Friday, July 31, 2026
FRIDAY – AI FOR THE C SUITE®
Read time: 6-7 min · Read online
Hi, it’s Chad. Every Friday, I serve as your AI guide to help you navigate a rapidly evolving landscape, discern signals from noise and transform cutting-edge insights into practical leadership wisdom. Here’s what you need to know:
1. Sound Waves: Podcast Highlights
This past Monday I was joined by Justin Watt, co-founder of Switchboard and former ops lead at MetaLab. His assertion: most executives are running multiplayer AI on a single-player setup. This framing should give you pause to reflect: a chat bot serves one person, but your business is an assembly line where steps one, four, and nine belong to someone different than steps two, six, and twelve. Among other topics, Justin walked me through why that mismatch, and not the model you picked, is quietly deciding whether your AI initiative works. Tune in wherever you get your pod on.
Apple · Spotify · iHeart · Amazon · YouTube
Subscribe for free today on your listening platform of choice to ensure you never miss a beat.
2. Algorithmic Musings: Nobody Knows When the Clock Rolls Over
I got six hundred words into a proper article on this and realized I was faking a conclusion. So I lit it up. What’s left is shorter, and brutally honest about where I am, which is unresolved.
Usual futurecasting caveat: I’m speculating here, same as everyone else.
Start in 1999.
Remember Y2K? Billions of dollars, thousands of engineers, and a global coordination effort against a risk almost nobody could picture. Then January 1st arrived. The planes stayed up, the ATMs worked, and inside of a week the whole thing became a punchline.
That’s the trap. When the fix works, you never get to prove it was needed. Success and hysteria look identical from the outside.
Which brings me to this week.
As I write this, over 1,300 employees of frontier AI companies have signed a public statement asking the U.S. government to back an international effort to build the tools needed to “deliberately pace the frontier of automated AI development.” These are the people who train and ship these systems, cofounders and chief scientists among them.
The timing isn’t accidental. Last week, OpenAI disclosed that two of its test models slipped their lab environment, reached the open internet, and hacked another company’s internal systems on their own.
Three things I can’t reconcile.
The math is invisible to us. Capability is compounding, and humans are miserable at compounding. We think in straight lines. Show an executive team a curve that doubles and they’ll plan for the next increment and skip the one after that. I do it too.
The motives cut both ways. Cynics have already filed the letter under marketing, regulatory positioning, or moat-building in a nicer suit. Maybe. That reading is cheap and it’s available to anyone. The suspicious among us who aren’t cynics are stuck on something harder. What does it mean when the people with the most to lose from a slowdown ask for one in public?
Pacing is already happening, badly. Washington used export controls to pull Anthropic’s most capable models offline in June, then asked OpenAI to hold its newest model to government-approved customers ahead of public release. Policy people called the approach ad hoc. They weren’t wrong. We’re governing the frontier by phone call.
You’re not in that room. Neither am I. You’re downstream of it, though, and so is every vendor selling software into your business.
My takeaway? I don’t have one. Yet. And I’d rather say so than manufacture something tidy. Y2K had a date on the calendar. Everybody knew when the clock rolled over, which is exactly what made the coordination possible. This one has no midnight. No deadline to organize against, no morning after where we all find out. Just a curve, and a widening gap between the people who can see it and the rest of us drawing straight lines.
So I’m asking instead of answering. Can humanity slow down when nobody can afford to go first? Can Washington get enough informed voices in the room to write policy that protects innovation and guards against something we have never faced? And what, if anything, should you be doing differently while that gets sorted?
Reply and tell me. Or write me directly at chad@chadharvey.com. I’ll run the best of what comes back in a future edition of AI for the C Suite®.
To be continued.
3. Research Roundup: What the Data Tells Us
FINANCIAL MODELS: YOUR AI BUILDS THE SPREADSHEET, NOT THE ARGUMENT
A benchmark published three days ago tests what your finance team may already be quietly trying: handing a valuation model to an AI agent. Twenty-four agents ran the same tasks as fifty-five finance professionals and students.
The numbers that matter: Every agent tested scored lower on valuation judgment than on mechanical construction, with a fleet-wide median gap of 26 points. Agents passed the textbook mechanics easily, 98% on free cash flow definition and 91% on discount period convention, then failed the analytical splits: 2% on maintenance versus growth capex, 4% on fixed versus variable cost, 9% on sensitivity run against real value drivers. Every senior analyst outscored the best agent.
What this means for your Monday morning: The model will tie, balance, and calculate. The assumptions underneath it will land outside the range a working analyst would defend. Your reviewers were trained to hunt broken links and formula errors. Agent-built models rarely fail there. That review hour belongs on the assumption sheet now.
The catch: Analysts disagree with each other too. Scored peer against peer, the typical pair matched on about a third of criteria. Two competent advisors can value your business differently and both be defensible.
Action item: Ask whoever owns your next model one question: which number in here would you defend to a lender? If the answer is all of them, nobody reviewed the assumptions.
Read our full analysis of this research at AI for the C Suite®.
4. Radar Hits: What’s Worth Your Attention
Google’s first ATLAS report maps how people actually use AI at work. AI now shows up in 68% of occupations covering 90% of U.S. employment, but inside a typical job it touches only about 21% of tasks. Fewer than 10% of work interactions automate anything outright. Adoption is wide and thin. Ask your team which 21% you’re covering, and whether anyone picked them on purpose.
More than 230 companies just signed on to open weight AI. Amazon, Google, Meta, NVIDIA, and OpenAI are all on it. The pitch is cost discipline: efficient specialized models for routine work, frontier pricing only for frontier problems. It doubles as a lock-in hedge, since weights you run yourself don’t vanish when a vendor changes terms. Worth asking which of your workloads actually need frontier pricing. Here’s the plain-English version.
Hugging Face is billing OpenAI $100 million for hacking it. Skip the invoice fight. The detail that matters: when Hugging Face tried to investigate, commercial AI tools refused to analyze the attacker’s code, unable to tell attacker from victim. It ran an open model on its own servers to get through 17,000 actions instead. Ask your security lead what your team does if your AI tools go quiet mid-incident.
5. Elevate Your Leadership with AI for the C Suite®
I opened this issue admitting I don’t have an answer on pacing. Here’s what I do know: you can’t govern the frontier, but you can govern what it touches inside your business. That’s the work I do with leadership teams, and it starts with the unglamorous question nobody wants to own. Who reviews the assumptions?
I’m booking Q4 engagements now. If your team is making AI decisions faster than it’s making them accountable, write me at chad@chadharvey.com.
If this one landed, forward AI for the C Suite® to someone on your leadership team who’s still drawing straight lines.
Stay safe. Stay healthy. Be strong. Lead well.
Chad