AI Requires that Architects & Product Owners Are The Same

So in the first few iterations of this dark software factory, the only agent that would interact with a human was the "Product Owner" agent. Once that agent received all the requirements, it would then hand off to an "Architect" agent who would attempt to find a solution to the problem.... And that is when the process started to break down. The architect would create new repos, lose context of how a new "feature" should fit in the current framework, and/or lack enough information to properly identify a solution that would match the requirements but push on regardless. Sometimes this was due to the requirements themselves being fantastical and not something that could be addressed by a technological solution. Having the architect agent run sequentially after the

This led to a revelations: for a dark software factory to work, the product owner and the architect have to be the same agent. These roles are usually considered seperate, but they need to be unified because they supply two different needs to a product and an architect and product owner need to reconcile with each other. Simply put:

  • A "Product Owner" is someone who can correctly identify opportunities
  • An "Architect" is someone who can identify solutions

If you can identify opportunities but have no idea how to achieve them or whether those opportunities are even possible, then you are stuck. Conversely, implementing "solutions" that don't actually address a given opportunity is also a failure mode that I have witnessed several times.

You need a product owner and an architect working in tandem as an single entity. That means the AI agent needs to be both a product owner and an architect...but that also means that users should also be reasonably versed in both product ownership and architecture themselves. so that they can validate both that an opportunity was correctly identified and that the proposed solution would actually address the opportunity.

Which means, that a software factory only works if used only by those with significant expertise in software development, architecture, product ownership, and business domain knowledge... AKA seniors.

It just feels like this is another observation that AI tools, rather than being a tide that lifts all boats, gives superpowers to those who were already experts while leaving more junior users in the dust. Trying to build out a software factory, which hypothetically was supposed to level the playing field and make it easier for non-technical users or technical users without business domain knowledge to bridge the gaps. Instead, it seems to have only exacerbated it. Or I could be wrong as these are the musings as I troubleshoot my own process inbetween builds :)

← All posts