From Coding to Supervision: The Developer as a Skilled Prompt Engineer and Guardian of Quality

With nine years of experience, Baptiste Fischer worked as a developer on a major e-commerce project, the very first agentic project at SQLI, and also took on the role of technical expert. He describes a profession in transition: writing less code by hand, taking on more of a supervisory role, learning to write precise prompts, and always keeping a human in the loop. While AI speeds everything up, it’s still the developer who is ultimately responsible for the quality of what gets delivered.

KEY TAKEAWAYS

• Value is shifting toward supervision. The developer writes precise prompts, reviews everything, and bridges the gap between the client’s needs and the technical solution.
• An ecosystem that encodes expertise. “Agents” and “skills” are specialized assistants integrated into the AI that ensure security, compliance (GDPR), and performance with every feature.
• Responsibility remains human. “I could never say there’s a bug because it’s the AI.” Systematic human code review before anything is delivered.
• Accelerated upskilling. A broader scope of responsibilities (including a period in the role of technical expert), with accelerated learning: “five months for what would have taken a year and a half.”

SQLI — How would you describe your role as a developer today, compared with before AI?

Baptiste Fischer — The biggest change is that we’re much more involved in discussions with the client. We have more time to analyze and go deeper into understanding the business: instead of receiving a raw ticket and executing it, we really dig into the request so that we can give the AI the right instructions. Based on our experience and what we’ve discussed with the client, we guide the AI toward the expected outcome. It brings us closer to a lead role: we bridge the gap between the business need and the technical solution.

SQLI — So where does your real value lie in this context?

Baptiste Fischer — Before, it was mainly about pure technical expertise: solving a task, turning it into a digital experience, quickly and at the right cost. Today, the value lies in bridging the gap between the request and the technology, translating it correctly, and above all, ensuring that we produce something of high quality. AI is an incredibly powerful engine, but it can only really harness its capabilities through us. My role is to write precise prompts and maintain control over what we deliver to the client. And because AI takes on a large part of the repetitive coding, I reinvest the time saved where I can be most useful: software architecture, compliance, and quality.

SQLI — With this acceleration, how do you ensure quality, security, and maintainability?

Baptiste Fischer — It starts as soon as we assign a task to the AI: we have to give it the right coding rules. But above all, we have an entire ecosystem, which is agents and skills, that encodes our experience of the project. A concrete example is the synchronisation of customer data flows on the website. Certain data needs to be encrypted, particularly for GDPR purposes, and there is a specific workflow that must be followed. So we create agents that systematically check security, compliance with coding standards, and performance for every feature or fix. When we generate a particular PHP class, the agent already knows what requirements to enforce. We’re the ones who provide that expertise because only a developer can tell the AI how to work without introducing regressions. And it means we don’t have to repeat ourselves with every request, so we’re less likely to forget something.

SQLI — Do you adjust these agents as you go?

Baptiste Fischer — Constantly. The project’s agentic configuration isn’t a fixed object: it’s a living asset that we refine week by week and that becomes increasingly relevant as the project progresses. In practice, we continuously enrich the rules, the patterns to follow, and the anti-patterns to avoid, just as a developer refines their working environment. Whenever we identified a weak point, often in the interface, where the AI was losing time or following the wrong instructions, we reworked our agents. We were updating ours again this very morning. It was SQLI’s first agentic project; since then, there have been many more in just a few months, and that’s a good thing.

SQLI — And where does responsibility lie now?

Baptiste Fischer — It remains entirely with us. I could never tell my project manager, “There’s a bug because it’s the AI.” The developer remains responsible: what we produce has to match what was requested, period. The machine executes, but it’s the human who must make sure nothing is overlooked and who is accountable for the result.

SQLI — It seems you’ve also expanded your scope of responsibilities?

Baptiste Fischer — Yes. I’ve been able to take on more responsibilities, sometimes beyond my usual role, for example, DevOps, meaning deployment automation. And I spent a month filling in for our technical expert, who was on paternity leave. Without AI, I wouldn’t have been able to take on that role with the same confidence, because it carries a lot of responsibility. More broadly, I feel I’ve learned in five months what would probably have taken me a year or a year and a half to learn otherwise.

SQLI — At this pace, how does it change teamwork?

Baptiste Fischer — We need to synchronise much more. Before, a feature would take two days, and we wouldn’t need to discuss it again after the morning meeting. Now it produces so much that we have two meetings a day, and I’m in touch with my project manager practically every hour. The speed forces us to communicate more, not to work in isolation, and to make sure everyone feels involved. That’s perhaps one of the positive aspects: we talk more. The trap would be trying to rush everything: AI produces quickly, with a good success rate, and you can find yourself thinking that a feature should be wrapped up in an hour. You have to resist that pressure. There’s still the review, testing, and expert validation.

SQLI — How do you continue to develop your skills? Did SQLI’s coaching help?

Baptiste Fischer — At the beginning, time was set aside for training: we were introduced to agentic delivery, and that was valuable. After that, a lot depends on the desire to lear. I review a huge amount of what the AI produces and constantly ask it, “Why?” That’s one of the tool’s strengths: AI is a fantastic way to learn because it often provides very good answers. As long as you use it to understand, not just to produce. For more junior profiles, it requires additional support, which is  essential. But with the right framework, it allows them to develop their technical skills much faster.

The Final Word

SQLI — Please sum up how you feel about this project and your role.

Baptiste Fischer — I feel like I’m in the right place. Honestly, I find it hard to imagine who other than a developer could have built this ecosystem of agents and skills, chosen the right technologies, and ensured the quality. What gives the role greater value is control over quality and what we deliver, and the connection between the client, the project manager, and the tool. AI is an incredibly powerful engine; we make sure it runs properly and that everything is right. The role is going to evolve significantly, and I find that more exciting than frightening.

After the project manager who became an orchestrator, we now have the developer who has become a supervisor, a precision prompt engineer, and a validator of AI-generated code. Two roles, one shared conviction: AI accelerates, but humans remain in control: the “human in the loop” that turns speed into quality. At SQLI, this is what agentic delivery looks like day to day: demanding, fast-paced, developmental, and firmly guided by our teams. Next episode: another role in the delivery chain, seen from the inside.