Friday, January 16, 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:
1. Sound Waves: Podcast Highlights
What happens when AI bots become the primary consumers of your content? This week I sat down with Brendan Norman, co-founder of Classify and now working with IAB Tech Lab on content monetization protocols for AI agents. We dig into his Ad Context Protocol, already backed by 30 founding companies and 700 builders creating the infrastructure for advertising in an agentic world. If you’re thinking about how AI agents will interact with your brand, this one’s worth your time. My next solo episode drops Monday, 1.19.26. Hit one of the below links to check out either episode:
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: AI Has a Clarity Problem
There’s a scene in Office Space where Tom Smykowski tries to explain his job to the two Bobs, the consultants brought in to decide who gets fired. He fumbles through a description of taking specifications from customers and bringing them to the engineers. The Bobs stare blankly. “So you physically take the specifications from the customers?” one asks. “Well, no,” Tom admits. “My secretary does that. Or the fax.”
Twenty-five years later, I watch executives have nearly identical conversations about AI.
“So which model should we be using?” “Well, it depends on the task.” “What kind of task?” “It depends on the model.” Blank stares all around.
If you feel confused about which AI tools or models you should be using, and for what, let me put you at ease: That confusion is not a failure on your part. It’s a failure of the AI industry to communicate clearly.
AI companies don’t have a capability problem. They have a clarity problem.
And that distinction matters, especially for leaders of mid-market organizations trying to move from curiosity to real leverage.
The Model Naming Maze
Today’s AI landscape is filled with powerful models whose names feel more like firmware updates than business tools.
GPT-4.x. o-series. Sonnet. Opus. Haiku. Gemini Pro. Gemini Flash.
Which one should you use? For what kind of work? With what level of risk?
I’ve been tracking this space obsessively for years, and even I pause to double-check which model does what before making recommendations. If that’s true for someone who lives in this world, imagine what it’s like for a CEO juggling seventeen other priorities.
These names make sense inside AI labs. They make far less sense inside boardrooms, operations meetings, or strategy sessions.
Leaders aren’t asking, “How was this model trained?” They’re asking, “Can I trust this thing to do meaningful work?”
Clever Names, Unclear Purpose
The confusion doesn’t stop with models.
Take features with friendly or whimsical names. “Gems” is a great example. It sounds approachable, maybe even delightful, but it tells you almost nothing about function, scope, or limits.
The problem is simple: Clever naming increases cognitive load when clarity should be reducing it.
Executives don’t need cute. They need orientation. What does this tool do? What should I use it for? What should I not trust it with? Those questions too often go unanswered.
Leaders Don’t Want Models. They Want Roles.
Most AI companies miss this entirely.
Executives don’t think in terms of models and benchmarks. They think in terms of delegation. They want to know whether this thing can research a market, draft a memo, summarize a forty-page contract, or challenge their assumptions without eating up an entire morning. They’re not evaluating technical architecture. They’re asking the same question they ask about any new hire: What can I hand off to you, and will you embarrass me?
In other words, leaders want roles, not raw tools.
A research analyst. A first-draft writer. A code reviewer. A meeting synthesizer. A strategic challenger.
When AI is framed this way, adoption accelerates. The mental model finally matches how leaders actually work.
From Tools to Coworkers
This is why recent moves toward more “coworker-like” AI experiences are worth paying attention to.
Not because they’re perfect. Because they signal a shift. The future of AI adoption isn’t about choosing the right model. It’s about assigning the right kind of work. When AI feels like something you can responsibly delegate to (rather than something you have to constantly supervise), real leverage begins.
Why This Matters for the Mid-Market
Large enterprises can afford experimentation. Mid-market organizations can’t.
You don’t have AI ops teams. You don’t have endless sandbox time. You do have real decisions, real risk, and real accountability.
When AI tools are unclear, leaders either stall adoption entirely or use them haphazardly and lose trust. Neither works. And you know it.
Clarity isn’t a “nice to have.” It’s the bridge between pilot projects and sustained advantage.
The Real Opportunity Ahead
The next competitive advantage in AI won’t come from a marginally smarter model.
It will come from helping leaders understand what to trust, when to use it, and where human judgment must remain firmly in control.
AI doesn’t need to be smarter. It needs to be clearer.
And the companies and leaders who recognize that will pull ahead.
If you’re wrestling with which AI tools deserve a seat at your table, and what roles they should actually play, I’d love to hear what’s tripping you up. Hit reply.
3. Research Roundup: What the Data Tells Us
One paper worth your full attention this week:
Hybrid AI Architectures: Why Your AI Shouldn’t Be Doing the Math
Source: Kodathala & Vunnam (2026), arXiv:2601.01774v1
Turns out everything we thought about AI doing calculations was wrong. New research testing six leading AI models across 100 engineering problems reveals a stark reality: when AI tries to crunch numbers directly, it fails catastrophically. But there’s a fix that cuts errors by nearly three-quarters.
The numbers that matter: Direct AI computation produced errors of 76% to 126%, making it useless for anything requiring precision. But when AI handles problem formulation while traditional solvers do the actual math, errors drop to 22-30%. That’s a 73.5% improvement on average. Electronics calculations saw the biggest gains at 93.1%.
What this means for your Monday morning: If your team is feeding engineering problems into ChatGPT and trusting whatever number comes back, stop. AI is brilliant at understanding what you’re asking and translating business problems into mathematical form. It’s terrible at arithmetic. The winning architecture uses AI as the front-end translator and hands off the heavy computation to proven numerical methods.
The catch: Even the hybrid approach still produced 22-30% residual errors from occasional equation misformulation. You need verification procedures before production deployment. Performance also varies wildly by domain. Fluid mechanics barely improved while electronics transformed completely.
Action item: Audit how your technical teams are using AI for calculations today. If there’s no numerical solver in the workflow, you’ve got a reliability problem waiting to happen.
Read our full analysis of this and all other analyzed research papers at AI for the C Suite.
4. Radar Hits: What’s Worth Your Attention
Why your LLM bill is exploding, and how semantic caching can cut it by 73%. If you’re running AI workloads, your engineering team should read this. The key insight: users ask the same questions in different ways, but traditional caching only catches exact matches. Semantic caching matches by meaning, not text. One team hit a 67% cache rate and cut LLM costs by 73%. Worth a conversation with your technical leads about whether this applies to your AI deployments.
Anthropic launches Cowork: Claude Code for non-developers. This signals where enterprise AI is headed, and it’s not coding-only anymore. Cowork gives Claude access to folders on your computer and lets it execute multi-step tasks autonomously: organizing files, creating spreadsheets from screenshots, drafting reports from notes. Currently Mac-only and limited to Max subscribers, but the direction is clear: agentic AI tools are expanding beyond developers. Start thinking about which business processes could benefit.
Claude Code 2.1.0 arrives with smoother workflows and smarter agents. For companies evaluating AI coding tools, this release shows Anthropic treating Claude Code as infrastructure, not an experiment. New features include session portability between devices, hot-reloading of skills, and agents that keep working after hitting permission walls. If your dev team is piloting AI coding assistants, ask them whether Anthropic’s approach to agent resilience beats what they’re seeing elsewhere.
5. Elevate Your Leadership with AI for the C Suite
If your organization is stuck in the “which model should we use” conversation instead of the “what work should AI be doing” conversation, that’s exactly where I help. I work with mid-market leadership teams to cut through the noise and build AI strategies that actually translate to your Monday morning.
(Shameless plug: Book a call and let’s talk about what roles AI should play on your team.)
Until next week—
Chad