Getting people to use AI.

I re-read an article I wrote in 2020 entitled Building buy-in for your design system. At the time design systems were the unsexy infrastructure work nobody wanted to do but everyone needed. Getting an organization to actually adopt one meant navigating politics with middle management, getting the front lines to use the tools, and convincing execs that the work was worth the calendar room.

Reading it now, almost every piece of advice maps directly to what I'm watching now with AI inside mid-sized companies. There's the same political friction, same type of gatekeeping, same middle management resistance, and some 3D chess to get folks to buy-in. The teams I see making real progress on AI aren't doing anything new, they're doing exactly what that article from 2020 prescribed.

The tech has changed but organizational power dynamics and politics didn't. Here's what I'd update if I rewrote the piece today, with what I'm seeing with clients right now.

This sounds simple, but a lot of companies skip right over it. You MUST have a clear vision that anyone can understand and you NEED a real answer to "what's the point?" and "why do we need this?" Almost everything I wrote in my 2020 article maps directly to AI in 2026. Folks license a tool, drop it into the org, send a launch email, and call it a vision or a strategy. That isn't a strategy, that's a procurement event.

The vision is the part that usually breaks first. If you can't articulate what your AI initiatives are for in 30 seconds and in a way that a salesperson, a finance person, and a customer service rep would all nod at, you don't have a vision yet, you have a vendor. The first question I ask in every workshop is some version of "why are you trying to do this?" and "Imagine it's Christmas and you're raising a glass to a successful year of AI. What would you be toasting to?" The silence that follows is usually the most useful five seconds of the conversation.

This might be a hot take, but I don't believe in the AI council/SteerCo approach. I think for any initiative there needs to be a single owner that is 100% accountable for the success of AI in the organization with a clear vision and a strategy that enables it. Having an accountable owner may also sound obvious, but I see this missing constantly. AI shows up as an IT initiative, a marketing initiative, and an innovation team initiative all running in parallel, with no one person responsible for whether the company is actually getting better at this. Without that owner, the program drifts toward whichever team is loudest that quarter.

Start small, focus on one component at a time, and don't try to boil the ocean. The fastest way to kill an AI program is to roll it out across several hundred people on day one and then try to figure out which use case is producing value. Pick one team, one workflow, one measurable outcome, and get that working before you scale.

In 2020 I said don't try to boil the ocean. In 2026 the ocean is bigger and the noise is louder.

In 2020, my advice was pretty straightforward.. find champions across the organization. Partner up with people who are trying to achieve similar outcomes. Include cross-functional representation in the decision-making. Make things together with the people who'll use the system. Resist the urge to go off on an island, because the fastest way to lose momentum and advocacy is to disappear, build something, and hand it to people.

That advice is more important now than it was then. AI sits at the intersection of every gatekeeping function in the company. Legal cares about it, and so do security, HR, and risk. The teams who'll use it day-to-day care about it most of all. If you build in a silo and hand it to them at launch, you'll spend the next six months unwinding objections that should have been resolved in week one.

I watched an internal AI champion at one company negotiate with their own legal team for months before a discovery workshop. Legal's first instinct was that they needed a battle hardened set of disclosures for every possible edge case before they could do anything with AI. The champion's reframe was that they aren't reading it today. Legal wasn't the enemy in that conversation. Legal was doing their job, and the champion's job was to translate between what legal needed to protect and what the business needed to ship.

By the time we got to the next conversation, legal was at the table helping prioritize where AI could go next, because the champion had treated them as a partner with a real concern instead of an obstacle to route around.

Find your champion in every business unit, and make sure they have permission to talk to the gatekeeping functions early. Build with them. Don't disappear into a working group and hand them a finished thing.

Don't mandate AI. Make using/having it sooo valuable that there's a magnetic pull that draws people in vs pushing them into it.

In 2026 there's a wave of "just make people use it" advice rolling through every leadership team I talk to. Tie AI usage to performance reviews. Make it part of the job description and force adoption from the top down.

What happens is sales reps open Copilot, type one prompt and then close it. Marketers run a single task through a sanctioned tool, then go back to using a personal Claude subscription on the side for the work that actually matters.

One team I worked with in the past was quietly working around their CRM for years. They weren't lazy, they just had a manual way of doing things that closed deals. Also the CRM had no real dedicated owner, the data was fragmented, and nobody had asked the sales folks what their actual workflow was before rolling it out. The same pattern is playing out with AI tools right now, and the fix is identical. Ask the people doing the work what would make their week easier, then build that.

Mandates make people defend the past. Carrots make them want the future.

The only exception is data policy. You can and should mandate where customer data goes, what gets logged, what's allowed in a prompt and what isn't. That's not change management, that's compliance, and there's no carrot version of compliance.

The 2020 framing I gave for design systems was that middle management pushes back on adoption because the system competes with their priorities/success metrics (usually how they get a bonus). Get buy-in from the top of the organization, show what's possible to the boots-on-the-ground, and squeeze the middle until they engage.

The 2026 version is harder. Mid-level managers aren't just defending their priorities anymore, they're defending their livelihood. The system that compresses their team's work also threatens the size of their team, the shape of their org chart, and their job. The pure squeeze doesn't work as cleanly, because you're squeezing people whose self-interest is to slow you down.

The fix is to give the middle a new role before you ask them to advocate. Middle managers become orchestrators, quality reviewers, and system/workflow designers for AI. They aren't replaced, their work changes. Once they have skin in the new game, the squeeze works again.

The execution play that still works is a highly visible early win that ICs can show off and that gives executives air cover. The biggest mistake I see is teams trying to retrofit AI into existing legacy workflows first. That's the slowest path to credibility and the one most exposed to political risk. The faster path is to build something on its own rails, which is an acute, visible pain point where AI is doing work the team couldn't do before. Once that wins, the rest of the organization stops asking "is AI real?" and starts asking "where else can we do this?" The middle squeeze then takes care of itself.

In 2020 I told folks to know their numbers, run surveys, do regular roadshows with executives, and speak the language of the business. All of that still holds.

The 2026 update is a missing column. AI ROI dashboards almost universally measure only the upside (i.e., hours saved, utilization, cycle-time reduction, time on task.. things like that). Almost none of them measure incident cost: hallucinations caught, data-leak attempts blocked, prompt-injection events, downstream rework from confidently wrong outputs.

If you only measure the upside, your CFO will fund the pilot until the first incident lands. When that incident lands, your CISO will undermine/defund the whole thing because there's no number on the downside to defend it. You don't get to argue hours saved when the question on the table is what one incident cost us.

A leadership team I work with picked their first AI project specifically because no other factor could claim credit for the outcome. They skipped the easy win in an area where they already had momentum and went straight to a place where the work literally wouldn't exist without AI. The logic was simple.. "When the program eventually gets board-level scrutiny, the attribution needs to be undeniable, and the incident column needs to be clean/clear, so they designed for that on day one."

Some example metrics for this:

In 2020, design system adoption was the infrastructure work that decided which companies stayed coherent at scale. In 2026 AI adoption is doing the same job, with higher stakes and worse incentives. The framework hasn't changed:

Onward & upward 🤘
Drew

P.s. If you missed it, we launched the CLT Startup House on Friday to 150 people! After a hard sprint to get everything ready for the last few weeks on top of wrestling with grief, it was surreal to be standing in the middle of what I dreamed about and see it actually exist. Never doubt what’s possible with God, community and grit. Much love Charlotte, big things ahead ❤️