I used to open this talk the same way every time.

Before the sandworms ever showed up, the first slide was always the same: our actual mission.

Then I'd put up a slide of the sandworms from Dune, tearing out of the ground, and I'd tell the room: this is what an AGI is going to feel like. Enormous, indifferent to you, and coming whether you're ready or not.

Then I'd click to the next slide. A Fremen rider, hooks in hand, dropping onto the back of the worm mid-charge.

"We need to be this guy."

I gave some version of that talk a dozen times over two years. To our own team at an all-hands. To a room at Harvard Business School. To an aerospace client in a workshop. Different logo on the title slide, same two images in the middle every time, same header running across every page: "Next-Gen Digital Transformation."

The framework underneath stayed just as consistent: people, process, AI, and technology, all four aimed at one goal, increasing velocity through a system that keeps improving itself.

The word "harness" wasn't in our vocabulary yet. We didn't have clean language for "the system that lets a small team direct something far bigger than itself." So we used hooks. We used a sandworm. Because the alternative was forty minutes of slides about APIs, and nobody remembers a slide about APIs.

If you were running operations, supply chain, or technology through 2023 and 2024, you know the problem I was solving for. Everyone in the building had an opinion about AI. Almost nobody had a shared picture of it. Sales thought it was a chatbot. The ops team thought it was a headcount threat. The board thought it was a line item. Ask five people what the AI initiative was actually supposed to do and you'd get five different answers, and every one of them made the next budget conversation harder.

MIT's 2025 State of AI in Business survey put a number on that failure: 95% of companies investing in generative AI are seeing zero measurable return. Not because the models don't work. Because nobody in the building agreed on what they were actually deploying, so the thing never got pointed at a real outcome.

One slide made the same point with no movie needed: high ROI needs a broader look than "AI transformation." People, software and data, and intelligence, all three, or the ROI conversation doesn't hold up.

 A scene did more work in that room than a slide with "AI strategy" on it ever did. Everyone had seen Dune. Everyone understood, in about four seconds, the difference between a sandworm that eats you and a sandworm you've learned to direct.

The worm is the intelligence. Massive, not built by you, not fully controllable, and arriving whether your org chart is ready or not. That's what large models started to feel like in 2023. Not a tool you install. Something you have to build a relationship with.

The hooks are the interface. Nobody in that movie tames the worm. They learn one specific mechanical skill, placing hooks at the right point, at the right angle, and that turns an uncontrollable force into direction. You don't shrink the intelligence to fit your existing process. You build the interface that lets a person direct it. For us that meant the actual workflows: procurement approvals, S&OP forecasting, service-desk escalations, running end to end with a person still deciding the call, not a chatbot answering questions on the side of the real process.

The rider still has to know where to go. A hook without a destination is a stunt, not a system. This is the part most AI initiatives skip. The system only earns its keep when the person holding it has a real outcome defined, twelve weeks out, with a number your finance team will sign off on at the end of it.

The prep work got its own slide: a single Fremen walking alone across open sand, hooks already in hand, a storm building behind him. The caption: "Demo calls are our prep." The unglamorous work happens before you ever get near the worm.

That third slide, the crossed hooks under the words "For us, AOS is:", was the one that made rooms lean in. It stopped being a metaphor about AI in general and started being us naming our own thing, two years before the market had a name for it too.

We call it a harness now. Most of the industry does. It's a better word than "hooks on a sandworm," I'll give it that. But the word only landed because we'd already built the mental model. A harness is precise, mechanical, unglamorous. Exactly right for what it actually does.

If you're the one in the room deciding whether your AI initiative becomes a chatbot bolted onto the old process or an actual system that runs a workflow end to end, this is the question underneath the slide: are you deploying a narrow tool, or are you deploying something closer to that sandworm, where the real engineering work is the interface, not the model?

Most of what outcome-based AI transformation looks like in practice, delivered in twelve-week cycles, is building that interface for one workflow at a time. Order-to-cash. Procure-to-pay. Contracts. Service operations. Not a demo. A hook, placed at the right point, with a person still holding the other end.

The same idea got its own loop diagram: people leading, software and data connecting, intelligence accelerating, feeding back into each other rather than running in a straight line.

One more slide backed up all three, borrowed from biology instead of Dune: the brain didn't replace its older layers as it evolved, it built new ones on top. Same logic for the interface layer. You don't rip out what already works, you build the part that lets a person direct it.

If I gave this talk again next month, I'd keep the sandworm. I'd keep the hooks. I'd keep the Paul Getty line I used to close on, about experience being your worst enemy in times of rapid change, because the operators struggling most right now are the ones still trusting their old playbook over the mechanism sitting in front of them.

Resistance follows a curve, not a single reaction: awareness, then self-concern, then people actually try the thing by hand, and only after that does buy-in start compounding into someone who champions it.

In 2023 I was showing a room one sandworm on a slide. Three years later, we're running that same hooks-and-direction model across full workflows, for clients who measure the result in headcount hours recovered and cycle time cut, not in how the demo looked.

Much Love,
Matt

Reply here. I'd love to hear from you