With AI becoming an integral part of every stage of the development cycle, the role of the developer is evolving rapidly. Automation, new trade-offs, a shift towards design skills rather than execution… the landscape is changing. To understand what is really changing in the day-to-day work of tech teams, we spoke to Karim Bourass, Lead Software Engineer at Berexia, who is observing this transformation from the inside and is already seeing its operational impacts.
In your opinion, what is the real impact of AI on the developer profession today?
“AI has not replaced the developer. It has shifted the value.
Everything that is boilerplate, scaffolding, format conversion, repetitive tests… yes, that is greatly accelerated. But understanding a vague business need, arbitrating between contradictory constraints, designing an architecture that lasts over time? That remains 100% human.
The real impact is this: the one who refuses AI loses a real competitive advantage. But the one who delegates everything without critical thinking produces technical debt at an unprecedented speed.”
What tasks does AI really help you accelerate in your daily life as a Tech Lead, and which ones do you refuse to delegate?
“Concretely, AI helps me enormously with:
- Scaffolding of backend/frontend modules (guards, interceptors, DTOs, basic structure, testing).
- Complex SQL queries, especially when working on multiple schemas with heavy aggregations.
- Writing technical documentation.
- Exploratory debugging when you come across a legacy system you don’t know.
What I refuse to delegate? Architectural decisions. API design between services. And the final code review (PR). AI does not understand your organizational context, client constraints, or the compromises you knowingly accept.”
Can you share a concrete example where AI has improved a delivery or unblocked a technical situation?
- On an ML scoring project, features needed to be extracted from five different schemas in the database. Complex joins, temporal aggregations, a lot of back and forth. Usually, that’s two to three days of pure SQL. Here, I described the business logic, the AI generated the SQL, I validated and adjusted. Completed in half a day.
- Another case: integrating a connector to a CRM API that I didn’t know in detail. AI accelerated the entire exploration phase, which allowed me to focus on robustness and error handling rather than reading documentation for hours.
How do you support developers (senior and junior) so that they use AI without becoming dependent?
“A junior who copies and pastes code generated by AI without seeking to understand what it does, he doesn’t progress, he accumulates technical ignorance without realizing it. AI is an accelerator, not a teacher. In PR review, I always ask the same question: explain to me what your code does. If you are not able to defend it, it has no business being in the codebase. No matter who wrote it.
For seniors, the trap is different but just as dangerous: it’s intellectual laziness. When you start letting AI think for you about the architecture or edge cases, you are regressing. And it’s insidious because the result looks clean.
Ultimately, the principle is the same for everyone: AI generates, you validate and you take responsibility.”
Does AI change the way you do code review or transmit best practices?
“Clearly yes. Before, a good part of the review focused on naming errors, structure, missing patterns. AI has reduced this noise. Today, the review focuses on the real subjects: consistency with the codebase, security implications, long-term maintainability. But there is a trap. AI-generated code is often “correct but generic.” It compiles, it passes basic tests, but it doesn’t necessarily respect your internal conventions or performance constraints. You have to be even more vigilant.”
For seniors, the trap is different but just as dangerous: it’s intellectual laziness. When you start letting AI think for you about the architecture or edge cases, you are regressing. And it’s insidious because the result looks clean.
Ultimately, the principle is the same for everyone: AI generates, you validate and you take responsibility.”
What risks do you observe when a team uses AI without a clear framework (quality, security, technical debt)?
“The first thing to avoid is pure vibe coding—throwing a prompt, getting a feature in one shot, and pushing without thinking. It works once, but at the scale of a project, it’s a ticking time bomb. Without a clear framework, you end up with three problems: invisible technical debt because the AI code looks clean but is verbose and inconsistent with your existing codebase. Data leaks because developers paste business code, credentials, or database schemas into prompts without asking questions. And an erosion of skills—a team that no longer knows how to write SQL, debug manually, or understand what the ORM is doing under the hood is a team that only stands as long as the AI covers its case. The day it doesn’t, there is no one to take over.”
For seniors, the trap is different but just as dangerous: it’s intellectual laziness. When you start letting AI think for you about the architecture or edge cases, you are regressing. And it’s insidious because the result looks clean.
Ultimately, the principle is the same for everyone: AI generates, you validate and you take responsibility.”
In your opinion, what human skills will remain essential for a developer in an environment where AI is omnipresent?
“For me, the most valuable skill tomorrow will not be writing code; it will be understanding a business need, often vague, and translating it into a clean technical architecture that can be implemented using AI tools. The 100% pure code part, we’ve moved past it. Using AI tools today is no longer a luxury; it’s an obligation. The one who doesn’t adopt them is putting themselves behind.”
