There’s a principle buried in the drag equation that explains more about AI failure than any Gartner report:
F = ½ρv²CdA
You don’t need to remember the math. You just need to understand one thing: Drag increases with the square of velocity. Go twice as fast, get four times the drag.
Current AI models are the most powerful engines for productivity most organizations have ever bolted onto their operations. And it is exposing, with mathematical cruelty, every bit of drag that was already there.
Executives treat AI like a bigger engine. Bolt it onto the rocket, go faster. Simple. But that’s not how rockets work. A more powerful engine doesn’t help if the rocket is shaped like a brick. And most organizations? They’re bricks.
4 Organizations Steps for AI Readiness
- Fix your data semantics before you build AI on top of them.
- Audit your approval chains.
- Measure outcomes, not initiatives.
- Finish things or kill them.
What Is the Shape of Your Organization?
The drag coefficient, Cd, is determined by shape. A streamlined object slips through the air. A brick does not. Same wind, same speed, wildly different outcomes.
Your drag coefficient is your culture, your decision-making structure, your incentive systems. It’s every layer of approval that exists because nobody ever removed it, every meeting that exists to prepare for another meeting, every initiative that justifies someone’s headcount without connecting to an outcome anyone can name.
I’ve spent a decade building machine learning systems inside large enterprises. The pattern was always the same: The technology worked, but the organization got in its own way.
In earlier generations, big data platforms, predictive models, recommendation engines and NLP systems moved slowly enough that this didn’t matter. The pace was forgiving. You could afford to be a brick at bicycle speed. But LLMs and agentic systems are not bicycle speed.
If your organization doesn’t take initiative to solve problems, AI won’t start solving them for you. But your competitor who does take initiative will now solve them at a pace that makes your quarterly planning cycle look like a geological epoch.
If your priorities are driven by org charts instead of outcomes, AI won’t fix the aim. A VP gets a budget, the VP picks a project, the project serves the VP’s visibility. The “strategic AI priorities” end up mapping one-to-one to reporting lines. Meanwhile, the cross-functional work that would actually deliver value dies in the gap between two leaders who can't agree on who owns it.
If you build for demos instead of users, AI will help you build the most jaw-dropping demo anyone has ever seen at a steering committee meeting. The deck is gorgeous. The chatbot is fluent. Someone says “art of the possible” without irony. Everyone agrees to a pilot. The pilot never gets past the sandbox and quietly dies when the champion moves on. Then the next team starts the cycle over with the same use case. AI doesn’t break this loop. It makes it shinier.
If your people are rewarded for looking busy instead of delivering outcomes, AI is the single greatest tool for performative productivity ever invented. More decks. More dashboards. More AI-powered everything. None of it connected to anything that matters.
If six teams all refuse to use each other’s tools, AI won’t unify them. You’ll get six separate AI initiatives, six vendor contracts, six competing chatbots and a Center of Excellence that exists primarily to produce a newsletter.
What Is Your Organizational Cross-Section?
The A in the drag equation is the reference area: how many teams, systems, handoffs, legacy platforms and half-finished migrations are exposed to friction.
If your data is messy, siloed or unreliable, AI won’t clean it up. And the problem goes deeper than dirty data. Most organizations don’t actually agree on what their own concepts mean. “Exposure” means one thing in risk, another in finance and a third in the regulatory reporting system nobody’s touched in five years. AI will take this semantic fragmentation and produce confident, well-formatted, completely wrong outputs. And because the outputs look authoritative, people will trust them. You aren’t just failing to get value; you're actively manufacturing misinformation about your own business.
If you don’t RTFM, AI won’t RTFM for you. It will hallucinate an answer that sounds plausible, and nobody in the room will know the difference because nobody read the documentation in the first place.
If you’ve hollowed out your subject matter expertise through attrition, early retirement packages and outsourcing, AI cannot replace what nobody documented. It will generate plausible-looking work that no one remaining has the expertise to evaluate. Institutional amnesia with a chatbot on top is not automation.
If your platform is three generations of tooling running in parallel because no migration ever fully completes, AI won’t fix the architecture. It’ll be one more thing duct-taped onto a stack that was already incoherent.
The Part About AI Nobody Wants to Hear
You can’t fix the drag later. At low velocity, a brick and a bullet experience roughly similar forces. The problems were always there. You just couldn’t feel them. AI makes you fast enough to feel them.
If you didn’t choose the right vendors before, AI won’t choose them for you. You’ll evaluate the wrong ones faster. You’ll onboard the wrong ones faster. You’ll realize they’re wrong faster, and then you’ll start a new vendor selection process, which will also be faster and also wrong.
