Friday, March 13, 2026
FRIDAY — AI FOR THE C SUITE
Read time: 8-9 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:
|
TL;DR New research tested 15 major AI models across 200,000+ conversations and found that performance drops 39% when people use AI the way they actually work (back-and-forth conversation) versus the way vendors demo it (one clean prompt). Every model tested showed the same pattern. One workaround recovered 15-20% of the loss. Details in Research Roundup below. |
1. Sound Waves: Podcast Highlights
My episode with Dr. Sam Zolfagharian dropped this week, which means her insights on AI implementation are now available for your use. We talk about how HR gets sidelined in AI initiatives, the most relevant findings from MIT’s Project Iceberg, and why your employees aren’t adopting AI tools faster. Check it out wherever you get your podcasts.
Apple · Spotify · iHeart · Amazon · YouTube
Subscribe for free today on your listening platform of choice to ensure you never miss a beat. New episodes release every two weeks.
2. Algorithmic Musings: Every AI-Driven Gain Follows the Same Arc. Most Leaders Stop Paying Attention After Phase One.
Your AI deployment worked. Estimates that took days now take hours. Operational costs dropped 20%. Your team absorbed a quarter more volume without a single new hire. The metrics moved. Leadership is excited.
Congratulations. You’re in Phase 1.
I call it The Thrill, and it’s the first stage of a pattern I’ve been mapping across dozens of middle-market organizations. The AI Advantage Arc describes three phases that every AI-driven gain moves through, regardless of whether that gain involves speed, cost savings, or capacity. The specifics differ. The structural pattern does not.
Here’s the uncomfortable part: Phase 1 is temporary. Not because the gains aren’t real (they are), but because the tools that created them are commercially available. Every competitor in your space can buy the same capability. The advantage you feel right now is a function of adoption timing, and adoption timing is a depreciating asset.
Which brings us to The Plateau (Phase 2). This is the moment your competitors deploy the same tools and achieve comparable results. Speed parity returns. Your improved margins come under pressure from a market that has repriced around the new cost structure. The deals you were winning on turnaround alone start going back to price and relationships. Most organizations stall here. They look around, see the same tools everywhere, and conclude that AI has run its course.
That conclusion is precisely the wrong one. Because Phase 2 isn’t the end of the story. It’s the setup for Phase 3: The Fork.
At the Fork, every organization faces two paths. Path A (the Compounding Loop) belongs to leaders who used their Phase 1 advantage to do qualitatively different work, not just faster or cheaper versions of the same work. They reinvested freed margin into new capabilities. They redesigned roles so their best people could do more meaningful work, not simply more work. They paired every efficiency target with a growth or innovation objective.
Where Path A leads next: The leaders furthest along this path are already thinking about recursive loops, AI systems that improve their own processes and build on their own outputs. That’s not speculative anymore. It’s happening now, and it widens the gap between Path A and Path B every month.
Path B (the Stall) belongs to organizations that treated AI as a project with a finish line. They implemented, celebrated, and moved on. When the Plateau arrived, they had nothing to show for their head start.
Consider a construction company that recently deployed AI in estimating. The speed gains were real and exciting. But the tool is commercially available. Every competitor can buy it. The real question isn’t “Can we estimate faster?” It’s “Now that estimating takes hours instead of days, what do your estimators do with that recaptured time that changes your competitive position?” That’s a Phase 3, Path A question. Simply being proud of the speed is Phase 1 thinking with a Phase 2 expiration date.
The bottom line: the tool was never the advantage. What you do with what the tool gives you is the advantage. And that’s a leadership problem, not a technology problem.
If you want to map where your organization sits on the Arc (and what to do about it), drop me a line at chad@chadharvey.com.
3. Research Roundup: What the Data Tells Us
Multi-Turn Reliability: Why Your AI Tools Perform Better in Demos Than in Practice
Your AI tool looked great in the vendor demo. The demo had one clean, complete prompt. Your employees don’t work that way, and new research from Microsoft and Salesforce just quantified exactly how much that gap is costing you.
