IndieFounder publication
Practical stories, product lessons, teardowns, and field notes for independent founders.
Startups
Classify decisions by the cost of reversing them. Move quickly on reversible experiments and spend more evidence on migrations, contracts, hiring, or du…
Archive
Startups
Classify decisions by the cost of reversing them. Move quickly on reversible experiments and spend more evidence on migrations, contracts, hiring, or du…
AI
MCP clients connect agent runtimes to external capabilities, but discovery does not equal authorization.
AI
Choose the model from a workflow benchmark. The real unit is cost per successful task after retries, tool calls, latency, and human corrections.
AI
A production agent needs separate ownership for request context, run state, sessions, durable business data, approvals, and audit events.
AI
An external API may return far more data than an agent needs.
AI
AI coding agents are taking on larger engineering tasks.
AI
Compare OpenAI and Claude on the runtime details that affect your workflow: tools, state, sandboxing, latency, cost, and evaluation.
AI
Anthropic's Managed Agents and Claude Agent SDK put the agent loop in different places. The trade-off is operational control versus managed infrastructure.
AI
External providers change response formats, endpoints, limits, and authentication rules.
AI
Build a focused agent around the Responses API with typed tools, server-side authorization, safe retries, and an evaluation loop.