Issue #21 · July 29, 2026 · The AI Playbook

The Botsitting Tax: Your AI Is Costing You 6.4 Hours a Week

productivityROIworkflowenterprise-AI

The Botsitting Tax: Why 88% of Companies Use AI and Almost None Are Winning

There's a number in McKinsey's 2025 State of AI report that has become a boardroom comfort blanket: 88% of organizations are now regularly using AI in at least one business function. Executives cite it to prove they're "doing AI." It reads like traction. It isn't.

The counterweight is buried in PwC's 2026 CEO Survey, which polled 4,454 executives. Only 12% of CEOs report both revenue gain AND cost reduction from AI. 56% report no significant financial benefit at all. Nearly nine in ten companies are using AI. Fewer than one in eight are winning with it.

This is the defining gap of enterprise AI in 2026. It has a name — the Adoption Paradox — and it doesn't come from bad technology. It comes from measuring the wrong thing.

The tax nobody prices in

The BCG AI at Work survey (4th annual, ~12,000 respondents across a dozen markets, June 2026) found that 42% of regular AI users save 8 hours a week — a full workday. In marketing, that number rises to 60%. In IT, 53%. In HR, 50%. Sounds extraordinary. Then read the next line.

66% of those employees receive little or no guidance on what to do with the time they save. (BCG AI at Work 2026)

The productivity is real. The organizational capture of that productivity is largely fictional.

Glean's 2026 Workforce AI Report puts the sharper knife in. The average knowledge worker now spends 6.4 hours a week "botsitting" — feeding AI context, reviewing outputs, correcting mistakes. That's often more time than they spend producing with the AI. And 69% of AI users admit to "botshitting" — shipping AI-generated output they haven't actually reviewed (Glean 2026).

Do the math on your own headcount. If you have 200 knowledge workers each losing 6.4 hours a week as unpaid QA, that's 1,280 hours a week gone. At a fully loaded rate of $50/hr, that's ~$3.3M/year of botsitting tax that never shows up in the "AI productivity gain" slide.

Why the ROI reports lie

The public conversation on AI ROI has come unmoored from what's actually happening inside enterprises. The data:

Read those two Glean numbers together. Individual gains: real and enormous. Organizational gains: missing.

The individual saves 11 hours a week. The org doesn't notice. Which means one of two things is happening: employees are reinvesting the time into more of the same task at half speed, or the time is being absorbed back into meetings, Slack, and ambient office noise. Neither of those shows up in the P&L.

The 5.3x unlock

McKinsey's 2026 research quantifies the actual unlock: companies that redesign workflows end-to-end around AI are 5.3x more likely to report real enterprise value than companies that just deploy tools into existing processes. Organizational readiness — not the models — explains 48% of the gap between value-capturers and non-capturers.

The maturity ladder is where it gets brutal (McKinsey 2026):

| Stage | % capturing meaningful value | |---|---| | Enablement (bought seats) | 13% | | Automation (some tasks handed over) | 24% | | Reinvention (workflow redesigned end-to-end) | 48% |

Nearly half of the reinvention-level orgs actually win. Almost nine of ten enablement-level orgs don't. And the median company today is proudly parking at "enablement" — where the productivity leaks through the floor.

Drucker had the line 60 years ago: "There is nothing so useless as doing efficiently that which should not be done at all." Your team spending 6.4 hours a week babysitting an AI to do the wrong task faster is exactly that.

What actually works

Here's the playbook if I were sitting on your side of the desk:

1. Kill the "showcase" AI projects this quarter. The ones that exist to be in the deck. They consume license budget, license attention, and — worst — organizational credibility you'll need later for the transformative projects.

2. Pick ONE workflow. End-to-end. Redesign it. Not "add an AI step." Redesign assuming agents can do 60% of the current human touches. What roles change? What roles disappear? What new roles appear? Write it out. Build to it. Ship in 90 days.

3. Set an explicit reinvestment target for saved time. If your team saves 8 hours a week per person, you need a specific answer: "What are we now producing that we weren't before?" If the answer is "more meetings" — you have a leak, not a gain.

4. Instrument the botsitting. Even a rough monthly survey number beats guessing. If you can't measure the tax, you can't manage it.

5. Adopt outcome metrics, not activity metrics. Kill "% of employees with AI access." Measure cycle time reduction on a specific process, or margin change on a specific line item, or first-response time on a specific customer channel. AI investments justified on activity metrics never survive the CFO's second-quarter review.

The strategic implication: Stop asking "How many of our employees are using AI?" Start asking "How many workflows have we redesigned around AI, and what is the measurable change in cycle time, decision quality, or cost per task?"

What I'm Watching

What I'm Building

I ship agents. The one I use every day is TextMyAgent.app — an AI agent that runs your email and calendar from text messages. Get a text when your boss emails. Forward flight itineraries and it books them. Text back and the meeting's on your calendar.

Built because I have six email addresses and got tired of missing signal in the noise. If your inbox is broken or you're losing hours to calendar back-and-forth, try it 14 days free →

Need something? Just reply.

— Keith

Sources: McKinsey State of AI (2025/2026) · PwC 2026 CEO Survey (4,454 execs) · BCG AI at Work 2026 (~12,000 respondents) · Glean 2026 Workforce AI Report · Writer 2026 AI Adoption Survey (2,400 global executives) · Business Insider AI Strategy Survey (July 2026)