What we wish we knew about building AI agents

kept by eddie mar 25

PostHog shares lessons from two years building AI agents, urging teams to validate demand with simpler interfaces like MCP servers before committing to custom agents.

The evolution of agentic surfaces: building with Claude Managed Agents
Claude Managed Agents decouples model reasoning from code execution to let teams deploy production-grade agents without building custom infrastructure.
also on Claude Agent SDK, Anthropic, #ai-agents, #developer-tools
Harness design for long-running application development
Anthropic engineer Prithvi Rajasekaran builds a multi-agent harness with planner, generator, and evaluator agents that produces full-stack applications over multi-hour autonomous coding sessions.
also on Claude Agent SDK, Anthropic, #ai-agents, #developer-tools
How much can you delegate to agents? - by Jina Yoon
The article presents a practical framework for deciding how much autonomy to grant AI agents based on task checkability and undo cost, outlining four levels from assistant to self‑driving mode.
also on PostHog, #ai-agents, #developer-tools
PostHog Code
PostHog Code uses production data signals to automatically diagnose issues and generate pull requests before users know there's a problem.
also on PostHog, #ai-agents, #developer-tools
How to Work and Compound with AI
Treating AI as a collaborator requires structuring context as infrastructure, encoding taste as config, shifting verification early, delegating bigger tasks, and closing feedback loops to compound improvements.
also on MCP, Anthropic, #ai-agents, #developer-tools