AI doesn't need to be in everything — EverGantt48px preference.<br>(No SVG favicon on purpose: an embedded-raster SVG won't render in tabs<br>and a pure-vector one was inconsistent — square keeps Google correct.) -->

← All posts · Perspectives<br>AI doesn't need to be in everything

Published June 11, 2026<br>I want to be careful about the claim here, because it’s a narrower one than you usually hear. AI is clearly useful for some things, and I’m not going to pretend otherwise. My problem is the default. Somewhere in the last two years, every piece of software you open decided it needed an assistant, and nobody stopped to ask whether you wanted one. AI doesn’t need to be in everything, and the tools that put it everywhere are paying a cost they don’t seem to be counting.

Every tool got an assistant

Project management tools converged on this faster than almost any category. Asana ships AI Studio and “AI Teammates.” Notion rebuilt itself around Agents. monday.com meters monday AI with a credit system. Atlassian put Rovo inside Jira and Trello. I keep a date-stamped map of where every major tool landed if you want the full table, but the summary is simple: among the big names, “no AI at all” has basically stopped existing.

The telling part is where the wave came from. Nobody’s support inbox was full of requests for a chatbot in their Gantt chart. The push came from the other direction: roadmaps, earnings calls, the fear of being the one vendor without an AI slide. Eight products adding the same feature in the same eighteen months looks less like eight independent conclusions about what customers need and more like a market following itself.

Features have weight, and AI is heavy

A feature isn’t free just because you can ignore it. This is the oldest lesson in software adoption, and AI features fail it harder than most.

Think about what one assistant actually adds to a tool you use forty times a day. A panel taking up screen space. A sparkle icon in every text field. A suggestion popup that interrupts what you were doing, which you dismiss, and which comes back tomorrow. A new admin question (what data is this thing reading, and where does it go?). And the most expensive part: output you have to check. When an AI drafts a summary or proposes a plan, someone has to read it closely enough to notice what it got wrong. The review work didn’t exist before the assistant did, so a chunk of the time it “saves” gets spent supervising it.

None of these costs shows up in a demo. All of them show up in week six, which is when adoption quietly dies. The pattern is well known by now: the moment a tool feels heavier than the work it manages, people stop updating it, the plan drifts out of date, and the team ends up coordinating in a spreadsheet while the official tool collects dust. Small teams already use about five features. Stacking an assistant on top of the ninety-five they don’t use makes the math worse, not better.

The noise compounds, too. One suggestion popup is a minor annoyance. The same popup across your PM tool, your email, your docs, your IDE, and your notes app is a tax on attention that you pay all day, in tiny amounts, without ever getting an invoice.

There’s a name for the result now: AI fatigue. It gets described as a backlash against AI, but I don’t think that’s quite right. Most people experiencing it aren’t against the technology. They’re worn out by the delivery mechanism, the assistant that arrives uninvited in every update, in every app, all at once. You can be perfectly happy using a model on purpose and still be exhausted by thirty of them volunteering.

The buyers quietly opting out

There’s a segment of buyers for whom this is sharper than annoyance, and it keeps showing up in agency and PM communities: agencies and studios that work with artists, illustrators, and other creative clients. For them, “does this have AI?” isn’t a preference question. Some client contracts now spell out that no generative AI touches the work, and a studio that signs one needs to be able to say, plainly, what every tool in its pipeline does. An AI feature they never use is still a thing they have to explain.

And the off-switch only half exists. Asana, Jira, and Trello let an admin disable their AI, which is genuinely better than nothing. But in tools like monday.com, ClickUp, and Notion, the AI is part of the product, metered and priced in, whether your contracts allow it or not. For a studio that has to certify its pipeline, “you can mostly ignore it” is not the same answer as “it isn’t there.”

You don’t have to be an agency to recognize the instinct. It’s the same one behind asking whether your project data trains someone’s model, or whether the meeting bot is recording. People aren’t anti-software. They’re tired of software doing things they didn’t ask it to do.

Where AI earns its place

The honest part of the argument: there are jobs where a model genuinely helps. Summarizing a forty-message thread. Drafting...