The odds of a December rate hike pushed above 65% after a jobs report that came in at 100,000 above the expected 65,000. If a rate hike does happen, AI budget’s will get a whole lot more expensive to carry. Of course, the big headline was Nvidia’s founder who was forced to resign as CEO, just as Anthropic told investors it had crossed one of its own safety thresholds days before its IPO. Vendors, regulators and labs all gave leaders the same message in different forms: nobody outside your organization is going to define what normal AI looks like for you. The cost of that, in both risk and money, lands on whoever signed the purchase order.
Last week’s coverage:
Concentration Risk Stopped Being Theoretical
One chip supplier, one productivity suite, one pool of electricity, and a very bad week to depend on any of them.
Nobody’s Setting the Rules, So You Are
Washington named no safeguards, a lab graded itself, and nearly half of security teams couldn’t say what normal agent behavior looks like.
The AI Bill Arrives While the Returns Stay Vague
Finance wants proof, the CEO wants transformation, and the money got more expensive.
The Jobs Math Works. The Feeling of Being Needed Doesn’t.
Gartner says AI nets out positive on jobs, and an HBR experiment suggests that’s the wrong thing to measure.
Here’s what I was reading.
Concentration Risk Stopped Being Theoretical
Nvidia’s board said it had confidence in Jensen Huang on a Sunday. By Thursday he’d resigned as CEO and left the board. A CIO Dive piece asked what that means for CIOs, and one analyst’s answer was blunt: concentration risk in a single supplier “has stopped being theoretical.” The company most enterprise AI plans sit on top of lost its founder-CEO in four days, with no board chair and no visible succession plan to absorb it. The practical advice was to diversify gradually, which is right, and also hard when switching costs and performance differences were barely weighed in the piece. (What Nvidia CEO Jensen Huang’s exit means for CIOs)
Microsoft 365’s outage from Thursday into Friday underscored the exposure. According to CIO Dive, downtime cost businesses $73.5 million in 2026. When identity, collaboration, security and Copilot all sit behind one provider, one outage takes AI offline along with everything it depends on. The better question than “how fast was it restored?” is what knowledge your people could still reach while it was down. (Microsoft 365 outage underscores enterprise resiliency issues)
Then there’s power. A Stratechery piece argued that opposition to data centers now cuts across party lines after the midterms, and that it could turn into a lasting cap on AI capacity. If that holds, compute stays scarce and expensive, and right-sized models become a budget decision instead of a technical preference. That “if” carries the whole argument (the piece offered little evidence on how fast attitudes or permitting rules might shift), but planning around abundant, cheap compute is the riskier assumption. (AI on the Ballot)
Three different suppliers. Same exposure.
Nobody’s Setting the Rules, So You Are
President Trump has said “I don’t mind guardrails,” but hasn’t named any. He warned against falling behind China and offered no framework. Without safeguards or rulebooks, AI policies that wait for one are already late. Employees aren’t waiting. They’re putting AI to work today, with whatever information they happen to have in front of them. (Trump signals openness to AI guardrails after ‘blue tsunami’)
Anthropic’s version of this was stranger. A Yahoo Finance piece reported that the company’s amended IPO filing says it has met a threshold in its own safety policy, the point where AI does enough of its research to roughly double how fast its models improve. METR, an outside testing nonprofit, estimated its models had been improving nearly twice as fast since October 2025. Crossing the line means more safeguards, not a shutdown, and Anthropic says humans stay in charge. The company that built the model decided for itself that it had crossed its own line, and disclosed it to investors, not to customers. The piece didn’t say what safeguards would change or how the acceleration is measured, which leaves buyers with a headline and no instrument. (Anthropic says a key AI safety alarm has been triggered, days before its IPO)
Closer to home, an Abnormal AI survey of 525 security professionals found that 46% couldn’t establish safe behavioral baselines for AI agents. If you can’t define normal for an agent, you can’t tell when it’s gone wrong. Defining normal means deciding what the agent is for, what it can touch, and what a sensible action looks like, and that’s a governance job, not a security tool purchase. The piece didn’t show that any approach reliably works, but the 46% is the number to bring to your next steering meeting. (Companies struggle to know what normal looks like for AI agents)
The AI Bill Arrives While the Returns Stay Vague
A Fortune piece quoted Mark Williams advising CFOs to set explicit measures of success or failure for AI projects, tie spending to financial results, and consider shorter-term financing as borrowing costs rise. Finance is asking for a definition of failure before approving the next dollar. With markets pricing a December hike at better than 65%, carrying an AI project that can’t show a result just got more expensive every month. The piece offered little evidence on what AI projects actually return, which is the problem in miniature. (CFOs face a tough choice on AI spending as borrowing costs rise, expert warns)
The CEO side of the table wants something different. A CIO.com piece cited an Avanade survey: 77% of business decision-makers said their organizations put too much AI investment into incremental improvements, while only 11% expected AI to fundamentally change operations within a year. CEOs want transformation and CFOs want proof, and both want it on a budget that costs more to hold. Those two demands only fit together if the organization can trust its own information and processes, which is what the piece argued: AI magnifies knowledge and data problems, so redesigning a process on top of records that disagree just makes the disagreement faster. (CEOs have had enough of incremental AI)
The Jobs Math is Working. The Feeling of Being Needed Isn’t.
Gartner projects 10.6 million new jobs and nearly 9 million lost to AI by 2030, a net gain of 1.7 million, according to a CIO Dive piece. Most workers will feel AI in their tasks, not their job titles, which makes the adoption plan a training plan before it’s a tooling plan. The headline is comforting. The piece’s own emphasis on reskilling and IT staffing is the part to act on. (AI will drive a 1.7M job boost, with big changes ahead for IT)
An HBR piece put into perspective something the headcount math can’t. In one experiment, people who were told that AI had created a public-health slogan were 13% less likely to offer their own slogan idea. In another poll of more than 6,000 employees, 44% worried their skills wouldn’t be needed. AI rollouts that hit their productivity targets leave employees feeling replaceable. If people stop asking colleagues to ask Copilot instead, the organization may be removing the exact interactions that make people useful to each other. HBR’s recommendations — set human-versus-AI norms, make contributions visible, point saved time at human strengths — are reasonable, though the piece didn’t show how well they work across different organizations. (How AI Erodes People’s Sense of Value at Work)
A rate-hike probability, a chip supplier’s missing succession plan, a lab’s own safety determination, a survey of security teams who can’t describe normal agent behavior... every one of those points at the same empty seat: someone inside the organization who owns the definition of what AI is for, what it’s allowed to touch, and what counts as working. Vendors won’t fill it. Washington hasn’t. Is an organization that keeps buying AI while that seat sits empty running a strategy? In my opinion, the answer is clearly no.
That’s it for this week’s BeAIReady brief!
If you appreciate the depth of reporting and how I connect the dots, please like, share this post, and subscribe (or share the Brief with a friend!). Thanks!
~erick



