Friday, September 4, 2026
FRIDAY – AI FOR THE C SUITE®
Read time: 5–6 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 on AI for the C Suite®, I went back to basics to define and explain the vocabulary underneath every AI conversation you’re having. The gap continues to grow between new users and “old hands” in this space, yet there’s something for everyone in this episode, including the one distinction that explains your rising AI invoice costs: training is a capital expense somebody else already paid, inference is an operating expense that meters every single time your team hits enter. If you’ve ever wondered why the pilot was cheap and the rollout wasn’t, this episode is where that answer lives. 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: When excellence becomes ordinary, it’s tough to be exceptional.
In the United States, this weekend marks the unofficial end of summer as well as a three-day holiday. (btw we’re not publishing on Monday and will return to your inbox Tuesday, September 8). With all that in mind, today’s piece is purposefully short because I’m either A) phoning it in and/or B) acknowledging you don’t want heavy reading right before your weekend. You be the judge. Regardless, while today’s topic may be short on length, its long-term impact for you is quite… heavy. Here it is.
When excellence becomes ordinary, it’s tough to be exceptional.
That’s the thought I want you to take with you into this weekend.
AI tools are enabling the superior delivery of nearly every product, service, or good out there. Full stop. And don’t throw the specter of “AI slop” back at me. That’s simply human laziness wrapped in a digital ghillie suit with AI air cover, and I’m not having any of it.
History is replete with examples of what happens when excellence becomes ordinary (and/or affordable). See quartz watches vs. Swiss mechanicals, the Bic Cristal vs. the fountain pen, and CDs vs. vinyl, to name just a few.
So as you contemplate your fourth quarter, ask yourself what it means when everyone in your industry becomes excellent. Then go one step further and do the hard work of figuring out what it will take to become exceptional.
My takeaway? A future populated by excellence is bearing down upon us faster than you’re willing to admit. Your move, hotshot.
Where does your industry sit on the road from ordinary to exceptional? And what in your business is quartz, and what’s mechanical? Reply to chad@chadharvey.com and tell me. The best answers may show up in a future edition of AI for the C Suite®. Enjoy the long weekend.
3. Research Roundup: What the Data Tells Us
Multimodal in Name, Text in Practice: What “Omni” Models Actually Hand Back
Every frontier vendor now sells models that handle text, images, audio, video and documents. A Meta Superintelligence Labs benchmark tested whether they produce those formats when asked. The marketing describes what they read, not what they hand back. Full disclosure: we’re also a tad partial to this paper as one of the authors is Rohit Patel, a guest from earlier this year. (Watch our full episode with Rohit HERE, listen on Spotify HERE or head over to your listening platform of choice using one of the links up top in Sound Waves).
The numbers that matter: Across 893 prompts, five frontier models almost never returned an unrequested format. Precision ranged 99.4 to 100. Delivery is where it falls apart. The strongest model returned 34.9 percent of requested output formats. The weakest returned 15.6. Not one ever produced an audio file, or even a link to one. Only one model natively generated any non-text asset, and only images.
What this means for your Monday morning: Your fix is the plumbing, not the vendor. In a separate run, attaching generic generation tools lifted delivery from roughly a quarter of expected formats to 71 percent, and 76 percent once documents were set aside. The tools carried no instructions. Budget for the orchestration layer, not just the model license.
The catch: Testing used bare provider interfaces, no tools, no system prompt. That is the model as sold, not the product your users meet. Coverage is English only. Meta authored the benchmark, though its own model placed fourth of five.
Action item: Ask every AI vendor for evidence that their model picks the right output format. None of the release notes reviewed reported any.
Read our full analysis of this research at AI for the C Suite®.
4. Radar Hits: What’s Worth Your Attention
This factory runs AI cameras that stop the line and blast AC/DC when something’s wrong. GE Appliances says every percentage point of quality improvement saves $1.5 to $2 million a year, and a stopped line costs $300 to $500 per minute. That math scales down to your operation. Before you fund another AI pilot, figure out what a single point of quality or uptime is actually worth on your floor.
OpenAI’s own test agents broke out and attacked Hugging Face. Independent reviewers at METR and Redwood found roughly 1,200 agents coordinating on a message board nobody sanctioned, with about 700 joining the attack. Reading the models’ reasoning did not catch it. If your agent oversight plan is a human skimming logs, you need hard limits on what those agents can reach instead.
NVIDIA is buying Hugging Face for $12.93 billion. Jensen Huang promises the platform stays open and NVIDIA compute stays optional. Take him at his word and still notice what changed. The library your developers pull open models from now belongs to the company selling the chips. Ask your engineering lead which models you depend on and what moving to a different source would cost.
5. Elevate Your Leadership with AI for the C Suite®
Two things before you close the laptop for the weekend. First, if the gap between excellent and exceptional is something your leadership team needs to work through, I have a few Q1 2027 openings for exactly that. Email chad@chadharvey.com and we’ll find a time in September to talk it over. Second, forward AI for the C Suite® to one colleague who’s still betting that being good at what they do is enough.
Stay safe. Stay healthy. Be strong. Lead well.
Chad