Why We Trusted AI With Everything Except Judgment

Briefly, AI — brought to you by Harry Sharman. Well, a version of him. The irony isn't lost on me either.

We've just invited AI agents into our workspaces. Not as assistants — as colleagues. And the strangest part? We're more comfortable letting them write code than decide what to write.

Right, so this week Notion launched something genuinely new. Not another chatbot. A developer platform that lets AI agents plug directly into your workspace — agents that pull data from other tools, run code, operate alongside your documents. They're calling it a multi-agent hub. The pitch is simple: instead of bundling one AI into everything like Microsoft or Google, Notion says bring your own. Pick the agents you trust for the tasks that matter.

And on paper, it's clever. It positions Notion against the giants without needing to build the models themselves. But here's what caught my attention. The agents they're showcasing aren't decision-making tools. They're execution tools. An agent that formats your database. Another that generates a first draft. One that pulls sales data into a report template. Notice what's missing? Judgment. Priority. Strategy. The actual hard bits.

Now, there's a reason for that. And it's not technical. Notion could absolutely build an agent that decides which tasks to prioritise, or recommends which project to fund, or tells you which meeting to skip. The models are capable. But nobody wants that. Or more precisely — we don't trust it yet. Even the people building these tools know we're not ready.

There's research that speaks to this directly. A study on white-collar AI resistance in 2026 found that the primary barrier isn't technological illiteracy — it's professional identity threat. When AI touches the work that defines who you are — the parts that require judgment, taste, relationships — people push back. Not because the AI is bad at it. Because letting it do that work feels like giving away the part of your job that makes you *you*.

And here's the thing. We're fine with AI doing the tedious bits. The formatting. The data-pulling. The transcription. Those tasks are annoying, and nobody's professional identity is wrapped up in being really good at exporting a CSV. But the moment AI crosses into judgment — which client to prioritise, which argument to make, which person to trust — we recoil. Even if the AI is right more often than we are.

Notion's bet, whether they'd say it this loudly or not, is that we'll adopt AI fastest if it automates execution but leaves judgment to us. That's the psychological sweet spot. You feel augmented, not replaced. You're still the one making the call. The AI is just faster at carrying it out.

But look six months ahead. Those execution agents are going to get very good. Fast enough, accurate enough, reliable enough that the time you spend checking their work starts to feel inefficient. And at some point — maybe not this year, maybe not next — someone's going to build an agent that makes the judgment calls too. Prioritises your tasks. Recommends which email to send first. Suggests which project to fund based on risk and return.

And the question we're going to face, whether we're ready or not, is this: if the AI's judgment is statistically better than ours, and we know it's statistically better, what exactly is our job?

Because right now, we're in this odd middle phase. We've accepted that AI can do the boring bits. We haven't accepted that it might be better at the interesting bits. And I think part of that is because the boring bits were never why we showed up. Nobody became a product manager to format slides. They became a product manager to make calls. To prioritise. To say no to the wrong thing and yes to the right thing. That's the identity bit.

Notion's platform is designed to live in that gap. Agents that make you faster without threatening what you think your job actually is. But the gap is temporary. The agents are learning. And at some point, we're going to have to decide whether we want to be right, or whether we want to be the one who decided.

Here's the question I'll leave you with. If an AI agent makes better calls than you do — not occasionally, but consistently — and you know it does, would you let it decide? Or would you override it anyway, just to feel like it was still your call to make?

That's your lot. See you Monday.