Remember three years ago when you were good at your job if you delivered consistent, quality outputs and were fun to work with? That has become increasingly irrelevant as our career success became defined by how quickly we can adopt new tools and take on other people’s jobs that were never our responsibility. Sure, AI makes it easier to copy write, draft reports, and build prototypes — but should we? For all of us the experience of working with AI has looked different but the one consistent has been that more and more of our time has been feeding and training AI to deliver less shitty work.
You’re spending way more time spotting AI bullshit than shipping consistently valuable deliverables. Glean’s 2026 Work AI Index put numbers on it. Workers spend 6.4 hours a week botsitting — supervising and correcting AI — which is more time than they spend using it to actually produce anything. 87% say AI saves them 11 hours a week, yet only 13% say their organization is performing significantly better because of it. And 69% admit they’ve shipped AI output they never verified, didn’t fully understand, or couldn’t confidently stand behind. That’s the horrible relationship the subtitle promised: hours spent supervising, plus hours spent shipping junk anyway, minus almost nothing to show for it.
If you work in tech you’re getting forced to adopt AI, with a gun at your head threatening your job if you don’t adopt it fast enough. As a result, every report on business AI use is showing the same shape: adoption trending toward 100%. McKinsey’s State of AI research tracked the climb — 55% of organizations reported using AI in 2023, 72% in 2024, 88% in the most recent survey. Executives aren’t asking whether that curve is producing value. They’re asking why it isn’t steeper.
Truth is, if you ask anyone on the street, AI adoption is nowhere near 100% for normal people. Pew Research found only 49% of American adults have ever used an AI chatbot — up from 33% in 2024 — and just 24% use one daily. Half the country has never touched the technology tech workers are being threatened over. The adoption curve isn’t universal. It’s mandatory in one building and irrelevant everywhere else.
Even inside that mandated majority, competence isn’t following adoption. Of the 67% of knowledge workers using AI weekly, Glean found only 5.5% are proficient. The 4% Pew identifies as constant users didn’t get there by adopting faster — they changed what they were doing: define correct output before running the model, build an evaluation function, let the model retry against its own failure, step back in only when it converges. Anthropic’s Economic Index found this group already delegates 50%+ of their tasks. That’s the difference between botsitting and actually shipping something valuable — and it has nothing to do with how fast you adopted.
None of this is making the relationship better — the mandates are making it worse. Writer’s survey of 2,400 executives found 60% plan to lay off non-adopters, and 54% admit the pressure is already tearing their company apart. Threaten people over usage and they will ship faster with less oversight and less certainty about what they shipped. You get more AI agents floating around with vulnerabilities nobody checked for. That doesn’t just threaten the business — it creates a new kind of technical and organizational debt, the thing Atlassian’s 2026 State of Teams research calls a fragmentation tax. Their number: $160 billion lost to it a year.
You’re seeing some shocking outcomes of all the poor corporate decision-making — teams are now being asked to reduce their token usage, businesses are cutting back on AI spending, and teams are shifting to Chinese models. It’s also showing up as mea culpas of layoffs, like Ford rehiring the 350 engineers it replaced and jumping from No. 15 to No. 1 in the JD Power quality rankings in a single year.
The reality check is clear. Rushing to AI adoption hasn’t led to consistent corporate AI-generated outputs, and consumers simply don’t understand what to do with AI aside from using it like a fancy Google search.
As we’ve hit the halfway point of 2026, we should all be forced to question what relationship we want with AI in 2027. The resources in this newsletter should help — and add a comment about what topics you want us to cover in the coming weeks.
We brought Molly Sands on because Atlassian’s Teamwork Lab just ran the most rigorous team-level AI research I’ve seen this year — 12,000 knowledge workers, 170 Fortune 100 executives. The finding that stopped me: teams that define how they’ll use AI together outperform teams with higher individual AI usage, with identical tools. The botsitting problem feels individual. The data says its root cause is collective — and individual training programs will not fix a team-level coordination failure.
Why AI adoption is still uneven — and what “drowning in forced change” actually looks like inside organizations
Why the governance gap — no CAIO, no policies, no connective tissue — is the real reason AI experiments don’t compound
How Fortune 500 companies lose $160 billion a year to coordination chaos, and why deploying more tools makes it worse
Why team-level AI thinking produces ROI that individual mandates can’t replicate — even when the tools are identical
What an AI working agreement is — a team protocol for how you’ll use AI together — and what Atlassian found happens to ROI when teams run one
How to read where your team sits on the AI maturity curve, and what the jump to orchestration actually requires
“The AI transformation is still an individual experience. And I think there’s so much potential that we really need to unlock together.” — Dr. Molly Sands, Head of the Teamwork Lab, Atlassian
Listen now: Spotify | Apple Podcasts | YouTube
Brittany Hobbs and I worked through four major 2026 AI research reports — OpenAI, Writer, Glean, Section — pulling the signal that cuts across all of them. Four years of botsitting at scale has become a finance problem: the organizations that cannot defend their AI spend right now are the ones that spent four years measuring seats and logins instead of outcomes. The window to course-correct is closing — and the budget corrections happening now are not reversible on the timelines most teams are planning around.
OpenAI, Writer, Glean, Section: four reports, one consistent signal — the adoption story hides the value failure
Glean 2026: botsitting is the dominant AI experience — more knowledge workers are losing time to AI than gaining it
67% of workers use AI weekly. Only 5.5% are proficient. The problem isn’t more training days. It’s the model of change.
Four years of measuring seats over outcomes has left AI leaders unable to defend their budgets. The window is closing.
