How the Use of AI Teammates Became Business as Usual and What Have We Learned in Their Usage?

We've now been working with our AI Team Mate, Azpertilo since June, so what have we learned?

Treat your AI agents like new hires, onboard them, guide them, and manage them using the exact same principles you use for your human team.

First of all ground them.  The more data, you give them the harder they work for you.  Just like humans if you had a new starter, you wouldn't let them work on the “training data”.  The data they were given before they joined you. There would be an induction process where you would explain to them the story of the company, the story of the founders, your convictions, your guidelines. The same for humans as for AI.

Every day reinforces what I’ve suspected from the start: AI agents need real supervision. That doesn't mean reinventing management from scratch; it means applying proven management tools and processes, tailored for machines.  That's why we built Azpertilo so there is proper application supervision and rigor. We do it so you don't have to. 

Here are the three core principles we rely on at Azpertilo:

👉 The Briefing: The make-or-break alignment phase. This is an active, two-way work session between professionals. The machine and I ask the critical questions needed to uncover blind spots and clarify requirements. When properly executed, this collaborative phase provides the single biggest force multiplier for productivity.

👉 The Plan: What we used to call detailed technical specifications, architecture choices, task breakdown, dependencies. While this work is fully delegated to sub-agents, speed is usually their top priority. They will try to take shortcuts, especially on complex tasks. Unchecked speed leads to tech debt, so maintaining strict human oversight and approval here is vital.

👉 The Debrief: I’ve run debriefs with human teams for years, but I made the mistake of skipping them with my AI agents early on. It cost us valuable time. Today, every session ends with a clear retrospective: what got built, where things stalled, and how we solved it. Just like an Agile sprint retrospective, the team documents key takeaways to ensure continuous learning.

As we’ve refined this workflow, our team's throughput has skyrocketed.

Over to you: What’s the most surprising or valuable lesson you’ve learned from debriefing an AI's output, good or bad? Drop it in the comments below!

Based on a blog written by Bertrand Godillot, CTO at Azpertilo.AI