The Unintended Inefficiencies of Giving Everyone AI

A new normal is starting to develop in corporate America, in legal departments and compliance offices across every industry. It goes something like this:

A business stakeholder thinks of a brilliant new way of doing something, asks Copilot if it's a good idea, get's the standard sycophantic response, and starts building. A few weeks later someone else hears about it, and points out that maybe someone from compliance or legal should weigh in. "Don't worry, I already ran it past Copilot" the brilliant businessperson responds. When the compliance or legal expert finally hears about, the plan's already halfway to launch, and now the expert is put in a difficult position-- stop the train, or get on board.

Enterprise AI adoption was supposed to make organizations more efficient. And in many ways, it has. Drafting emails faster, summarizing documents, generating first-pass code — the productivity wins are real. But there's a cost hiding in the margins that nobody put in the ROI presentation: the slow, grinding erosion of expert bandwidth as legal, compliance, and regulatory professionals are increasingly pulled into exhausting arguments they should never have to have.

When the Wrong Answer Arrives with Confidence

The core problem isn't that AI tools give bad answers. It's that they give bad answers convincingly. A general-purpose assistant like Microsoft Copilot or ChatGPT is trained to sound authoritative. It presents nuanced, jurisdiction-specific, context-dependent regulatory questions the same way it presents a recipe for banana bread — cleanly, confidently, and without any of the appropriate hedges a human expert would instinctively include.

So when a product manager asks Copilot whether a new data-sharing feature requires a DPIA under GDPR, they get a tidy paragraph that sounds like an answer. It may be technically half-right, completely wrong for their specific situation, or missing three relevant variables entirely. But it reads like an answer. And that's enough.

The business stakeholder arrives at the expert's door not with a question, but with a conclusion. The expert is no longer being consulted — they're being audited.

The New Burden: Arguing Down Overconfidence

What used to be a fifteen-minute consultation has become a forty-five-minute negotiation. The expert now has to do two jobs simultaneously: provide the correct guidance and dismantle the false certainty the AI already installed.

This is a genuinely new kind of cognitive labor. It requires patience, diplomacy, and a careful resistance to the temptation of sarcasm. Because the business person isn't being malicious — they genuinely believe they did their homework. The AI told them something. They read it. It made sense to them. And now this expert, who should be an ally, is making things complicated.

"Why does it have to be so complex? The AI explained it in two sentences."

That sentence — or some version of it — is becoming a defining feature of expert life in the AI enterprise era. Complexity, in the eyes of the newly AI-informed layperson, has become a sign of inefficiency rather than rigor.

The Hidden Tax on Institutional Knowledge

What's getting lost in this dynamic isn't just time. It's trust, and eventually, talent.

Regulatory and legal experts are among the most expensive, hardest-to-replace people in any organization. They carry years of domain knowledge, hard-won professional judgment, and an understanding of nuance that simply cannot be compressed into a prompt response. When those people spend their days playing defense against Copilot printouts, the organization isn't saving money. It's burning it in a less visible place.

And the best experts notice. They start to feel like their expertise is a liability — something to be challenged rather than valued. Some will quietly disengage. Some will leave. What replaces them? Probably more AI queries, cycling the problem forward.

The Fix Isn't Banning AI — It's Rebuilding Trust in Expertise

Organizations deploying enterprise AI need to reckon with a cultural side effect nobody planned for: they've accidentally taught employees to trust machines over people. The answer isn't to roll back AI adoption, but to be deliberate about where AI ends and human judgment begins.

That means communicating clearly — and repeatedly — that AI tools are starting points, not authorities. It means creating clear escalation norms so experts aren't relitigating AI outputs from scratch every time. And it means recognizing that the fastest answer isn't always the right one, no matter how confident it sounds.

The expert in the room is still the expert. It would be worth listening to them again.