“You have an agent that's working fine in production and starts to drift a little bit. Why does it start to drift? Because it's grounded on some data on refund policies that just got changed. But somebody wrote it in a way that it's not very clear. Somebody updated the document, the agent's not working. … you're sending refund checks that are bigger than the amount that was needed.”
Product Growth · Product & Design · November 2026
Context
Asked to define the agent development lifecycle versus the usual software lifecycle, he gave a concrete example of why a live agent needs monitoring. The goal, he says, is an agent that can detect what context changed and correct itself.
Transcript
…that you're building or testing the system that you're testing? Can you set a bar on what good looks like? Because that's what translates into evals. And then, like you said, I think access to all the prototyping tools, which I think is the easy bit now. I think that pretty much everyone gets.
…pretty easy. The roles, we've given you some more clarity on. I think that third layer now is the life cycle that all of this is under. And you created this new term, ADLC. Yeah. Not the SDLC that we all know. So can you define those two terms for us and explain to us what makes the ADLC different?
…do you deploy it? And then having deployed it, how do you observe it so that it continually improves? So these five elements are taken together is what I call the ADLC lifecycle. The Holy Grail, of course, is an agent that you've built and it's in production and can self-heal to some extent. And I'm not talking some fancy rocket science sci-fi stuff on self-healing. I mean simple things. Like you have an agent that's working fine in production and starts to drift a little bit. Why does it start to drift? Because it's grounded on some data on refund policies that just got changed. But somebody wrote it in a way that it's not very clear. Somebody updated the document, the agent's not working. The refund, or you're sending refund checks that are bigger than the amount that was needed. Can it self-heal? It can. It needs to detect what context was changed, whether that context was changed the right way or wrong way. That's the knob that you want to get to. You want to get to a point where if you've designed this right, there are two loops. There's the inner loop, which is you conjuring…
Does the ADLC only apply then if you're building agents?
I think it's broader than that. I think it has to do with an AI development lifecycle, which has an eval mindset baked in, which understands that anything built on data can drift because data can drift, because data itself drifts. And then having the right tool sets to observe that drift in…
So it sounds like part of what is involved in the ADLC is moving a lot faster. If you were on a six-month ship cycle, it sounds like you need to almost go to a continuous deployment. It can self-heal whenever it wants cycle.
More from Product Growth
