Architecting AI Systems for the Messy Reality of Enterprises: Why Agentic Compute is the Missing Layer

code red code red

Enterprise AI doesn’t fail because of models or frameworks , it fails because of architecture.
In this talk, we’ll explore how agentic compute, a layer for orchestrating and building compound AI systems, solves the integration, scalability, and adaptability challenges that plague enterprise AI deployments.

Drawing on real-world experiences from  multi-country Agentic PaaS deployments

  • Why retrofitting green field compound AI Systems crumble in the face of heterogeneous data, political resistance, and changing business requirements.
  • The personas that need to be enabled by the architecture and why systems alone is not enough
  • The agentic compute architecture has core layers, lifecycle management, and fault isolation patterns. While enabling the existing teams within an enterprise
  • Open source components you can use to bootstrap your own agentic systems.

Whether you’re dealing with legacy systems, multinational data sources, or complex organizational politics, you’ll leave with architecture patterns and a reference design to build AI that actually works in the messy real world.

Interview:

What is your session about, and why is it important for senior software developers?

It’s about shifting from cobbling AI tools together to treating AI as a real computing platform. Most teams are forced today to build piecemeal: a model here, a tool there, brittle integrations everywhere. That approach collapses in messy enterprises. What’s needed is a foundation , a system view ,  that makes AI reliable, scalable, and usable in the real world.

Why should attendees prioritize your session?

Because you’ll leave with patterns for building AI systems that don’t collapse under enterprise complexity. This isn’t about demos. It’s about how to think in terms of platforms instead of patchwork.

What are the common challenges developers and architects face in this area?

The tooling is exploding. Every week there’s a new model, framework, or library. Teams try to stitch them together, but end up with brittle, piecemeal systems buried in complexity. What we need isn’t more tools it’s less. A platform view of computing, not a toolbox view.

What's one thing you hope attendees will implement immediately after your talk?

Stop thinking about “adding AI features.” Start designing around an agentic compute layer ,  a foundation that turns models into systems that last.

What makes InfoQ Dev Summit stand out as a conference for senior software professionals?

It’s where builders who’ve carried systems into production meet. Less hype, more scars and patterns. That’s what makes it valuable.

What does being part of InfoQ Dev Summit mean to you?

A chance to share what we learned building one of the first multi-country agentic platforms , and to help others avoid the dead ends of piecemeal AI.


Speaker

Arun Joseph

Founder @Rhizome Foundry GmbH | Co-Founder @Masaic AI (AgC™, Agentic Compute) | Envisioned & Lead @Eclipse LMOS | Former Head of AI Engineering, Deutsche Telekom AG

Arun Joseph is the co-founder & CEO of Masaic, a stealth startup building AGC , the Agentic Compute: a new computing substrate for knowledge work. AGC is open by design and built for scale, providing the architectural foundation for decisioning and actioning systems that integrate with existing enterprise stacks.

Previously, he headed engineering at Deutsche Telekom AG’s Central AI program,AICC,  where he envisioned & led one of Europe’s first agentic PaaS platforms, deployed at multi-country scale and now part of the Eclipse Foundation. Known as Eclipse LMOS, the platform demonstrated how empowering existing teams could produce best in class, cost-efficient systems that made work meaningful and accelerated results delivery.

He also founded Rhizome Foundry GmBH, a builder’s movement and foundry for agentic computing companies and systems, shaping the principles and infrastructure of this emerging paradigm.

Read more
Find Arun Joseph at: