Most companies haven't adopted AI. They've licensed it.

Hiya 👋 

"We adopted AI" has become one of the least informative sentences in business. It can mean anything from "we bought 40 seats of ChatGPT" to "we rebuilt three core workflows and ship internal agents weekly." This week's stories sit across that entire range, from a system we just put into production to a McKinsey study on what separates the top 6% from everyone else.

Let’s get into it 👇

Opening a tab and describing your problem works. But you still have to go get the answer. You're still the bottleneck.

Here's a submission triage system we built for an insurance brokerage getting 100+ quote requests a day that a dozen people deal with. It's wired into their HubSpot, where the work was already happening. Names changed of course, but the workflow is real.

A new quote request comes in by email. The system reads it, pulls the data points that matter, matches against the carriers they write, ranks the best fits, and flags considerations before anyone opens the thread. Then they approve it and AI + automation handles packaging it up, sending it off to the right team, and updating the deal in their CRM.

Work that took 5 to 20 minutes per submission now takes about 90 seconds to review. Accuracy is holding above 95%.

The shape of this problem shows up in every business. Requests come in, someone reads them, applies the rules in their head, routes them forward. Insurance submissions. Inbound leads. Support tickets. Vendor invoices. Grant applications. Same pattern, different industry.

Humans can stay in the loop to catch what the system may miss and handle edge case judgment calls. The rules AI uses are theirs to change. Change them once, applies to every request after. That's a dozen people getting dozens of hours back every week.

That's the difference between AI as a prompt and AI as a system. A prompt in a chat window is single-player. A system is something that runs in the multiplayer environment of an actual business.

You don't have to rebuild the company to get there. A lot of teams look at AI and assume the answer is starting over. But purpose-built workflows wired into the tools you already run (HubSpot, Salesforce, whatever the stack is) move faster and pay back sooner. Classic "Walk before you run."

HBR ran a piece this week that opens with an image worth reading twice. An official at a large central bank told the authors that when employees work on their secure, no-AI, bank-issued PCs, they often have their personal laptops open next to them with ChatGPT running.

The piece documents what happened at BBVA, one of Europe's largest banks, when they stopped treating AI adoption as a top-down rollout. Every big-bank gen AI program you've read about follows the same script: central team picks the tools, builds a clunky internal version, mandates training, rolls it out slowly, measures nothing useful. BBVA did the opposite. They followed their employees.

The result: over 11,000 active users, 4,800 custom internal tools built by employees themselves, and reported time savings of two to five hours per person per week. The lesson the authors pull out is blunt. Centralized mandates produce slow rollouts and unimpressive results. Employees produce actual adoption.

The pattern BBVA faced is running inside most companies right now, whether leadership sees it or not. The demand for AI is already in the building. Whether any of it is being channeled into something useful is a different question.

If you run a small or mid-market company: Before you commission another AI vendor evaluation, ask five people on your team what they're already using and what they're trying to do with it. The people closest to the work have usually figured out which problems are worth solving.

📄 Read the HBR piece → The Hidden Demand for AI Inside Your Company

Every company says it's adopting AI. Ramp published the numbers that show what real adoption looks like.

Geoff Charles, Ramp's Chief Product Officer, put out an essay this week with the kind of data almost no company shares publicly. 99.5% of Ramp staff active on AI tools. 84% using coding agents weekly. 1,500+ internal applications shipped in six weeks by 800+ different builders. 12% of human-initiated pull requests on their production codebase now come from non-engineers.

For context: PwC's 2026 Global CEO Survey found 56% of CEOs report getting nothing from their AI adoption efforts. McKinsey pegs high-performer companies, the ones getting 5%+ EBIT impact from AI, at around 6% of the market.

Ramp is a tech-forward company with a lot of engineers, so a direct comparison to a 200-person operational business doesn't quite work. But the structural lesson does. Ramp uses the same Claude, the same GPT, the same off-the-shelf agents everyone else has access to. What's different is how they designed the organization around them. An internal agent platform called Glass, built by four people in under three months, authenticates once and connects to 30+ tools automatically. When someone has an idea, they ship it themselves instead of filing a ticket and waiting six months.

Charles closes with a question worth borrowing: what's the last workflow your team changed because of AI, and who changed it? If the answer is the AI team or the engineering team or nobody, the company isn't AI-pilled. It's AI-adjacent.

📝 Read Geoff's essay → How to get your company AI-pilled

The pattern of the winners fits on a page, and it's not what most companies are doing.

McKinsey dropped their AI Transformation Manifesto earlier this month. The firm studied 20 companies they consider leaders in AI-driven transformation. Average results: 20% EBITDA uplift, break-even in one to two years, and $3 of incremental EBITDA for every $1 invested.

What those companies had in common wasn't tooling or headcount. It was focus. They picked one to three business domains and reinvented them. They didn't run twenty pilots.

The piece names twelve themes. Three of them matter most for mid-market operators:

• Economic leverage points beat long lists of use cases. Every business has a few places where small operational improvements show up directly in the P&L. Winners concentrate there. Everyone else spreads thin across an ideas backlog nobody is accountable for.

• Senior business leaders have to own the transformation, not IT. McKinsey couldn't find a single success story where the work was run out of a technology function. The leaders driving it were one to three levels below the CEO.

• Trust and guardrails are the condition for deployment, not an afterthought. You don't scale an agent across a business function until you can show the work, audit the decisions, and intervene when it drifts.

The manifesto is written for the Fortune 500, but the pattern scales down cleanly. Pick one workflow where fixing it would show up on your P&L. Put an operational leader on it, not the person "in charge of AI." Give them a quarter. If nothing got measurably better, either the workflow was wrong or the ownership was.

📚 Read the manifesto → The AI Transformation Manifesto (McKinsey)

Know someone who’d enjoy this newsletter?
Forward it along, or send us a note if there’s a topic you’d like us to unpack next.