Ninety-five percent of corporate AI projects, reportedly, go nowhere. According to MIT, the reason had nothing to do with the tools. It was the companies.
Hold that against two conversations, placed side by side.
The first is a friend who runs UX at a hundred-year-old manufacturer — real factories, a brand your parents would recognize. She's sharp. She figured out how to build with AI fast.
Her company did not. It only recently decided digital mattered, and leadership's entire AI strategy has been to tell everyone to move faster without telling anyone how.
So her product managers open Claude and start building things before anyone has agreed what they're building. Dozens of half-baked products, blooming with no strategy.
She is the head of UX, senior to those PMs, the one person who could point all that energy somewhere useful. And she can't.
She's stuck waiting on IT to approve access to the same AI tools her team is already using — asking again and again for permissions that never come, while people with a fraction of her judgment ship mediocre products that solve no real problem.
The second conversation is with an engineering lead on a small product team, the kind where the founder is still in the room for the calls that matter.
They were working out how design gets approved. The default on small teams is to let every stakeholder weigh in and land on design by committee, which reliably produces the worst version of anything.
He killed that idea before it started. Day-to-day, he and the designer would decide together and keep moving. Only the genuinely big questions go up.
Then he told the founder her job was to be a filter that removes the small stuff — so her attention stayed reserved for the few decisions that actually needed her.
Same tools. Claude and Cursor on both sides. Opposite fates.
The difference isn't capability. It's whether anyone is allowed to decide. One company is small enough to let a good decision happen among trusted people. The other spent a century building machinery to guarantee that no decision happens without permission.
The difference isn't capability. It's whether anyone is allowed to decide. One company is small enough to let a good decision happen. The other spent a century preventing it.
That's the constraint flipping in real time. For a decade, the bottleneck was tooling — scarce engineering time, slow production, expensive prototypes. AI is dissolving that. What's left exposed is the older, harder problem: organizational will.
And here is the trap inside the good news. The autonomy designers begged for is arriving. It's just arriving as a private burden, not a collective win.
Andy Budd names the mechanism precisely. The bull case is that designers now need less permission — you can move from "we should fix this" to "I fixed this, and pushed it live." Prototype the onboarding flow, test the copy, clean up the design debt without waiting three months for a roadmap slot.
But autonomy, he writes, has teeth.
"A lot of designers are good at noticing what is wrong. Fewer are good at deciding what should happen instead. Fewer still can make that alternative real enough for other people to judge. AI will expose this gap."
Read that as a description of a whole professional posture, not a few weak individuals.
Design spent years perfecting the critic's stance. You annotate the flow. You bring the research clip. You point to the support tickets. You explain why the "small edge case" is actually the first-run experience for half your new users. Everyone nods, agrees it matters, and moves on to whatever was already on the roadmap.
That posture was comfortable because it lived safely in opposition to the shipped thing. The better idea never had to survive contact with details. "We never got the engineering time" was both a real grievance and a permanent alibi.
AI removes the alibi. If you can prototype the recommendation, the recommendation has to get better. If you can fix the small piece of design debt, you have to decide whether it was worth fixing.
The deciding, crucially, now happens alone. No roadmap to blame. No committee to diffuse the call across. Just you and the thing you shipped.
And a profession built on committee cover, shared critique, and permission-seeking has almost no muscle for solitary judgment. We trained for the argument, not the verdict.
This is the honest reading of that 95 percent. Those projects don't die from bad models. They die because most companies are engineered, top to bottom, to prevent any single trusted person from deciding — and AI doesn't touch that wiring.
Gary Hamel's argument, cited in that same piece, is that hierarchy was our answer to scarce information. When the person on top was the only one who could see the whole picture, routing every decision up made sense.
Neither of those conditions holds now. Everyone can see the forest. Making isn't the hard part. So the machinery we built to promote and protect the controllers has become the single biggest thing standing between a company and the people who already know what to do.
Now widen the fault line, because the same forces isolating the individual are quietly eroding the culture that used to carry judgment for us.
Crit. Community. Shared standards. The apparatus that let a group hold a decision together — instead of dropping the whole weight on one person's terminal at 11pm.
You can feel the strain in the fatigue. Consider the designer and front-end developer who needed an expenses app, searched the App Store, and found it flooded with vibe-coded apps he could immediately tell weren't human-made — and couldn't stand to use any of them.
So he fired up Claude Code and built his own. He'd done it plenty of times before, for work and for fun. This time felt different.
At work he uses AI all day: coding sites with Claude Code, generating designs in Figma, drafting the client emails that ship the AI-made work with an AI-made message. Then he sat down to build his personal expenses app, and it landed.
"I was offloading my free time into an AI workflow as well."
That's the cost of rugged individualism arriving without the collective to hold it. When every improvement is something you could ship yourself, the workflow expands to fill your evenings, and the sense of accomplishment thins out until it feels hollow.
His fixes are telling, because they're all about reclaiming judgment as a human act. Do the small tasks yourself. Do the fun parts yourself — the AI won't have fun for you. Never outsource the writing, because writing is thinking. Invest in the work, not the pipeline that got you there.
That's one person rebuilding a boundary alone. The more durable answer is collective, and it's already visible in the community.
Look at what Rogie flags about the tools designers are now building for themselves. What he loves is that it's connective.
He points back to what Dribbble, the Figma community, conferences, the blogging scene all did — push people together, celebrate each other's specific skills, see each other. "I hope we do more things like this with our newfound development powers."
Set that instinct next to the engineering lead who made his founder a filter. They're the same move at different scales: deliberately manufacturing the trust and shared repertoire that let a decision happen at all.
That's the real design skill of this era. Not prompting. Not pixel-pushing. Both are now table stakes, and the second one is being absorbed into the execution layer anyway.
The scarce, valuable thing is building the organizational conditions in which a trusted person is actually allowed to decide — and building the community that keeps that person from having to decide alone.
Which loops back to the question we left open last issue: who gets staffed to own outcomes. That was framed as a role question. It's also a human one.
Because autonomy is not a gift you unwrap. It's a weight. The manufacturer's head of UX has all the judgment and none of the permission. Her PMs have all the permission and none of the judgment. Neither is winning.
The small team isn't winning because it has better tools. It's winning because someone did the unglamorous work of deciding who gets to decide — and then got out of the way.
Most hundred-year-old companies could choose that as freely as any startup. Most of them just won't. That refusal, not the tooling, is where the 95 percent goes to die.