Why our developers still need to be engineers, not just implementers
Short answer: AI coding agents make our developers much faster, and we use them every day. But speed only holds if a human still reviews the logic, can read the code without help and understands the fundamentals underneath it. That is the difference between an implementer and an engineer, and it is what clients pay for when something breaks.
Most of our developers now use Claude Code day to day. They pick up a ticket, give the agent the ticket and the codebase context, and it can build the whole ticket in one go. It works. It is fast. And that is exactly what I keep an eye on.
Where review quietly disappears
When an agent builds a whole ticket at once, the easy path is to let another agent review it. The output looks clean, the AI review passes, and the change ships. AI writes the code, AI reviews the code, and fewer human eyes sit in between.
It gets harder as the output grows. When a single generation runs to thousands of lines, no person can review it properly in the time a sprint allows. That is where the drift creeps in. Not in the code that fails, but in the code that passes and nobody really read.
Where the human belongs
A human has to stay in the loop at both ends.
- On generation: work in smaller pieces, so the developer can review each one before it moves on
- On review: do one pass with AI, ask it to flag the important logical parts, then review those parts yourself
Here is the hard question I ask myself. If a developer only passes tickets to an agent and approves whatever comes back, why hire a person at all? I could run two or three agents, one to pull the ticket and generate and one to review, and keep adding skills to them. The human has to bring what the agents cannot.
What "stay human" means
I do not want my team typing out boilerplate. Nobody writes every line by hand anymore, and they should not. But if AI generates everything, you slowly forget the syntax. Then you lose the ability to read what was written.
AI is very good at following documentation. If a Redis setup is broken, an agent can read the Redis docs and find where it went wrong. The trouble starts when the bug is a logical flaw that only the product owner understands, and the agent has already failed to fix it after several tries. At that point the developer is the last line of defence. If their reading has gone rusty, they cannot even see where the logic breaks.
Keep the muscle working
So we keep the hands-on skill in practice. Short, low-stakes, build-it-by-hand sessions with no agents. They take very little time away from client work, and they keep syntax and reasoning sharp for the day they are needed.
What I tell junior developers
A junior who leans on AI from day one can become a good implementer. Becoming an engineer, or an innovator, takes more. So I ask them to learn it properly:
- Object-oriented programming and how syntax actually works
- Why a language was invented and what problem it solved
- The core ideas behind compilers and interpreters
- The fundamentals beneath the frameworks
Learning the hard way makes you sharper, and that edge shows up in a crisis. Everything feels smooth while you keep implementing, right up until something breaks.
The fitness analogy
There are two ways to get a strong-looking body. One is the shortcut: supplements, the best gym, every piece of gear, fast results. The other is slow: clean food, running, proper training. From the outside, both look the same.
The difference shows when it matters. When real strength is needed, the body built the slow way holds up. Take away the shortcuts and the other one struggles.
AI is the shortcut, and a very good one. You can build a polished product fast, and the agents get stronger every month. But on the day the agents are down, or simply wrong, you are no longer an implementer. That day, you have to be an engineer.
The only question is whether you can
How this shapes the way we build
We are an AI-native studio. AI is part of how we design, build and test every product, including our own platforms, Mayavi and SCAI. That is exactly why we hold the line on human review and fundamentals. Clients get the speed of AI, with engineers who can still read, question and fix every line that ships.
See how we build AI systems or talk to us about your product.
