18 Months Building an Autonomous Agent: What Worked
In early 2025, we started with a deceptively simple goal: build an AI system that could plan,...
Tag archive
In early 2025, we started with a deceptively simple goal: build an AI system that could plan,...
Quando i modelli migliorano, le regole a mano diventano debito tecnico. Tre principi per costruire...
My inbox agent skips anyone who already has an open opportunity. It looks like good manners. Then I counted the CRM it writes into: twenty open opport

Giving an AI agent ten tools often makes it worse, not smarter. Here is how to scope tool access so agents stay fast, cheap, and easy to debug.

Originally published at norvik.tech Introduction Deep dive into the trade-offs between...

Agent tool call cost jumped after you gave it better tools? GitHub hit that on Copilot code review and won ~20% back with a prompt rewrite, not new tools.

Agent-to-human delegation is the pattern nobody designed for: your AI messaging a coworker. Why it lands badly, and the handoff rules that fix it.

Agent harness design decided a benchmark: OpenAI's ARC-AGI-3 score went 13.3% → 38.3% with zero model changes. What that means for your agent loop.
Prevent AI agent overspending. Umair shares "laziest senior dev" patterns like budget cap prompting and tiered actions to build cost aware AI agents. Stop ru...
Anthropic added the /goal command to Claude Code in v2.1.139. You set a completion condition; the...
The May 15 piece on access-rents drew a line through every services industry. On one side:...
CMU's General AgentBench finds giving agents more turns often hurts. Learn why context ceiling and verification gap limit test-time scaling for LLM agents.