I am thinking of an AI-First Living Architecture Governance Framework where the architecture documentation is no more static but is a living thing used by AI with dual purpose leading the low level design and development and acting as a guardrails maybe integrating directly into development loops and CI/CD pipelines as automated guardrails to prevent code drift and enforce consistency across the entire system.
We have a lot of tools out there: arc42 document template, c4 model, ADR, ArchUnit, Spec Driven Development, BMAD, Agent skills and more.
I made a research but I haven't found a n architecture Governance Framework that put the available tool all together to get most out of them, at least in ny humble opinion.
In short, I'm using arc42 document template for architecture documents, ADR every time there is a decision to be made, for both I drvelopoed AI agents and skills to support me in architectural work, arc42 agent suggest me when I should create an ADR.
But I still need to work to the link between the architectural stuff and the development.
That's the crux of it — and I think the problem is that we ask one artifact to do two jobs. arc42, C4 and ADRs are intent: context for humans and now for agents. Guardrails are executable: ArchUnit rules, dependency checks, contract tests in CI. They're what holds when nobody is reading.
So the missing link isn't another document, it's a translation step. What helps is adding one question to the end of every ADR: how would we know if this decision was violated? If the answer is a test, it goes in the pipeline. If no test is possible, it's a guideline rather than a guardrail — and we should expect drift.
Docs guide the agent. Tests constrain it. We need both.
In my experience, the way you perceive code changes when you have to manage the design of the entire system, trying to find a compromise between high-level aspects, including system architecture and infrastructure, technology choices, costs, and more. It's crucial to lead multiple teams so that everyone is aligned on your architectural vision, to achieve the desired consistency.
It's therefore very difficult to keep up with the programming details that every developer deals with on a daily basis, yet, as an architect, you have to find an effective way to do so. In my opinion, investing in the development of AI tools that help ensure consistency is very valuable.
This is where we get the disconnect. Someone might be a greater developer, then “move up“ to management, and not do “lower level“ activities (like coding) anymore, but spend their time with Excel sheets, PPT, meetings… that’s how we end up losing talent…
I am thinking of an AI-First Living Architecture Governance Framework where the architecture documentation is no more static but is a living thing used by AI with dual purpose leading the low level design and development and acting as a guardrails maybe integrating directly into development loops and CI/CD pipelines as automated guardrails to prevent code drift and enforce consistency across the entire system.
We have a lot of tools out there: arc42 document template, c4 model, ADR, ArchUnit, Spec Driven Development, BMAD, Agent skills and more.
I made a research but I haven't found a n architecture Governance Framework that put the available tool all together to get most out of them, at least in ny humble opinion.
What’s your current approach?
In short, I'm using arc42 document template for architecture documents, ADR every time there is a decision to be made, for both I drvelopoed AI agents and skills to support me in architectural work, arc42 agent suggest me when I should create an ADR.
But I still need to work to the link between the architectural stuff and the development.
That's the crux of it — and I think the problem is that we ask one artifact to do two jobs. arc42, C4 and ADRs are intent: context for humans and now for agents. Guardrails are executable: ArchUnit rules, dependency checks, contract tests in CI. They're what holds when nobody is reading.
So the missing link isn't another document, it's a translation step. What helps is adding one question to the end of every ADR: how would we know if this decision was violated? If the answer is a test, it goes in the pipeline. If no test is possible, it's a guideline rather than a guardrail — and we should expect drift.
Docs guide the agent. Tests constrain it. We need both.
The worst architects are the ones who stop being developers.
Yes, they get out of touch with the pains of maintaining an ivory tower architecture.
In my experience, the way you perceive code changes when you have to manage the design of the entire system, trying to find a compromise between high-level aspects, including system architecture and infrastructure, technology choices, costs, and more. It's crucial to lead multiple teams so that everyone is aligned on your architectural vision, to achieve the desired consistency.
It's therefore very difficult to keep up with the programming details that every developer deals with on a daily basis, yet, as an architect, you have to find an effective way to do so. In my opinion, investing in the development of AI tools that help ensure consistency is very valuable.
That’s a crucial point - leading multiple teams, to avoid inconsistencies in architectural vision.
As you mentioned, AI is a game-changer. Now our job is no longer coding, but rather specifying requirements and architecture to AI…
What do you see as the direction of architects & AI?
I never understood why some architects and technical managers view coding as some sort of "peasant" activity.
It's an activity that requires high skills and good cognitive ability. I'm proud of my coding skills and I still do a lot of it.
I can't imagine myself ever thinking it's "beneath" me.
This is where we get the disconnect. Someone might be a greater developer, then “move up“ to management, and not do “lower level“ activities (like coding) anymore, but spend their time with Excel sheets, PPT, meetings… that’s how we end up losing talent…