“BDD, Planning Poker, refinements: it always comes down to talking to each other.” - Richard Seidl
I believe we are currently using AI for the wrong problems. Generating test cases, writing boilerplate code, auto-fixing broken tests. All nice, all efficient. And don’t get me wrong: I use these tools every day and would not want to give them up. But honestly, that is just our old job, done faster. There is that famous quote attributed to Henry Ford (probably not authentic, but it hits the mark): “If I had asked people what they wanted, they would have said faster horses.”
Do you remember the first iPhone? It displayed websites in miniature. A browser on a pretty device. It only got interesting when people stopped rebuilding the old. When things emerged that could not have existed before. Shazam, Uber, and so on.
In software development, we actually know this moment well. Assembler was replaced by high-level languages, frameworks came on top of the high-level languages, and each of these layers did the same thing: it devalued one skill and upgraded another. If you could juggle registers in your head, you used to be a hero. A few years later hardly anyone needed that; instead, we needed people who could design systems. Frameworks then devalued reinventing the wheel and upgraded assembling. Each time, a piece of craft was lost; each time, a different piece of thinking became more important. And each time there were colleagues who swore that without the old skill, nobody could seriously build software. Some time ago I asked here whether AI even needs our programming languages anymore. Today the other side of that question interests me: what does this new abstraction do to us?
Because that is exactly what AI is: the next abstraction layer. This time, execution itself moves to the machine. And which skill gets upgraded? Formulating intent precisely. What should the system do? For whom? Why at all? What must never happen? We have spent decades turning vague wishes into software. Requirements specifications, user stories, acceptance criteria: all attempts to nail down intent. Mostly mediocre. Again and again we tried different forms to make this transfer work. And what worked best? Talking to each other. Planning Poker = talking to each other. BDD = talking to each other. Refinements = talking to each other.
Finding a shared perspective. Whoever could do that could always build good software. Now it works without the detour. And it democratizes massively: anyone who can put an idea into words can build. Great and dangerous at the same time.
Lately I often hear a kind of career bet: as long as the AI hallucinates, they need me at the quality gate. I would not put too much on that. With grounding and verification the error rate drops, and betting against the learning curve of technology is a bet our industry has rarely won.
The quieter danger is a different one. If AI takes over the doing: where does the skill come from that we use to judge it? Judgment grows through doing, not through watching. If you outsource thinking too early, you unlearn exactly the ability that was supposed to make you irreplaceable.
One thing the machine cannot do, though, no matter how good it gets: want. It recombines its training space, brilliantly and tirelessly. But there is no goal, no interest, no stake in the result. The value of our work never lay in execution alone. It lies in the intent behind it.
And the right problems? Those are human problems. What is this good for? For whom? Should we build it at all? Who even needs all those generated apps? The exciting questions in my projects were never: how do I write these test cases? But: why do business and development keep talking past each other? Why is this requirement sitting in the backlog when nobody remembers who wanted it? Questions like these get decided between people, in meetings, at the coffee machine, in arguments about priorities. Not in the model. Quality is a social system, not a technical one. And AI amplifies what is already there: a team with shared understanding gets faster. A team full of misunderstandings, silos and distrust gets the same problems, just at a higher clock rate.
That is something we should do at some point: talk to each other.