From project manager to AI-augmented delivery orchestrator

At SQLI, agentic delivery is no longer a side promise: it's an operational reality, already proven on customer projects. To prove it, we give the floor to those who practice it on a daily basis. First portrait of a series: Davy Ducrocq, project manager, former developer, embarked on a major e-commerce project carried out with the Agentic Delivery approach. His job has changed in nature: from project manager, he becomes an orchestrator of AI-augmented delivery. Tickets generated in series, 80% automated tests, days cut in half to keep up... He tells us, without filter, what AI has really changed in his role — and what it does not replace.

AT A GLANCE — TAKEAWAYS 

• A real and measured acceleration. Creating tickets, specs, tests, dashboards: tasks that used to take days are now done in hours. "I don't create a single ticket by hand anymore."
• The human remains in control. AI handles 80% of the tests; the project manager validates the 20% most sensitive. Management, arbitration and responsibility remain human.
• Context makes quality. Feeding the agent with the right code and rules has become the key gesture of delivery — exactly what the approach intends.
• A lucid method, not magic. Davy also shares the limitations observed and the team adjustments: it is this honesty that makes the approach controlled and reproducible.

SQLI — On a daily basis, how would you describe your role today, compared to before? 

Davy Ducrocq — It's become a little more technical. In the morning, I get the up-to-date code from the developers so that the agent can rely on it when I ask him. Then I use it a lot: exports from Jira (our task tracking tool), Excel files, document templates, and especially the creation of tickets — units of work — I no longer create a single ticket by hand. The same goes for the specifications: we talk to the AI, it asks us questions, we answer, and in the end we push the specification (the "spec") directly into Confluence, our shared documentation space, without rewriting it. Transforming a spec into user stories and then into tickets was done by hand before and it took a lot of time: the gain is phenomenal.

SQLI — What has changed most profoundly in delivery? 

Davy Ducrocq — Definitely the testing phase. On a ticket that I used to spend a whole day on, the AI now does 80% of the work. I test the most specific 20% by hand, the ones that really require business judgment. Anything that is fast and systematic, the AI does it very well, and I don't need to go back to the top. It's a real gain.

SQLI — So there is no longer a human review?

Davy Ducrocq — Yes, there still is — but much less than before. Let's be honest: on this project, we started without a dedicated tester, with just the AI and me. When we give the feedback, we'll see that it's probably not ideal and that a tester is still needed — not necessarily to do manual testing, but to drive the AI well, to write good test scenarios. It was a first try: not perfect, but it allowed us to test a lot of things and move quickly.

SQLI — It is said that the more AI accelerates, the more critical coordination becomes. Is that how you experience it?

Davy Ducrocq — Yes! In concrete terms, instead of just one "daily" in the morning, we do two: one in the early morning, one in the early afternoon. Because in a single morning, a lot of things now happen. It allows you to coordinate and react much faster. It's sporty, it's sustained, but I have a team that immediately embraced AI. And since I'm more of a techie, I'm really into it.

SQLI — At this fast pace, how is the team holding up over the long term? 

Davy Ducrocq —Pretty well, because everyone bought in from the start. It's precisely to preserve everyone's attention that we divide the day in two: we take breaks, we don't engage with the AI nonstop. And it was the team itself that calibrated its rhythm — we reduced the number of daily updates when we felt it was too much. Above all, we've never been so united or so interactive: we talk to each other a lot more, we are constantly helping each other. And I, as a project manager, really find myself at the center of all these interactions — between the design, the BA, the technical expert and the developers.

SQLI — What does a typical day look like?

Davy Ducrocq — I get up, and pull the latest code so I have it on my machine. Then, team stand-up: who's doing what, how yesterday went. Then I set the AI running on writing specs, user stories, or tests — and while it's running, I go to my meetings, talk with the client. While we're talking, I've got an agent running that's doing tests on the site. There's no downtime.

SQLI — You also write specs — isn't that the business analyst's role?

Davy Ducrocq — Yes, and it will clearly stay their job. But on this project, with a very ambitious schedule, we needed reinforcements: the business analyst — the "BA," responsible for translating business needs into specifications — wouldn't have met the deadlines alone. I also wanted to test what AI could do — I started with a basic request, "I know nothing about e-commerce, create a spec for me for managing a gift card": the AI built an outline, asked me questions, and produced the final spec. That's telling. Since the team moves faster in development, specs need to be fed in much faster too. Where one BA used to be enough, tomorrow it might take two.

SQLI — What distinguishes, in your view, a project manager from a true AI orchestrator?

Davy Ducrocq — Context. The most important thing is giving the AI good context: the more accurate it is, the more accurately it addresses the need. That's why I pull the code every morning — when I ask for a spec on the gift card, it analyzes the existing code to propose an integration that doesn't break anything. My background as a former developer helps a lot. Not every project manager will go dig into the developers' code every morning, but it's become hugely important. AI will probably push some PMs to become a bit more technical.

SQLI — And the client relationship, is that changing?

Davy Ducrocq — The relationship stays the same, and the client knew perfectly well that we were using AI — we didn't do anything behind their back. What's changing is the pace: since we're producing more specs faster, the client needs to be more responsive, especially during their testing phase. You need a client who's even more involved than usual. On the business-testing side, I actually think human validation from the client remains essential: they're the ones who know their business rules.

SQLI — How did you train for this? Coaching, formal training, on the job?

Davy Ducrocq — A two-hour micro-training session for developers at kickoff, where the essentials of using AI were explained. After that, a lot of it happened on the job: learning to prompt comes with time, gradually. It's largely self-taught, with support along the way.

A final word

SQLI — If you had to sum up what AI is changing in your job?

Davy Ducrocq — The role is really evolving, with a constant sense of acceleration: I barely notice my days going by anymore. It can be a bit scary, but at the same time — being someone who's very active — I love it: I kick off one topic, move on to another, things keep moving, there's no downtime. You even have to relearn how to force yourself to take breaks, which is also why we break the day into chunks. AI is a real accelerator; it helps orchestrate things better and even fill in the gaps — suggesting the prerequisites to ask the client for before a production release, for example. It's truly an assistant.

Agentic delivery at SQLI is no laboratory concept. It's an everyday reality: demanding, fast-paced, exciting — and mastered by teams who know its levers as well as its limits. Davy is the first in a series of profiles. Next episode: the floor goes to a developer, to see how, from the other side of the keyboard, the profession is reinventing itself too.