We believe the truth is close to the opposite. Experts matter more now, precisely because the volume of plausible, competent, forgettable, sloppy, work has gone up massively.
Mediocrity is the objective - not a failure
Generative systems are trained to find the optimal answer which is the one that would satisfy the 99.99%. That is what the objective function rewards in genAI training. It is also turns out to be a fair definition of mediocre. We believe ingenuity lives in the remaining 0.01 per cent, in the miniscule that is yet unexplored. A model has no route there mainly because the training procedure that makes it reliable is the same procedure that sands the edges off. That 0.01% is simply and mostly unlikely in the eyes of the statistical functions that make up the artificial neural network.
One further danger is that the slop or namely noise that is generated compounds as a false bias. More generated output means more of the 99.99%, which tips the scale further in that direction, which makes the next round more biased still. Spotting the 0.01 per cent now takes a more practised eye than it did five years ago.
Platforms have already worked this out
YouTube's inauthentic content policy, since renamed generic or repetitive content, strips monetisation from mass-produced, templated, low-variation uploads, and in January 2026 the platform deleted sixteen faceless channels outright rather than merely demonetising them. Google Search treats scaled content abuse the same way. Neither is a ban on AI as a tool. Both are a ban on output with no original perspective in it.
The organisations whose entire business depends on separating signal from noise have decided that unoriginal generated content is the noise. They also have to keep their own models clean as a false bias is a heavy tip of the scale.
What a creative process actually involves
To a non-expert, generating an artwork is an enormous leap. The skill gap is so wide that the result looks like genius, and the fact that they produced it themselves makes it more convincing. The reaction is honest but it is also the problem.
A creative process is altogether a different kind of problem-solving that involves research, contemplation, internalising the problem, living with it for days, losing sleep over it, going out into the world and seeing everything through the frame it has given you to name a few. Then one day, hopefully within the timeframe of the project, a stroke of insight arrives. Luck is the right word early on in an expert's career; it gets replaced by expertise as skill accumulates. Not every job is that accentuated of course, but the shape of the process holds: elaboration, observation, analysis, research, exploration, ideation, trial and error, and on top of all of it the artisan craft to actually make the thing.
A neural network at speed can only represent the statistical residue of all this. It is a compression/reduction of the process.
Speed is and always have been the enemy of design
Designers can write better prompts than non-designers, as they should with regards to the skill gap. Yet speed is not a neutral improvement to the design process because with generative processes you arrive at the bottom line without the build-up. You get the strapline without the dialogue that earned it. The obvious question where does innovation fit into that where there is no minute for it.
The real difference between the expert and the non-expert is not access to the tool. It is critical thinking: the knowledge required to evaluate what comes back. Its errors, its underpinnings, its borrowed assumptions, its zero days. If it looks like genius to you, that may be because you do not yet have what you need to see the seams and the fault lines. That is false confidence and we believe it is very expensive. To an expert eye, the same work reads as amateurish.
None of this discussion is new
Herbert Simon and the expert systems of the 1970s and 80s ran into exactly this wall. Expertise was going to be extracted from practitioners, encoded as rules and handed to everyone else. It worked in narrow, well-defined domains and fell apart everywhere the problem had not already been framed. The bottleneck was never the tooling as there are intelligent systems that inherit this frame still today - this is the non-generative side of artificial intelligence.
But these systems cannot obtain the expert judgement, which is not a mere set of rules you can copy out.
It is dialectic
This week OpenAI released GPT-6 Astra, with its president suggesting we may now be in the AGI (artificial general intelligence) era. The pace is of technological articulation is increasing by the minue but designers move too. We believe. in some respects faster, because the tedious technical layer is precisely the part these tools take away.
What remains is what the job always was: keeping creativity sharp, keeping a childish curiosity alive, wandering through the world spotting problems nobody has framed yet and devising solutions nobody has tried. As creatives stay a step ahead they push these systems to get better. It is dialectic, and there is no version of it in which the expert becomes unnecessary, only further crucial.
We use these tools daily but we do not hand them the judgement.
