up of the agentic inflection — verifiability as the thing that decides where AI moves fast. Useful framing for Software 3.0: natural language as the new program, and why “vibe coding” is different from disciplined agentic engineering. I keep coming back to the jagged-intelligence idea when explaining what models are good at today.
ow they scaffold planner / generator / evaluator loops so Claude can ship multi-hour apps. The interesting part is the harness itself — sprint contracts, handoffs, and evaluation gates — not just “give the model more tools.” Practical reference if you’re building agents that have to survive beyond a single chat turn.
till the best map of agent building blocks — planning, memory, tool use — before the hype caught up. Reads like a systems cheat sheet: what each component is for, how they connect, and where agents usually fail. I send this when someone wants foundations instead of another framework tour.