Most conversations about AI fall into one of two camps. Either AI is going to do everything for us, or it's overhyped and can't be trusted. Both miss the point.
The real story is messier and more interesting. AI is genuinely powerful, but only when someone knows how to fly the plane.
The Autopilot Myth
We love the idea of autopilot. Set it and forget it. Hand off the work and move on to something else. The promise of AI often gets framed this way: describe what you want, press a button, receive a polished deliverable.
That's not how it works. At least not yet, and probably not for a while.
Here's what actually happened this week. I needed to create a proposal for a new client engagement. Not a simple email, but a comprehensive package: workshop curriculum, supporting research, pricing options, and a shareable deliverable the client could navigate and share with colleagues.
I used AI for all of it, specifically Claude's Cowork mode. And it took real work.
What the Process Actually Looked Like
The session started with context. I explained the client, the relationship history, and what we'd discussed. The AI helped me structure a three-part workshop series, but I had to push back on the first draft. It was too formal, too much like a traditional proposal. We iterated.
Then came the research. I needed statistics on AI adoption, failure rates, and the skills gap to make the case for why this engagement matters. The AI searched, synthesized, and compiled. But I had to validate sources, cut the ones that felt weak, and decide which stats actually landed for a private equity audience.
Pricing was next. I pulled in context from existing client engagements to calibrate rates. The AI helped me build out time estimates for each component, but I made the judgment calls about what felt right for this relationship and this scope.
Along the way, we created local files: a master overview document, research compilation, pricing breakdown, outreach drafts for email, Slack, and text. Each one went through rounds of refinement. The AI wrote, I edited. The AI suggested, I decided.
Finally, we needed a delivery mechanism. Something the client could explore, not just read. We built it in Notion: a main page with quick summary, linked subpages for program details, research, and pricing. Every stat on the research page links to its source.
The whole thing took a few hours. Without AI, it would have taken a day or more. But it wasn't hands-off. It was deeply collaborative.
The Pilot Matters
Here's the uncomfortable truth for anyone hoping AI will just handle things: the quality of the output depends entirely on the quality of the operator.
Every step required judgment. What tone fits this client relationship? Which statistics actually matter for this audience? How formal should the pricing section be? Where does the narrative need to breathe, and where does it need to move faster?
The AI didn't know any of that. I did. And the collaboration only worked because I could articulate what I wanted, recognize when it was wrong, and guide the iteration toward something good.
This isn't a knock on AI. It's a recognition of where we actually are. The technology is remarkable. But it's a tool, and tools require skill to use well.
You have to know how to fly the plane to put it on autopilot.
What This Means for Organizations
If individual AI use requires operator skill, organizational AI adoption requires it at scale. This is why 95% of enterprise AI projects fail to show ROI. It's not the technology. It's the gap between having tools and knowing how to use them.
The organizations seeing results aren't the ones with the fanciest AI investments. They're the ones investing in their people: building AI literacy, creating space for experimentation, and developing the judgment that turns AI capability into actual outcomes.
This is the work that matters now. Not finding better AI, but becoming better operators.
The Deliverable
At the end of that working session, the client received a link to an interactive Notion site. Clean, navigable, professional. Every claim backed by linked sources. Easy to share with colleagues.
The client doesn't know or care that AI helped create it. They see a thoughtful, well-researched proposal that clearly addresses their needs. That's the point.
AI didn't replace the work. It amplified what was possible in the time available. But only because someone was flying the plane.