The 5-Minute
AI Decision
How I decide AI in production. One email a week. One signal, one decision it forces, one framework that explains why. Five minutes to read. Use it all week.
Past Issues
The 6% Who Redesigned the Work
Deloitte's 2026 Human Capital Trends survey of more than 9,000 leaders found 66% say designing how humans and AI work together matters, but only 6% are doing it. The gap holding back AI returns is not capability, it is work design, and the leaders who redesign the work are twice as likely to beat their AI ROI targets.
The Model Was Never the Bottleneck
Box's July 14 survey of 1,640 IT leaders found 96% say their AI agents need company knowledge to work, but only 36% have connected agents to trusted internal content. The model was never the real constraint, and wiring an agent to ungoverned knowledge is its own liability.
The Model Is Not Neutral
Booz Allen found four Chinese code-generation models wrote more vulnerable code for U.S. government users and refused far more often on topics Beijing treats as sensitive. The AI writing your software is not a neutral tool, and most CEOs cannot say what is inside their stack.
Capability Is Not Permission
A better model is not a promotion for the agent running it. Gartner expects 40% of enterprises to demote autonomous agents by 2027, and the ones who backtrack granted autonomy by capability instead of earned track record.
Nobody's Coming to Set Your AI Standard
In one month the EU delayed its high-risk AI rules to 2027, Colorado repealed its own AI Act, and a White House order ruled out federal AI licensing. The deadline isn't coming, so a higher-than-required governance bar becomes the advantage competitors are free to skip.
The Off Switch You Don't Hold
Anthropic released a guarded version of its Mythos-class model on June 9. Three days later a US export-control directive led it to suspend access for all customers. The tools you depend on can be switched off by someone who isn't you.
Your Model Has an Expiry Date
Four vendors shipped competing coding models in one week, and one of the leaders got a shutoff date. Model choice just became procurement on the vendor's calendar.
Adopt Second
Every platform spent this week telling CEOs to deploy AI agents now. The most governed adopters are quietly reversing. The real decision is timing, not speed.
The Hidden Operator Audit
AI adoption does not always show up as a new tool. Sometimes it shows up as an old job title that no longer describes the work. Your org chart shows roles. It does not show who owns the AI-mediated workflows your company now depends on. The risk this week is operating continuity risk, and the systems your company runs on may not be on the IT map, the risk register, or the org chart.
The Stakes Threshold
At SAP Sapphire 2026, Christian Klein launched the Business AI Platform with more than 200 specialized agents and 50 Joule assistants, anchored, in his words, in the business processes, data and governance of mission-critical ERP. The keynote named accuracy, compliance, KPIs, and adoption challenges. It did not name the line above which a human still decides. Every newsletter this year has circled the same question without naming it. Call it the Stakes Threshold.
Governance Delay Just Got a Price Tag
On May 11, Alation launched a governance product whose existence is the signal. The reason it exists, in the company's own words: when a board or regulator asks about compliance, most CDOs and their teams spend weeks manually assembling evidence. WRITER's 2026 survey of 2,400 executives puts numbers around the lag: 67% believe they have already had a data leak from unapproved AI, 35% can't pull the plug on a rogue agent, 36% have no supervision plan. Governance delay used to be a future risk. This quarter, it became a line item.
Autonomy Is Not Bought. It Is Earned.
ServiceNow's Autonomous Workforce launched May 5, AI specialists designed to complete entire processes inside enterprise workflows. Anthropic's production data shows the autonomy ceiling is earned by operators over months, not handed over by a vendor at install. Does your oversight maturity match the autonomy your vendor is selling?
The 1,000-Grad Bet Salesforce Just Made
Marc Benioff posted on April 25 that Salesforce is hiring 1,000 new grads and interns to power Agentforce and Headless360. Anthropic's March paper found entry-level employment in AI-exposed occupations dropped 6 to 16 percent, primarily a hiring slowdown. Benioff did the opposite of what most CEOs are doing this quarter.
The Structural Call the 20% Made
PwC surveyed 1,217 executives and found 20% of companies capture 74% of AI's economic value, generating 7.2x more returns. Technology delivers only 20% of that value; the rest is workflow redesign, governance, and workforce reskilling. EY's CEO made the structural call before the results were in.
Why Your AI Keeps Feeling Disposable
Anthropic's latest enterprise expansion celebrates bigger models and bigger deals. But the companies pulling ahead this year aren't winning on model choice. They're winning on the memory layer around it. The model is rented. The harness is yours.
72% of CEOs Now Own the AI Decision. Most Don't Have a Framework for Making It.
BCG surveyed 2,360 executives and found that 72% of CEOs are now their company's chief AI decision maker, double from last year. Half believe their job is on the line. The 15% pulling ahead aren't spending more, they're learning faster.
The Org Chart Is the Bottleneck. Two Companies Just Proved It.
McKinsey deployed 25,000 AI agents alongside 40,000 humans and cut 25% of back-office roles. Zuckerberg is building an AI deputy to bypass his own management layers. Two very different companies reached the same conclusion: the structure is the constraint, not the technology.
Your Team Hit the AI Tool Wall. They Just Haven't Told You Yet.
BCG and UC Riverside studied 1,488 workers and found a tipping point: productivity gains from AI tools collapse at four or more. The cost is 39% more errors, 33% more decision fatigue, and a third of your best people planning to quit. The question isn't how many tools you've deployed. It's how many your people can actually absorb.
94% Will Keep Spending. Half Think Their Jobs Depend on It.
BCG surveyed 2,360 executives and found CEOs doubling AI budgets while half believe their careers depend on the results. The difference between winners and the rest isn't budget size. It's whether the money builds autonomy or deepens dependency.
Get the next one
One email a week. Five minutes. Unsubscribe anytime.