AI Can Write Code, But 22 Experts Say Humans Must Still Decide

AI tools can now do a lot of the work that software engineers used to do by hand — writing code, suggesting designs, and even fixing bugs. So a group of 22 software professionals gathered at a European conference in 2026 to answer one nervous question: what's left for humans? Their finding was surprisingly clear. The big-picture decisions — what to build, how it should be structured, and who gets blamed if it fails — remain human jobs. Think of it like construction: robots can lay bricks, but a human architect still signs the blueprint and carries the liability.

The group also coined a term that should worry any manager: 'cognitive debt.' It's the slow loss of understanding that happens when teams accept AI-generated work without really thinking it through. Like a driver who lets the car steer for years and forgets how to handle a sharp turn, companies can end up with software nobody internally truly understands. That's dangerous when something goes wrong at 2 a.m. and the AI isn't around to explain itself.

Their recommended fix is something they call 'harness engineering' — basically, building your own company-specific rules, knowledge, and safety checks around the AI, rather than trusting whatever the tool spits out. They also offered a simple rule of thumb for when a human must step in: 'criticality,' which means how uncertain the situation is combined with how expensive a mistake would be. Low stakes, low uncertainty? Let the AI run. High stakes or high confusion? Get a person in the loop.

Worth noting: this is a small focus-group report, not a large study, so it reflects the opinions of 22 people rather than hard data. Still, it's a useful heads-up. If your company is racing to use AI for technical work, the takeaway is to keep the judgment, the accountability, and the documentation firmly in human hands — and to make sure someone on your team can still explain how the thing actually works.