The numbers that matter: Across 15 major AI models and 200,000+ simulated conversations, performance dropped an average of 39% when conversations unfolded naturally versus starting with complete instructions. The bigger problem isn’t lower average scores. It’s unpredictability. Variance between best- and worst-case responses on identical tasks more than doubled (112% increase), with performance swings of 50 percentage points on the same task in different runs. Your team can’t predict when they’ll get a great answer versus a bad one.
What this means for your Monday morning: If your teams are using AI tools for customer support, workflow automation, or internal knowledge work, they’re operating in the degraded zone most of the time. The fix isn’t a better model. All 15 tested showed the same pattern, including GPT-4.1, Gemini 2.5 Pro, and Claude 3.7.
The catch: There’s no quick vendor solution here. Reasoning models and temperature adjustments didn’t close the gap. The researchers found one practical workaround that recovers 15-20% of the performance loss: building a recapitulation step into workflows, where the AI summarizes everything discussed before generating a final output.
Action item: Tell your teams two things this week: front-load requirements into a single prompt before starting any AI conversation, and when an AI goes off the rails mid-conversation, start a fresh session rather than trying to correct it. Then ask your AI vendors for multi-turn reliability data, not just benchmark scores.
AI Video Ads Are Coming, And Your Brand Isn’t Ready
Researchers just demonstrated a system that embeds brands into AI-generated videos so naturally that human reviewers rated the results 4.70 out of 5. That’s not a gimmick. That’s a new advertising channel taking shape while most middle-market marketing teams are still figuring out static image generation.
The numbers that matter: BrandFusion achieved a 94.7% brand presence rate across test scenarios, while cruder integration methods dropped to 64%. The real story is resilience: when the brand-to-scene fit was awkward (think automotive brand in a bedroom music video), basic approaches collapsed by 66% in quality. BrandFusion dropped just 10%.
What this means for your Monday morning: If you’re evaluating text-to-video platforms for marketing, the integration method matters more than the video quality. Bad brand placement doesn’t just waste money, it turns off viewers. And here’s the wrinkle for middle-market brands: the system needed extra onboarding for brands without broad cultural recognition. Your brand likely needs that step.
The catch: This is early. Regulations around AI-embedded advertising don’t exist yet, and there are real questions about user consent and transparency. You don’t want to be the cautionary tale when those rules arrive.
Action item: Add one question to every video platform evaluation: “How do you handle brand integration for companies outside the Fortune 500?” The answer will tell you whether they’ve solved the hard problem or just the easy one.
4. Radar Hits: What’s Worth Your Attention
Microsoft launches Copilot Cowork, turning AI chat into an execution engine across M365. This is the moment Copilot stops being a search box and starts doing actual work: rescheduling your calendar, building meeting decks from your email threads, running competitive research. Built on Anthropic’s Claude technology, it’s currently in limited preview with broader access late March. If you’re a Microsoft shop, get on the Frontier program waitlist now. This shapes your 2026 productivity stack.
At Anthropic’s hackathon, a lawyer, a cardiologist, and a road technician beat 500 developers. Out of 13,000 applicants, the winners built real software products in a week with zero coding background. The first-place app automates California building permits. The takeaway for your hiring and internal AI strategy: domain expertise now outweighs technical skill when it comes to building AI tools. Your best AI builders might already be on payroll in operations, legal, or finance.
Ramp spending data shows Anthropic now captures 2x OpenAI’s share of business AI chat spend. Corporate card data across 50,000+ U.S. businesses shows Anthropic went from 10% of combined AI chat subscription spend to over 65% in twelve months, driven mostly by Claude Team adoption in the SMB and mid-market. If you’re negotiating an enterprise AI contract, you now have real leverage to play vendors against each other.
5. Elevate Your Leadership with AI for the C Suite
You just read about the AI Advantage Arc. Here’s a question worth sitting with this weekend: is your organization building toward Phase 3, or are you still celebrating Phase 1 wins that your competitors will match by Q3?
That’s the exact conversation I have with leadership teams in my consulting engagements. My Q2 calendar is committed, but I’m now scheduling Q3 strategy sessions. If you want a clear-eyed assessment of where you sit on the Arc and a concrete plan for what comes next, reach out now so we can get you on the calendar. chad@chadharvey.com
Until next week, keep building.
Chad