[Blog Series] Productionalizing Agents

[omrishiv]

We want to propose a blog series that the community can start working on together around building agents:

Part 1: What are agents/when are they appropriate. There is no shortage of literature on what agents are, but a comparison of agents to traditional systems/LLM augmented systems are fewer. As this is part of a series, we will dive into a specific use case of when agents are appropriate and compare it to traditional systems to highlight the benefits of moving to a more autonomous architecture. We can use the booklet on agentic architecture as further reading for this post. The first part will set the stage for building out an agentic system with the community's working group lens in mind: development, interoperability, operationalization, and security in mind.

Part 2: We will introduce the agentic platform the community is developing and use it to build out a production agentic system. We will dive into the components of the platform and why they are needed for large scale agent deployments. This will coincide with the release of the agentic platform deployment/document that can be used to build agentic deployments. We will go into the architectural decisions and why they were made as well as how to swap components based on preferences. How to take a simple example and move it from client to production.

Parallel Posts (building on the same use case):
Part 3.1: A deep dive into agent development. Architecture, meta agents, large vs small agents, when is it appropriate to use which?
Part 3.2: Interoperability components: agent mesh, MCP, A2A, AgenticGateway and why they're needed for agentic systems.
Part 3.3: Agent Evaluations: Moving from traditional software testing to context based testing. Agent metrics (tool use accuracy, context adherence, context engineering), token utilization, cost, and latency and how to understand performance given these competing metrics.
Part 3.4: Security: identity, governance and what to keep in mind when developing agents.

Part 4: Bringing it all together, a preview of future use cases we're developing and lessons learned as a community

[rr-paras-patel]

I can help on Part 3.2.

[omrishiv]

Thank you @rr-paras-patel. Let's coordinate with @ratnopamc to align on content here

[dineshdm57]

I can help on part 3.3