AI Autonomy Is Redefining Architecture: Boundaries Now Matter Most

March 4, 2026·52 min·3 clips

Jesper Logren says generative AI needs boundaries, not runtime control.

1. The InfoQ Podcast episode "AI Autonomy Is Redefining Architecture: Boundaries Now Matter Most" centers on the Next Generation Architecture Playbook and the shift to the AI era. 2. The hosts bring back Jesper Logren, Enterprise Architect Lead at DXC Technologies and author of Design or Be Designed, after referencing an earlier conversation with Grady Booch. 3. The episode asks how architecture changes when generative AI, autonomy, and multi-agent systems enter existing software platforms. 4. Jesper says he has spent the last two years almost obsessively focused on generative AI, businesses, people, and processes. 5. Jesper describes his DXC role as 100% focused on generative AI, frameworks, models, proof of concepts, and customer experiments. 6. Jesper argues that the real shift is not tools but autonomy, because autonomous systems behave differently from robotic process automation. 7. Jesper uses the example of digitizing paper sales orders to show the difference between partial automation and end-to-end digital workflows. 8. Jesper says trying to embed generative AI inside procedural constructs creates the costs without the benefits. 9. Jesper connects the failure mode to a 2025 MIT report that he says found 95 percent of proof-of-concepts failing. 10. Jesper says the architectural answer is to understand the boundary around the agent instead of controlling logic at runtime. 11. Jesper compares an agent to "a gene in a bottle" and says the boundary needs tightly controlled names, holes, and interfaces. 12. Jesper says he has identified seven things that define an agent boundary and claims they can contain an agent with high confidence. 13. Jesper says the more agents you have, the more control of the boundary you need because emergent behavior increases. 14. Jesper says governance, design, and architecture must be "joined at the hips" and designed at the same time. 15. Jesper describes maturity levels from ad hoc to repeatable and says level 3 multi-agent systems need a new operating model. 16. Jesper says authority and decision rights are a guardrail, along with scope, goals, policy, risk, semantics, and evidence. 17. Jesper describes a design workshop for a listed Australian company where an LLM produced a 27-agent end-to-end process. 18. Jesper says the team tested edge cases, found one around national policies and buyer protection, and expanded the design to 33 agents. 19. The conversation is practical and interview-driven, with long technical explanations, analogies like the merry-go-round, and repeated returns to boundary design. 20. Architects, platform leaders, and enterprise teams working on AI governance would get the most from this episode, while listeners wanting light AI trends may skip it.

As heard by us

A practical architecture discussion on agent boundaries, guardrails, and keeping AI autonomy under deliberate control.

AI autonomy is framed less as a product feature than as an architecture problem: where boundaries sit, what agents are allowed to do, and how LLM-driven systems connect to existing procedural, microservices-based platforms.

Read the full review in PlayNext →

Why you'd press play

When your AI layer needs guardrails before it gets a vote.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The InfoQ Podcast