Your phone's AI writes better legal language, debugs faster, and answers questions more accurately than whatever your company deployed. This is not an accident. It's what happens when procurement replaces judgment.
Why Do Companies Default to Copilot?
Because nobody owns the decision. That's the honest answer.
Copilot integrates with Microsoft 365, legal approved it, IT understands the data governance story. When no one in the organization actually knows what the good options are or what differentiates them, the safe default wins. The problem is that the output is often terrible for anything substantive. We ran Copilot across our team for months. Nobody came back with a use case worth talking about.
The real issue isn't risk appetite or procurement timelines. It's that the role of "person who actually understands AI tools and makes decisions about them" barely exists in most companies right now. It's becoming its own job. And most organizations are filling it with nobody.
Does It Actually Matter Which AI Model You Use?
It matters more than most people realize, and the wrong choice doesn't just slow you down, it convinces people AI doesn't work at all.
Two recent examples from my own work:
I needed service addendums for terms and conditions covering AI products. Rather than pasting into ChatGPT and hoping for a usable template, I loaded all our existing T&Cs into Claude locally and asked it to write addendums that referenced real clauses and language already in those documents. The output wasn't generic. It was a real legal document with specific cross-references to our actual terms. I handed it to our lawyer to review and edit, and it held up. Copilot wasn't going to produce that. Neither was ChatGPT without a lot of cleanup and manual reconstruction.
The second example is more technical. I was building a SaaS app using Claude in a Roo Loop and it broke Azure SAML authentication. I tried Opus two or three times. Claude is strong on architecture and reasoning, but this was a specific integration bug, and I knew Opus was going to struggle with it. I dropped the problem into Codex instead. It one-shotted the fix.
That's the point. Claude Opus is exceptional for certain tasks. Codex is purpose-built for specific code generation problems. Without someone in your organization who knows which model to reach for and when, you deploy a catch-all, get mediocre results, and then either mandate more usage or quietly abandon the investment.
Do Mandates Like Accenture's Work?
Tying AI adoption to promotions is strange to me. You're telling someone their career depends on opening a specific app a certain number of times. That's not adoption. It's compliance theater.
The AI Daily Brief recently covered Accenture doing exactly this, alongside Amazon logging tool usage across teams. You can track how often someone opens Copilot. You can't track whether they did anything useful with it.
Forced adoption also misses how this actually works: people find AI where it fits their role, or they don't. The ones who do will naturally stand out. Their work gets faster and more thorough. That distinction shows up in performance reviews on its own, because the output reflects it. You don't need a dashboard to see who's innovating. Promotion becomes obvious.
Mandating usage before people have figured out the right tools and the right use cases is backwards. You're measuring the behavior before you've created the conditions for it to matter.
What Does Natural AI Adoption Look Like?
At TeleCloud, we post a simple message in Teams every month or two: "What use case are you using AI for lately?" That's the whole program. People share what's working, others pick it up, ideas spread without any mandate or adoption metric.
We also let employees use our TeleCloud paid AI subscriptions for personal use. No forced separation between work AI and home AI. If someone finds a workflow at home that changes how they approach their job, that's the outcome we want. The artificial wall between personal and professional use is part of why the home-to-work gap exists in the first place. When people can explore freely, they bring that fluency into work.
We run learning slots where people log into Replit and actually build something. Not a training video. Not a policy document. They open an environment and try to make software work. When you build something yourself, even something small, you understand what these tools can and can't do in a way that carries over into real decisions. A webinar doesn't do that.
The harder question isn't how to force adoption. It's whether you have someone in your organization who understands these tools well enough to make good decisions, and whether you've created space for people to discover where AI fits in their actual work. Most companies right now have answered no to both. Their employees know it, because they go home and use something better.