Salesforce agreed to acquire Fin for $3.6B. What they built before that exit is the lesson most orgs are ignoring.
Boris Cherny no longer prompts — he builds loops. What that means for every team not yet running autonomous evaluation.
Listen now: Spotify | Apple Podcasts | YouTube | Episode page
The Playbook for AI Value Creation
Four years of measuring AI adoption has produced a generation of organizations that track everything and know nothing about what they have produced. The 68-point gap is not a vendor problem or a model problem — it is what happens when leadership confuses a procurement decision for a strategy. The Playbook for AI Value Creation lays out four things that need to happen for a program to survive the shift from adoption to value:
1. Be honest about what is actually happening. Most people don’t understand why the AI pressure feels so forced — you can’t mobilize anyone for a shift you haven’t explained. Say the real stakes plainly: whoever masters this first consolidates an advantage the way the internet did to industries in the early 2000s. A training-day slide deck won’t do that.
2. Champion what success looks like — and show it running. Not a deck about AI. A live, rebuilt workflow that’s visibly faster, better, or cheaper. People need to see what it feels like to work somewhere their judgment is the scarce resource, not their capacity.
3. Build collective knowledge. Individual power users don’t compound. Shared tools, shared methods, and institutional expertise do. CAIO hiring hit a record in Q1 2026 — 47 new appointments — precisely because no existing role owns this end-to-end, and someone needs the authority to challenge how work is structured, not just train people on tools.
4. Keep the actual goal in view. The goal was never AI adoption. It was building better products and solving customer problems better than your competitors. Measuring the path instead of the destination is how organizations end up with impressive dashboards and nothing defensible to show for them.
Read: The Playbook for AI Value Creation
Companies Are Rehiring the People They Fired for AI.
39% of business leaders cut jobs citing AI, according to Orgvue — and 55% of them now admit it was the wrong call. Ford is rehiring hundreds of experienced engineers to fix quality issues its automated systems couldn’t catch; VP Charles Poon put it plainly: “Artificial intelligence is a fantastic tool, but it’s only as good as the information you use to train it.” Commonwealth Bank of Australia cut 40+ customer service staff for an AI voice bot, watched call volume spike, and reversed the cuts. IBM’s CHRO is now warning what happens to the talent pipeline in three to five years if entry-level hiring stays frozen. The layoffs were the easy decision. Reversing them is the expensive one.
Read: Employers Who Laid Off Workers Citing AI Are Already Starting to Regret It
Tokenmaxxing Is Dead. The Pullback Is Real, but Slower Than You’d Think.
Uber’s CTO confirmed the company burned through its entire 2026 AI budget in four months as Claude Code adoption jumped from 32% to 84% of its 5,000 engineers — Uber now caps spend at $1,500 per employee per month. Duolingo’s CEO walked back tying performance reviews to AI usage after employees pushed back on being told to “use AI for AI’s sake.” Meta’s own internal memo admits the same exponential problem — tracking toward billions in internal AI spend in 2026 — but the real fix isn’t arriving yet: “In 2027, we expect Meta will move toward managing AI tokens in a more structured way.” The mandate created the mess. Fixing it is still a roadmap item, not a policy.
Read: Tokenminimizing: Meta Moves to Curb Employee AI Usage as AI Costs Reach Billions
Coinbase Cut Its AI Bill in Half — Without Cutting Usage.
While other companies cap spend by capping people, Coinbase’s Brian Armstrong published the alternative: better defaults (open-weight models like GLM 5.2 and Kimi 2.7, not usage caps — 91% of engineers never hit their caps anyway), smarter routing to the cheapest model that can do the job, and aggressive caching that took its hit rate from 5% to 60%. The result: AI spend cut nearly in half while token usage kept growing. This is what the Playbook above calls keeping the actual goal in view — the goal was never less AI, it was more value per token.
Read: Brian Armstrong on How Coinbase Kept AI Spend Flat While Usage Grew Exponentially
AI Strategy Resources
Five signals from the research and practitioner layer.
Americans and AI 2026: Chatbots, Smart Devices, and Views on Impact — Pew Research. 1 in 4 Americans uses AI daily; the majority use it for tasks they were already doing, which is why adoption metrics and value metrics keep telling different stories.
Economic Index: Cadences — Anthropic. Half of the 9,700 Claude users surveyed already delegate 50%+ of their job tasks to AI — but the survey captures people whose work AI is augmenting, not the workers whose roles it is replacing, which is exactly the gap most workforce forecasts miss.
Small and Medium Businesses Aren’t Waiting for an AI Invitation — They’re Already Leading — Microsoft. The SME end of the AI value curve moves fastest because smaller organizations can redesign how they work without the coordination overhead that stalls transformation at scale — and they are building structural advantages right now that will not be easy to replicate later.
UBS: 60% of Companies Have Already Started Curbing AI Spending — UBS. Model routing and open-source defaults are how serious enterprise teams respond to bill shock — the $35K/month individual user and 200% quota overrun are the pattern landing in finance teams right now, not the outlier.
Work AI Index — Glean. Botsitting — spending more time supervising and correcting AI than gaining anything from it — is now the dominant AI experience for knowledge workers, and it is a product design and change management failure more than a skills gap.
If you’re measuring AI by seat counts or weekly active users, you are not measuring AI value — you are measuring exposure. AI Value Acceleration works with product teams and their leaders to build the measurement and workflow systems that turn that exposure into defensible outcomes. If your finance team is starting to ask questions you can’t answer, let’s talk.
Browse all two seasons of episodes at productimpactpod.com.