Thoughtworks Technology Podcast · Thoughtworks

Exploring AI agent platforms

·38 min·3 clips

Ben O'Malley argues whether a machine can think is as irrelevant as whether a submarine can swim.

AI is the pressure behind the episode. Ken McRage opens with Ben O'Malley and Fabian Nonmacher, two Thoughtworkers writing a book about AI agents, and the conversation keeps snapping back to production. Agents are interesting, sure, but the useful question is what happens when they can touch internal data, run code, or face the open internet. The security risk is described in practical terms. Internal data access, code execution, and internet exposure make a dangerous mix. That takes the AI chatbot demo out of the toy box and turns it into a release problem. Someone has to design for ownership and security before the thing goes live. Internal use cases get a little more room, but not a free pass. The guests point out that many customers are trying AI inside trusted organizational environments, where the first layer of exposure is lower because the work stays inside existing controls. Lower is not the same as harmless. The episode still pushes zero-trust thinking, inherited permissions, and visible human responsibility, even when the use case stays inside the company. Coding agents make the concern sharper. The merge PR or push point is where generated work stops being a neat demo and starts becoming part of the codebase. Somebody still has to stand behind it. That is where platform design turns into guardrails: who can do what, with which permissions, and who answers for the result. Low-code and no-code come up near the end. The transcript only reaches the start of that thread, but the direction is obvious enough: when building gets easier, guardrails have to get better. The book is still in motion. O'Reilly has moved the expected date to 2027, with the guests hoping for Q1, and chapters are being released for feedback on the O'Reilly platform. That feedback loop fits the episode. Specific criticism is treated as part of making the argument usable, not as a nuisance after the fact.

As heard by us

Agent platforms are judged here by permissions, guardrails, and production risk.

The episode focuses on what AI agent platforms are really for once they leave the demo stage and enter an actual organization. It keeps coming back to the hard edges: what data the system can reach, what code it can run, which permissions it inherits, and where a person still…

Read the full review in PlayNext →

Why you'd press play

You want the part where agent platforms become a permissions and risk problem.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Thoughtworks Technology Podcast