You adopted AI. Did the judgment come with it?

The collision: speed vs. accountability. AI makes individuals dramatically faster. It probably doesn't reduce the work-load. And the org-level gain from doing more faster keeps not showing up in ROI numbers.

I run my whole operation on AI and I do more varied work and am faster at almost everything than I was a year ago. So are a lot of you. And yet the team-level, org-level return on that speed is far murkier (data below).

A lot of that is because coordination is bottlenecked — both between agents but more interestingly between the people responsible for the work. As I see it there are two primary sources for that bottleneck. The first is integrating AI native workflows with more "traditional" workflows. That shows up as working with different primitives and uneven review/turnaround speeds.

!A race where the AI-native kart boosts on markdown, code, and agents while the others spin out on banana peels labeled Excel, PowerPoint, and SharePoint

The second component is ensuring that the work product retains human accountability and judgment.

IBM famously said: "A computer can never be held accountable, therefore a computer must never make a management decision."

Today though in both individual and team-level workflows human accountability is far from a given. Interestingly it may well get harder as models get better.

I had an interesting experience with this recently where I'd written a PRD for a landing page that had draft copy, visual elements, sections, styling call-outs and more. It was robust. But it was not complete. I had not mocked it up in HTML, spent hours fine-combing every line and every style and layout choice. That was the job I was handing off to someone with more talent and experience in that space.

What came back was a finished landing page specced exactly to the PRD. It wasn't wrong exactly. The copy was serviceable and the layout was clean. What was missing was the hundred small calls about emphasis and structure and which capability claims were actually true enough to put in front of a customer. The expertise required was in the finish, and in this case the finish never happened.

By the way, this isn't about people being lazy either. Iterable's 2026 Customer Engagement Report found that 40% of marketers report *more* work since adopting AI.

In the end the model can draft the thing, ship the thing, even be right about the thing. It still can't own the thing. When the output is wrong, nobody points at the model. They point at you.

I think the accountability needs + work-product translation + the actual jagged edge of AI capabilities helps explain the panic over ROI when enterprises adopt AI. And that panic really took off this week.

• Derek Thompson called it the "wait, is this worth it?" phase
• Gary Marcus has been cataloging the companies admitting it.

Of those, Uber is the loudest recent example. It put AI coding assistants in front of 5,000 engineers in December and burned its entire 2026 AI budget by April, about $3.4B in tokens at $2,000 per engineer per month. Worse, the COO admits he can't tie the spend to meaningful ROI.

Part of this is the bottleneck at review we talked about (e.g. Wikipedia/public repos drowning in AI commits). The 2025 HBR "workslop" study summarized it well: AI-padded work just "transfers the effort from creator to receiver," and leads the receivers of that work to regard their colleagues as LESS reliable and less trustworthy. It's tempting to chalk this up to AI-pilled super humans getting slowed down by slower adopters. This is NOT that story. AI is wonderful *around* a decision (gathering, drafting, routing, summarizing, producing) and dangerous *as* the decision or final pass. Yikes!

New accountability mechanisms and stated/unstated cultural norms need to surround AI-enabled work. The simplest — and for now most practical — way to start is that the human is still responsible for every facet of the delivery. Period.

AI is cutting the entry rung — the exact place judgment used to get built. The collision: corporate incentives vs. employee ones — cut headcount now, against the org's need to keep producing the seniors it'll need later.

Back in April I wrote "The money moved. The jobs didn't follow". It basically told the story of the jobs market as AI capex climbing while hiring wasn't.

This month showed more stories around this macro story.

• Oliver Wyman's CEO survey found the share of executives planning to cut junior roles doubled from 17% to 43% in a year, with hiring tilting toward older workers.
• Tomasz Tunguz said on the GTMnow podcast that SDR and BDR automation is a one-way change. The first rung of three different GTM functions is going at once.
• One recent study pushed back on "the story is AI taking jobs" narrative. The study found 63% of the uptick in youth unemployment might trace back to COVID and remote work. It's hard to train someone new, and harder when you're never in the same room. So some of what looks like AI eating the entry rung is really a hangover from how we reorganized work five years ago.

But the trend is undeniable. In addition to the human and social story today (the most important part), it's also an entry into arguably the most important philosophical questions of our time.

How do we train and develop people if machines are better at stuff than us? What happens if we don't? Will we have people with great judgment in 5 years? 10?

Deloitte for example, flagged a "missing middle" where AI adoption actually *declines* with age. So the senior people who'd model good judgment are often the ones using AI least, while the juniors who'd learn it are getting automated before they start.

It's also not like more AI by default reduces work or the need for humans, despite the CEO mania around efficiency. After all, if automation just deleted work, the most AI-aggressive companies would be the smallest. They're not.

Every, an AI-focused media company/incubator, for example, tried to automate everything it could with Codex and Claude Code, then doubled headcount from 15 to 30. Dan Shipper, Every's CEO, put it well when he said making expert work cheaper does not simply replace experts.