Thoughtworks Technology Podcast Exploring AI agent platforms
114 snips
Jan 22, 2026 Ben O'Mahony, a principal engineer at ThoughtWorks and co-author of the upcoming book on AI agent platforms, joins Fabian Nonnenmacher, a software engineer with a rich Java background turned AI practitioner. They dive into the crucial role of AI agent platforms in future work, discussing definitions of agents and their characteristics. The duo emphasizes extending existing platforms, the importance of observability, and security measures for autonomous agents. They also explore prototyping and the merging of AI engineering roles, offering insights into the evolving tech landscape.
AI Snips
Chapters
Books
Transcript
Episode notes
Observability Needs Per-Interaction Tracing
- Observability must evolve for generative AI because individual traces matter more in conversational flows.
- Ben O'Mahony says extracted interaction datasets become separate data products to manage and prune.
Trace Agent Control Flow
- Use tracing to reveal invisible control flow when agents decide execution paths.
- Fabian Nonnenmacher recommends tracing during development, evaluation, and production for agentic apps.
Design Autonomy With Guardrails
- Treat autonomy with deliberate design: evaluate goals, add guardrails, and avoid naive openness to the internet.
- Ben O'Mahony urges zero-trust permissions and human checkpoints for high-risk actions like code merges.
