Three AI Experts Said the Same Thing This Week. It Changes What 'AI Skills' Actually Means.
Within five days in June 2026, three independent practitioners each reached the same conclusion about what professional AI skill actually requires. It is not prompting.
By Forge Team
Within five days last week, three people who did not coordinate — an academic researcher, a SaaS CTO, and a developer-journalist — each published the same conclusion about which AI skill matters most. Not from the same conference. Not prompted by the same news event. Independently. Ethan Mollick, Charity Majors, and Simon Willison each arrived at the same place: the bottleneck in professional AI work is not generating output. It is judging it.
What three independent voices said in five days
On June 14, Willison pointed to an essay by Narayanan and Kapoor documenting that not a single company in New York's 2025 WARN Act filings attributed layoffs to AI — because the real bottleneck is deciding, verifying, and understanding output, and AI does not do those things. On June 16, Mollick told Simon Sinek that "taste may become the most valuable skill of the AI era," arguing that domain expertise now matters more precisely because it lets you evaluate what AI produces, not because it helps you produce it. On June 17, Majors, CTO of Honeycomb, published "AI demands more engineering discipline. Not less" (425 Hacker News upvotes) — her argument: when any AI can generate code as good as the median engineer in seconds, the professionally valuable skill is the discipline to verify, iterate, and hold quality standards.
Majors framed the shift specifically: code has gone from "treasured, reused, cared for" to "disposable and regenerable." The same is happening to marketing briefs, legal summaries, financial models, and customer communications. If another draft is eight seconds away, the skill you need is not prompting another draft. It is knowing whether the draft you have is good enough.
What to do differently Monday morning
Three independent signals converging in five days point at three specific skills.
Evaluation speed: Can you tell in under 30 seconds whether an AI output is good enough for your purpose — not perfect, good enough? AI generates faster than most people can currently assess. Closing that gap is the practical skill.
Failure-mode recognition: Do you know the specific ways AI fails in your domain? A financial analyst knows which categories of error appear in AI-generated models. A content director knows the patterns of AI copy that read as generic on closer inspection. A project manager knows where AI-generated timelines quietly slip. This knowledge comes from domain experience, not prompting fluency.
Quality criteria: Have you written down what "good enough" looks like before you start generating? It is easy to skip this step — you know bad output when you see it, so it feels unnecessary. But naming your criteria before you prompt changes how fast you can judge what comes back.
Write the quality criteria for one recurring AI task before your next prompt — three specific properties that make an output good enough to use.
Marcus: the content director who fixed inconsistency without fixing his prompts
Marcus is a content director at a 55-person fintech company. He manages three writers and runs the blog, email campaigns, and social content. Since December, he has used AI to accelerate first drafts. He was getting inconsistent results — sometimes the draft needed one edit pass, sometimes it needed to be discarded entirely. Reprompting helped at the margins but did not solve the core problem.
The change came when he wrote down what a usable first draft actually required: a specific claim in the opening two sentences, one concrete data point per section, and no statement that would embarrass a fintech company if quoted out of context. Three criteria, written before he prompted.
With those criteria on paper, he could assess any draft in 90 seconds. The AI still produced drafts he discarded. But he could now tell in 90 seconds which pile a draft belonged in. That changed his workflow more than any prompt adjustment had.
Run the same task prompt twice, then assess which output better meets your criteria — document how long the decision took and which criteria you used to decide.
Priya: the HR business partner who stopped using AI — and why that was also the wrong call
Priya is a senior HR business partner at a 200-person professional services firm. She tried AI-assisted drafting for performance reviews and found the outputs too generic. She went back to writing everything herself.
The skill she was missing was not a better prompt — she had tried reprompting the same content many times. It was a framework that let her use AI on the parts of the review where generic language was acceptable (structure, transitions, standard compliance language) and her own judgment on the parts where it was not (specific examples, evidence, development recommendations). AI for the scaffolding. Domain expertise for the substance.
That is failure-mode recognition applied to workflow design. Knowing where AI fails in your domain tells you where to use it — and where not to.
The conclusion all three of them pointed at
The professionals who will get the most from AI in the next two years will not be the ones with the best prompts. They will be the ones with the sharpest judgment. All three of them pointed at the same thing: judgment is built from domain experience, and domain experience is harder to automate than prompting.
Pick one task you now handle with AI that you used to think through yourself. Check whether you could still do it without AI — and decide whether that matters for your work.
Like this post?
Get the next one in your inbox. Practical AI skills, no filler.