Past issues. Every edition of Agents at Work, newest first.

🏗️ Agents Are Infrastructure Now

OpenAI entered managed agent operations on October 20. KPMG and Microsoft committed seven years to agentic AI on October 21. Only 21% of large enterprises see how their work actually flows.

🛡️ Lawful by design

Three enterprise platforms shipped major AI agent launches this week, with governance built into the runtime.

🤝 Your AI colleague got an inbox

Google gave its agents four kinds of memory and, for coworker agents, an inbox of their own. Cohere shipped memory with usage controls.

🔑 Who says “go for it”?

Meta's agent is getting a device, payments and more shops. The approval step is what keeps you in charge of it.

🤝 Agents in Public

Two launches this week put AI agents in front of your customers. What they say is on you.

🛡️ Governance Is the Product Now

Last week OpenAI and Meta both led their AI launches with safety architecture — following a line Microsoft drew ten months ago. That's not coincidence; it's what enterprise teams are buying.

🧠 The memory layer

Three companies built AI that gets smarter every time an expert corrects it. One PM built it for himself.

🗂️ The handoff

Workato launched free project-owning agents. AccuKnox shipped the governance stack to match. Here's what both mean for teams responsible for AI at work.

🔐 The Five Things That Go Wrong

This week, three organizations named the failure modes for enterprise AI agents — and one of them built a product category around the job of managing them.