AUTOMATION

9 min

Agents in production: what actually breaks

Agents in production: what actually breaks

Agents in production: what actually breaks

Agents rarely fail at reasoning. They fail at retries, permissions and the edge case nobody logged.

In production the interesting failures are boring: a stale credential, an API that answers slowly instead of failing, a tool call that succeeds with the wrong scope. Observability beats cleverness.

EM

Elena Marchetti

Managing Partner · Writes on operating models for applied AI

Working on something like this?

Alyssa

Consulting, data and applied AI for organizations that intend to lead.

NEWSLETTER

One considered email a month. Nothing else.

© 2026 Alyssa. All rights reserved.

Alyssa

Alyssa

Consulting, data and applied AI for organizations that intend to lead.

NEWSLETTER

One considered email a month. Nothing else.

© 2026 Alyssa. All rights reserved.

Alyssa

Alyssa

Consulting, data and applied AI for organizations that intend to lead.

NEWSLETTER

One considered email a month. Nothing else.

© 2026 Alyssa. All rights reserved.

Alyssa

Create a free website with Framer, the website builder loved by startups, designers and agencies.