Friday, August 14, 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 Sean Campbell, CEO and founder of Cascade Insights and now a business professor at George Fox University. His firm does deep research for Microsoft, Google, and AWS, and AI adoption work for mid-market companies that look a lot like yours. We tackle the AI question every CFO eventually asks out loud: how do you actually know the AI spend is producing anything? Sean’s answer isn’t a dashboard. It’s a pilot group, a frozen comp plan, and an honest look at whether outcomes moved, plus a blunt warning about what happens to the whole initiative the moment it gets handed to IT. Tune in wherever you get your pod on. Hit one of the below links to check it out:
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: Where’s the “Unwind” Button?
I recently hopped on a quick call with a leadership team asking me which AI approvals to keep and which to launch downrange based on risk. It was a good conversation, but the focus felt wrong. Sorting tasks by how risky they felt is roughly the equivalent of sorting a highway by how fast the cars look.
STOPPING DISTANCE.
Ask any driving instructor what puts people through a windshield and you’ll get a lecture on stopping distance. Reaction time plus braking distance. The ground you cover between the instant you recognize the problem and the instant you come to rest.
Your workflows have one. It’s the distance the work travels between the moment your AI gets something wrong and the moment a human being sees it. Measure it in invoices, quotes, customer emails, purchase orders, or whatever unit stings most when you have to claw it back.
Almost everything written on this subject right now is about where to place approval gates, which treats it as a design exercise. I’m just as guilty as the next consultant. Approval gates work, and the instinct behind them is sound.
The trouble is that at your scale, a gate is a claim on somebody’s calendar. And it lands on the one person who understands the workflow well enough to catch a subtle error, which means it lands on your controller, your ops lead, or you. So your autonomy is governed by the hours your best three people have left. Your risk appetite has very little to do with it.
Stopping distance also stretches on you quietly. The system gets faster. Volume climbs. Your reviewer, forty clean approvals in, starts skimming. None of that shows up on a dashboard. The gate stays on the org chart, stays in the SOP, stays the thing you’d point to if a regulator or a plaintiff’s attorney came asking. It just quit stopping anything.
THE UNWIND TEST.
One question, asked of every workflow you’ve turned loose: If this runs wrong for a week, how many people and how many days does it take to undo?
A bad meeting summary costs an apology. A bad pricing rule costs a quarter and a customer. Set your distance against the cost of the unwind, not against how nervous the technology makes you.
And you have an advantage here worth using. You can walk your own operation and name the three workflows where a mistake compounds instead of just sitting there. That takes an owner about an hour. It takes a Fortune 500 two quarters, a steering committee, and a high-priced enterprise consultancy with a slide about maturity models.
My takeaway? Speed is what everybody’s buying this year. Room to stop is what determines whether you get to keep what you bought. Find the gap between the mistake and the human. Then decide whether you can live with it.
What’s your stopping distance? Hit reply and tell me at chad@chadharvey.com. I read everything that comes into AI for the C Suite®, and the best of it ends up shaping what runs here next.
3. Research Roundup: What the Data Tells Us
Preference Learning: Your Experts Are Answering Questions When They Should Be Teaching
Most AI tools that learn from your people run the same setup. The system offers two options, your expert picks one, it updates. Carnegie Mellon researchers showed that arrangement throws away the one asset your expert holds and the machine does not: knowing what good looks like.
The numbers that matter: Across fifty simulated trials, letting the system pick the questions ranked last at every round and both complexity levels. The paper’s scaling math puts the expert-led advantage near 4.5x on a twenty-factor judgment and 6.3x on a forty-factor one. The harder your quality standard is to write down, the more you lose.
What this means for your Monday morning: Thumbs up and thumbs down is a rating interface, not a teaching interface. Worse, the gain collapses at handoffs. When four instructors rotated every two turns, each arrived with a stale picture of what the system knew, and performance fell below every corrected condition. Shift changes and territory reassignments are where your AI quietly drifts.
The catch: This is simulation, not people. Human trials are in development, and the advantage multiples come from the paper’s stated relationship, not measured output. For a tool judging one or two clear criteria, the payoff is small.
Action item: Ask your vendor this week: can our operators submit their own examples, or only approve and reject what it proposes?
Read our full analysis of this research at AI for the C Suite®.
4. Radar Hits: What’s Worth Your Attention
Claude now watermarks everything it writes. Models launched on or after August 2 embed an invisible mark in generated text plus signed metadata in files, with no opt-out. The mark survives copy and paste. Anthropic hasn’t published accuracy thresholds or a dispute process, so work you wrote and Claude only lightly polished can come back flagged. Get your AI disclosure policy in writing before a client or an HR case forces the question.
OpenAI’s CFO on building an AI-native finance function. It’s a product pitch, but the framework travels. Sarah Friar’s scorecard for AI spend: did it complete work that mattered, what did it really cost including employee time and rework, was the output good enough to use, and did it actually speed up a decision. A teammate who had never written code now builds the tools that turn her advertising forecast into weekly and daily plans. Ask your CFO which of those she can answer this quarter.
Google’s weather model just bought forecasters an extra day. WeatherNext predicts cyclone track, intensity, and wind structure roughly a day earlier than anything before it, verified in Nature with the National Hurricane Center and the UK Met Office. The weights are now open source and free. You won’t run it, but your insurers, carriers, and plant managers will. Ask what your business continuity triggers are keyed to.
5. Elevate Your Leadership with AI for the C Suite®
Here’s the uncomfortable truth underlying my thoughts this week: your approval gates are only as strong as the three people holding them, and nobody has put that on a slide for you. If you want an outside read on where your autonomy is actually capped, my Q1 2027 calendar is open and I’m booking now. Your real workflows, your real people, no maturity models.
And if someone on your leadership team is still sorting AI by how nervous it makes them, send them AI for the C Suite®.
Stay safe. Stay healthy. Be strong. Lead well.
Chad