Skip to content
▲ Blog

What I learn
running on agents.

Field notes from César Soto on building software with agents without losing quality: what works, what does not, and how the team lives it.

▲ Agents in production

· 9 min read

How do you keep your AI assistant from seeing another customer's data or acting without permission?

An in-product AI assistant should reach only what the user's own screens reach, and a person applies its changes. This is how we launched it at Simón.

Read article →
▲ AI productivity

· 9 min read

Does AI make developers slower? What METR measured, and what it corrected in 2026.

In 2025, METR measured 19% more time with AI; in 2026 it published new data that is not conclusive either. What to measure in your team meanwhile.

Read article →
▲ Agents in production

· 6 min read

Human approval without the bottleneck.

How I decide which AI agent actions wait for a person: each tool is classified by impact and reach, approval is asynchronous, and autonomy is earned.

Read article →
▲ Agent-based development

· 6 min read

Refinement agent: criteria an agent can verify.

How a refinement agent reads a requirement against the code, asks the product owner, and leaves acceptance criteria another agent can verify.

Read article →
▲ Agents in production

· 6 min read

The agent proposes, the backend executes.

Where I draw the line between an AI agent and a transactional backend: typed output with Zod, a domain service, tools with an injected tenant, and idempotency.

Read article →
▲ Agent-based development

· 6 min read

What a harness is and how it makes an AI agent trustworthy.

A harness is everything around the model in an AI agent. What it is made of, and how an acceptance agent cut what QA sent back to almost zero.

Read article →