Agentic Product Development: Governance at the Speed of AI

Does that mean that you have to spend more time as the product owner and the human in the loop really refining the context and the goal in a more effective way to allow them to really be able to say, is this still valid, this assumption or these sets of assumptions?

Yeah, I've got to be more explicit around when we get to setting a sprint.

I've got to be more explicit about the sprint goal and the relationships between the items in the backlog and why these items are going in the sprint.

And that took me a while to learn how to actually specify.

But part of sprint planning as we go through it is, again, some governance files in order to be able to help them to provide that kind of structure.

That's also asserted through a hierarchy of files.

that become basically an extended set of the definition of done.

So rather than, for example, going that the task is supreme and I'm going to do everything to execute the task and bypass all of the rules, there is an explicit hierarchy around what they can't touch and what is flexible.

And so that as well with the sprint goal then helps to create basically this bounded authority about where their work starts, where it finishes, and importantly, then what decisions that I should be making.

So yes, being more explicit, we've worked through that after, and that's taken something that's probably about six months worth of work with Raptors to work out what works in terms of being more explicit.

and working through then how they then interrogate those things without violating the definition of done, for example.

How they move through the escalation paths of single, double, or triple-loop learning, and at what point should I actually be involved, as opposed to them going, oh, there's a problem, that's okay, Matt will fix it.

Or, oh, I've done some copy, and I'm not really happy about this particular part, so we'll all stop.

And we'll escalate it to Matt.

But importantly, we won't use Al.

We'll just write it in the JIRA file.

And then we'll just sit back and have a virtual coffee and wait for him to find it.

So all of the normal social aspects that you've got with a human team, where humans can, through just simple social interaction, kind of keep tab on each other.

And product owner can walk through or over the cubicle or even pop up in a chat and go, oh, yeah, how things are going.

None of that the Raptors can do because they're just agents.

So being more explicit helps me as a product donor know where I need to focus.

And now I've finally got to the point of going, great, as the human, the team, it's called the irreversibility boundary.

What's the last sensible moment for me to act? And typically it's publishing or it's releasing software.

It's a great, do your work, and then I will look.

Sometimes throughout the sprint, sometimes I'll wait to sprint review.