Skip to Main Content
Talk Intermediate MIT License

Agents as Backend: Reimagining Backend Architecture

Proposal status is Rejected
Session Description

We’ve spent years building backends around the same ideas: APIs, controllers, middleware, serializers, versioning, and request-response pipelines. But as software starts interacting through agents instead of fixed user flows, backends themselves become increasingly stateful — continuously managing context, memory, intent, and long-running execution.

This talk explores a different way to think about backend systems: treating the backend itself as a network of agents and capabilities rather than a collection of endpoints and services. Instead of manually wiring routes to business logic, requests are dynamically routed based on intent, context is progressively hydrated as needed, and specialized agents coordinate execution while results stream back to the frontend in real time.

The session will cover how these systems work in practice: capability-based architectures, intent routing, contextual memory layers, multi-agent orchestration, streaming communication, and protocol layers such as AG-UI, MCP, and A2A. We’ll also look at how familiar backend concerns like observability, security, caching, governance, and scalability evolve in agent-native architectures.

Along the way, I’ll demo a fully working agent-native backend running on open-source infrastructure with interchangeable models, showing real-time routing, multi-agent execution, and streaming interactions end to end. The goal is to give attendees a practical view into where backend architecture is heading as language models become part of the execution layer itself.

Key Takeaways

  • A new mental model for backend architecture where capabilities and agents replace traditional endpoint-driven systems.

  • How state, execution, and coordination move deeper into the backend layer as applications become increasingly agent-native.

  • The emerging role of interoperable protocols and event-driven communication in modern AI systems.

  • Practical insights into building modular, extensible, and model-agnostic backend infrastructure for the next generation of applications.

References

https://agentend.ai/
https://gok03.com/slm-talk/

Session Categories

Technology architecture
Engineering practice - productivity, debugging
Talk License: MIT License

Which track are you applying for?

Main track

Speakers

Gokulavasan Murali
Gokulavasan Murali Head of Engineering | Emma Robots Inc.

Gokul is a product engineer and architect with over a decade of experience developing autonomous systems, specialising in the convergence of computer vision and agentic AI. He currently serves as Head of Engineering at Emma Robots Inc., where he builds Vision-First RPA and Computer-Using Agent (CUA) systems that perceive and operate software interfaces the way a human operator would.

Previously, as Head of Engineering at Plena (acquired by Chief), he architected AI sales agents that performed autonomous multiplatform outreach. His technical roots trace back to the mobile edge: he co-founded Flytta and shipped some of the first mobile-native deep learning models for logistics and driver safety. His career began in embodied robotics, where he led teams to design semi-autonomous navigation systems on embedded compute platforms and to build autonomous underwater vehicles.

He is a long-time member of the open-source and Python communities - a speaker, builder, and PyCon India organiser with a particular interest in agentic AI and efficient on-device intelligence.

https://www.linkedin.com/in/gok03/