If you don’t have the skills to build with AI, you’ll ship something that looks like it works in staging and collapses in production. And it’ll collapse faster than your previous failures because you built it faster.
If you can’t distinguish between activity and outcomes, AI will flood the zone with activity. The dashboards will be prettier than ever. The signal-to-noise ratio will be worse than ever.
If you don’t have governance, AI won’t govern itself. You’ll move faster toward decisions that nobody reviewed, approved or understood. And when something goes wrong, “The model recommended it” will be the new “that’s what the spreadsheet said.”
One more thing the drag equation won’t tell you: AI doesn’t accelerate everything equally. It amplifies interfaces, which are the places where clean, structured work touches a tool. Support ticket triage, document drafting, code generation: these see benefits fast. But strategy, cross-team coordination, ambiguous decision-making will barely budge. Any plan that treats AI as a uniform accelerant is already wrong.
What Do You Actually Do With AI?
You fix the shape of the rocket before you upgrade the engine. I know that’s not the answer anyone wants. It doesn’t make a good keynote. It doesn’t fit on a vendor slide. But it’s physics.
I should be honest: “fix your organization before adding technology” has been the standard prescription for decades. People said the same thing about ERP, cloud and digital transformation.
So why does it matter this time? Because the penalty scales differently. Previous technology waves added capability roughly proportional to investment. If your organization was messy, you got less value from your ERP, but you got some. The cost of dysfunction was roughly linear. AI’s relationship with organizational dysfunction is steeper. The well-organized ones compound their advantages. The poorly-organized ones compound their confusion.
Don’t Be a Brick
Look at what happened at C.H. Robinson. When Dave Bozeman took over as CEO in mid-2023, revenue had fallen nearly 29 percent in a single year. The freight market was in the worst recession in more than a decade.
The company didn’t start with AI. It started with the boring stuff. An operating model borrowed from Bozeman’s years at Amazon and Caterpillar. The quote-to-cash workflow mapped end to end. Waste identified and removed. Decision-making flattened. Thousands of roles cut. Explicit productivity targets set. The unsexy stuff.
Then the team deployed AI agents into those clean processes. Not as a pilot, but wired directly into workflows they’d already fixed. An agent that generates freight quotes in 32 seconds instead of hours. More than 30 agents processing more than 3 million tasks. Eight consecutive quarters of outperformance, during a recession so deep the Cass Freight Shipment Index hit its lowest Q4 level since 2009.
The honest version is that a meaningful chunk of that 40 percent productivity improvement comes from fewer people doing the same work. Revenue is still shrinking. The real test comes when freight rates recover and Robinson has to absorb volume without proportionally re-hiring. But what matters for this argument is that the AI worked because the organizational drag was already gone. They reshaped the rocket first.
Shopify tells the same story from a different angle. The company spent years of time and billions of dollars building out logistics, including a $2.1 billion acquisition of Deliverr that CEO Tobi Lütke eventually called a “side quest.” In 2023, they sold it to Flexport and went back to their core: commerce software. An expensive correction.
But the decision to subtract before adding is the part that matters. By the time the 2025 memo mandated AI as a baseline expectation, Shopify had already been running Copilot for over a year before ChatGPT launched. The AI had somewhere coherent to land.
Neither of these is a story about predicting the future. Robinson was fixing a company in a freight recession. Shopify was unwinding a strategic mistake. They weren’t preparing for AI. They were reducing drag. And when AI showed up, their rockets were already the right shape.
What Reducing Organizational Drag Looks Like
1. Fix Your Data Semantics Before You Build AI on Top of Them
If a core business concept means three different things in three different systems, reconcile them. Build data contracts between teams: explicit agreements about what fields mean, what freshness guarantees exist and what happens when they break. This is unglamorous. It is also the single most important thing most organizations can do before deploying AI.
2. Audit Your Approval Chains
For each one, ask: What decision does this enable and what would happen if we removed it? Many exist because someone created them during a crisis years ago and nobody ever revisited them. Remove the ones that don’t catch actual errors.
3. Measure Outcomes, Not Initiatives
If your teams are measured on how many AI pilots they launch rather than what those pilots produce, you will get a proliferation of pilots that never reach production. Count the things that move the balance sheet, not the things that kick off.
4. Finish Things or Kill Them
Every unfinished migration, every half-built platform, every pilot that’s been “about to scale” for 18 months: each one is exposed surface area. Subtract before you add. Then automate what remains.
A well-shaped rocket with a modest engine will outperform a brick with Saturn V strapped to it every single time. And a well-shaped rocket with the engine running doesn’t need permission to leave the pad.
Fix the shape. Then turn up the engine. Or don’t, and wait for your competitors to do it so you can watch them fly.
