The Question Most Leaders Skip When Choosing AI Tools
There’s a conversation I keep having with leadership teams, and it goes roughly like this.
They evaluated tools. They did the research. They looked at capability benchmarks, sat through vendor demos, maybe ran a pilot with their most technical people. They picked the one that scored highest. They rolled it out across the organization.
Six months later, adoption is sitting somewhere between 10 and 20 percent. The engineers and analysts love it. Everyone else quietly went back to doing things the old way. And leadership is trying to figure out why their AI investment isn’t showing up anywhere in the numbers.
The answer is almost never the tool. It’s the selection criteria.
Most organizations are asking “what’s the most capable?” when the question that actually predicts ROI is “what percentage of our people will still be using this 90 days from now?”
Those are different questions. They produce different answers. And right now, most teams are optimizing for the wrong one.
Why 90% Adoption at 80% Capability Wins Every Time
I was debating this with Matt Graham a partner I work closely with on AI training and organizational readiness and we were going back and forth on Lovable, the vibe coding platform that just raised $400 million at a $13 billion valuation. I’ve had a complicated history with it. I spent a miserable weekend last year fighting with it, and it left a mark.
Matt made an argument I couldn’t fully push back on.
When they run AI training sessions with organizations and show people a range of tools Cursor, Claude Code, Vercel, Lovable Lovable is consistently the one that non-technical people gravitate toward. The HR manager. The admin who’s convinced they’ll break anything they touch. The ops lead who’s been skeptical of every technology initiative for the last decade.
They see Lovable and something clicks. They can actually do something with it. The interface is prompt-based, the output appears while they’re building, and it doesn’t require them to open a terminal or manage a local file system or understand what Supabase is.
Matt put it plainly: getting 90% of your organization using a tool at 80% of its theoretical capability is more valuable than getting 10% of your organization using a tool at 100% of its capability. The math isn’t close. The organizational impact isn’t close.
The most capable tool that only your technical people will adopt is not a technology decision. It’s a change management failure with a good-looking feature list.
Here is the specific claim worth making explicit: When the goal is organizational productivity not individual output adoption rate is the primary variable. A 5x increase in the number of people using AI tools compounds across every department, every workflow, every week. A marginal capability advantage in the hands of a small subset of your workforce does not. Organizations that understand this are making fundamentally different tool decisions than organizations still running capability benchmarks.
Listen to the full episode:
What This Means for How You Evaluate Tools
This isn’t an argument for always choosing the simplest tool. It’s an argument for being honest about your workforce before you make the call.
If 70% of your team is non-technical and in most mid-market companies, that number is higher then you are not buying a tool for your engineers. You are buying a tool for your HR department, your operations team, your marketing coordinators, and the finance manager who has strong opinions about Excel and deep suspicion of everything else.
For that organization, the evaluation criteria have to include: Can someone with no technical background build something useful in their first session? Is the output good enough to change their workflow, even if it’s not perfect? Will they come back to it on their own, or will it sit unused after the training day?
Those are usability questions, not capability questions. And most vendor evaluations never ask them.
The practical test I’d suggest: before any organization-wide rollout, give three non-technical employees access to the tool for two weeks with minimal instruction. Watch what they actually do with it. Their adoption behavior not the feature comparison sheet is your real data.
There are real trade-offs to acknowledge here. Tools optimized for usability often come with constraints that more technical tools don’t have. In the Lovable conversation, I pushed back on vendor lock-in: once your data is established in their environment and your team has built 30 or 40 internal tools on the platform, your switching costs are real. The code may be exportable, but the organizational muscle memory built around a specific platform is not.
The answer to that isn’t to avoid usable tools. It’s to go in with eyes open. If you’re choosing a high-usability platform, build two habits into the rollout from day one: maintain a proper version control practice so your code is backed up independently, and make deliberate decisions about where your data lives. Those aren’t complicated steps. They’re the boring infrastructure work that keeps you from a painful conversation in three years.
The Broader Decision Your Organization Is Actually Making
The Lovable debate is a specific case of a more general problem. As agentic AI tools start entering the market — tools that don’t just respond to prompts but take autonomous actions, log into systems, move files, send communications — the gap between technical and non-technical users is going to become even more consequential.
Matt and I spent time in this episode on Grokbot, XAI’s new persistent agent framework that lets you spin up bots that operate independently in the cloud. It’s genuinely interesting. It’s also a category of tool where the governance stakes are meaningfully higher than a vibe coding app. An app that produces a slightly wrong output is annoying. An agent that makes autonomous decisions with access to your receivables system or your email is a different kind of risk entirely.
This is the trajectory. Tools are becoming more capable and more autonomous at the same time they’re becoming more accessible. That combination means the organizational decisions you make now which tools you adopt, how you govern them, what permissions you grant, who has oversight are going to determine how much risk you’re carrying in 18 months.
For most organizations, the right move is still to start with the usability end of the spectrum. Build the habit of using AI tools across your workforce. Let people experience the wins. Create the organizational muscle memory. Then, as the category matures and governance frameworks catch up to the capability, you’ll have a team that’s ready to go deeper not a team that’s still trying to figure out why nobody opened the platform you bought two years ago.
The companies that are going to win at AI adoption aren’t the ones who picked the most powerful tools. They’re the ones who did the unglamorous work of getting most of their people actually using something and built from there.
If you’re trying to figure out where your organization actually stands before you make the next tool decision, the AI Readiness Assessment at Ascend Labs is a practical starting point.
And if you want to talk through the specifics of your situation before you commit to anything, I set aside time every week for exactly these conversations: tidycal.com/kevinwilliams
