Cesar Toscano

My job is to create an environment where agents do their best work.

My coding agents check their own work. That's how I built an MCP server a startup's sales team ran outreach through, and an LLM pipeline with evals for therapeutic programs.

workflowharnessagent runverifierhuman gatestart from the work, not the APIfails go back, not forwarda person sends

Four ideas, one essay each

I

Workflows define the spec

Start from the work people do, then the data it needs, then the tools.

II

The harness carries the standards

Agents copy what they see, so issues, checks and docs are the job description.

III

Loops need a verifier

A change stays only if something outside the agent says it's better.

IV

Humans own the irreversible step

The agent drafts and queues. A person sends, and can undo.

Now
Founder, Weller Labs · AI engineering and GTM systems engagements
Before
Mahana Therapeutics, Bitmatica, StackLead, AVG
Based
San Francisco
School
MIT