Collborative Agentic AI strategy

[lakruzz]

As described in Collaborative Agents, setting up Agentic AI support for a purely spec driven approach implies a lot of complexity. Not to mention adoption to specific languages, tool stacks etc.

Investigate, what MoM principles should be applied. Why and How?

Make this easy instead of hard!

[blikest]

Details mentioned:

• "Componentize" codebases (disambiguation: GNU KISS, modularity? package?)
• Reproducible, declarative development environments with Docker
• Trainable workflow for artificial delivery (CICD -> AICD? jk.)

"Collaborative" may use disambiguation too

[keozx]

I read your article a few days back and I figured this topic is a great fit for MoM which I would love to dig deeper.
I have a few highlights that resonated with me:

“AI agents should work against the same expectations that the system will enforce later” just like Developers
"the ugly truth is often that the team is not yet mature enough to utilize them, because of how the development process is organized."
"Instead of writing different kinds of tests, you write tests that express behavior, and then you control what they run against."
"The simple point to make here is really: When you build a mature flow, the AI is suddenly not immature anymore."

This is what I deal with daily in this new world. I have been pushing for maturing the development process and teams with modern software engineering practices. I'm convinced that people in the desert are the majority still, and I think we are forest dwellers (I hope), as Kent Beck would put it.

Likewise, I think related to this I also wrote something along these lines: AI will further increase the gap between teams with strong engineering discipline and teams that rely mostly on isolated code production (or no collaborative engineering), as I elaborate in my article

So I believe this is a major problem that all companies are facing, and it will have a bigger impact on their production costs as time goes on and prices rise; perhaps, as I said in the meeting, this is a pain that companies cannot afford, at least not the majority; it is not sustainable to keep up with non-mature AI workflows because the engineering workflows are not mature from the beginning. It's not about the best model or the best spec-driven framework; it's about how sustainable companies can work with AI, working with what we know works, AI or not.