The Agentic Architecture Framework: community-driven, governance-first architecture guidance for building safe, reliable, and scalable agentic AI systems. Adapted from the Cloud Well Architected Framework, the AAF aims to sit above orchestration frameworks and provide architectural principles and guidance to humans and agents alike. Contributions very welcome!
No reviews yetBe the first to leave a review for Agentic Architecture Framework
Maker
📌
I spent 10 years doing cloud solution architecture. AWS Well-Architected Framework, infrastructure governance, cost optimization at scale; that's the mental model I used to think about systems.
Then I started seeing teams build AI agents and it was chaos. Same problems cloud solved 15 years ago; no governance, no cost visibility, no clear boundaries around what the system can do autonomously, but nobody had translated those lessons to this space.
So I took what worked in cloud architecture and adapted it for agents.
Most teams are using LangChain, LangGraph, Claude SDK, orchestration tools are figured out. But the higher-level architectural thinking isn't. You're picking up pieces from blog posts, Discord threads, people's experience. Nobody's got a coherent framework.
This framework puts it in one place:
Epistemic gates: Where exactly does AI output become real action? What needs to gate it?
Cost: Not just cheaper models. Why agents cost what they cost. Context bloat is where most of the waste lives.
Autonomy: How much can this agent decide on its own without breaking things?
Context: Understanding what's actually needed vs. what you're paying for anyway.
I also built an MCP tool to start applying this: agenticaf.io/tools
Looking for maintainers, contributors, people to test the MCP tool and reports, folks willing to review the tools against real agent builds.
Feedback wanted. What breaks. What's missing?
happy building - Dave
Report
Maker
This is a seriously exciting launch, Dave. Agentic AI is moving fast, but the frameworks for building agents that are secure, reliable, and cost-aware haven’t kept pace. Love that the Agent AI Framework tackles those challenges at the architecture level rather than treating them as an afterthought. Looking forward to seeing how the community builds on this!
Having had the chance to contribute some input into this framework, it’s great to see it launch publicly.
One of the things I think Dave has got particularly right is treating agentic systems as an architecture problem, not simply an orchestration problem. As agents move from generating outputs to actually taking actions, questions around autonomy, governance, cost, context and epistemic boundaries become fundamental.
There’s a real gap between “we can build an agent” and “we can engineer an agentic system we’d actually trust in production.”
Artificial Intelligence use in business is the wild west right now. AAF is an important step towards giving us a shared architectural language for that gap.
Excited to see where the community takes it.
Report
Maker
@joe_griff that difference between architecture and orchestration is a nuance thats really key to understand. I can orchestrate a poorly architected agent...
@davem_0 It's probably the biggest problem with AI use right now. Tokens can get expensive if your ability to orchestrate is flawed. It should be grounded on a well architected framework....
This is a seriously exciting launch, Dave. Agentic AI is moving fast, but the frameworks for building agents that are secure, reliable, and cost-aware haven’t kept pace. Love that the Agent AI Framework tackles those challenges at the architecture level rather than treating them as an afterthought. Looking forward to seeing how the community builds on this!
@muhammad_hussain40 Thanks , the community aspect is the key here
Clarity: Caffeine and Focus
Having had the chance to contribute some input into this framework, it’s great to see it launch publicly.
One of the things I think Dave has got particularly right is treating agentic systems as an architecture problem, not simply an orchestration problem. As agents move from generating outputs to actually taking actions, questions around autonomy, governance, cost, context and epistemic boundaries become fundamental.
There’s a real gap between “we can build an agent” and “we can engineer an agentic system we’d actually trust in production.”
Artificial Intelligence use in business is the wild west right now. AAF is an important step towards giving us a shared architectural language for that gap.
Excited to see where the community takes it.
@joe_griff that difference between architecture and orchestration is a nuance thats really key to understand. I can orchestrate a poorly architected agent...
Clarity: Caffeine and Focus
@davem_0 It's probably the biggest problem with AI use right now. Tokens can get expensive if your ability to orchestrate is flawed. It should be grounded on a well architected framework....