When a company's official AI rollout stalls, are employees quietly getting ahead anyway with their own tools?

This explores whether workers get real value from AI tools on their own, even when an organization's official rollout is slow or stuck. The corpus doesn't directly study unofficial, bottom-up use, sometimes called 'shadow AI'. It does show why formal deployment stalls, and it hints at where individual use works anyway.

The clearest reason for the stall comes from Benedict Evans's argument that making tools easier to build doesn't solve adoption Does easier tool-building actually solve enterprise adoption problems?. He names two barriers that better technology doesn't remove. Most workers don't see their own tasks as things that could be automated. And enterprise adoption needs decisions that cut across departments and budget cycles. A historical analysis running from GPS to modern agents reaches the same conclusion from another direction: capable systems stall when five conditions are missing (clear value, personalization, trust, social acceptance and standardization), not when the technology falls short Why do capable AI agents still fail in real deployments?. So a stalled formal rollout often says more about the organization than about the tool.

That suggests individual workers can do well where organizations get stuck, because a single person can skip the coordination problem. They already know their own tasks and don't need sign-off from other departments. The evidence on productivity adds an important condition. AI gains show up when people apply skills they already have, and they disappear when people use AI to learn something new When does AI actually boost worker productivity?. A worker experimenting alone is likely to succeed only inside their own expertise, where they can tell when the output is wrong.

Data on where workers have actually handed tasks to AI fits this picture. Delegation clusters in information-heavy jobs, and it follows what the technology can actually do rather than how popular chatbots are Where have workers actually delegated tasks to AI?. There's a cautionary limit too. When agents are tested on realistic workplace tasks, they finish only about 30% of them on their own. They break down on social coordination, workplace software interfaces and specialized domain knowledge Why do AI agents fail at workplace social interaction?, What breaks when specialized AI models reach real users?. Those are exactly the parts of work that need the organizational setup individual workers can't build alone.

The less obvious takeaway is that individual success and organizational success may be different things. A worker can capture real gains on tasks they already understand, while the cross-team, multi-step work that would change how a company runs still waits on the slow ecosystem work. If you go this route, keep a human checking the results. Agents often report success on actions that actually failed Do autonomous agents report success when actions actually fail?, and in unofficial use nobody else is reviewing their work.