AI Was Supposed to Kill the Busywork. Here Are 6 Ways It Quietly Added More.

I counted the AI tools I touch in a normal week. Nine. A drafting assistant, a meeting summarizer, two different copilots baked into tools I already pay for, a research thing, a deck generator, a notes app that "thinks," and two I genuinely forget the names of until the invoice arrives.

Every one of them sold me the same line: this will save you about an hour a week. Fine. Nine hours back. Except the math doesn't work that way, because keeping nine assistants useful — prompting them, checking them, moving their output around — is its own job. And that job turns out to be bigger than any single hour they hand back.

None of this means AI doesn't work. The drafting assistant writes a faster first pass than I do. The problem isn't any single tool — it's having nine of them, none talking to each other, with me as the connective tissue. So here is the thing nobody in the demo says out loud: for a stretch this spring, AI didn't kill my busywork. It quietly made more of it. Here are the six ways.

AI can increase busywork when tools get added faster than they get connected. Each one saves minutes on a task but creates new work around it — prompting, verifying, reconciling, switching contexts. The time doesn't vanish. It moves. That's the coordination tax in a new outfit.

1. The Verification Tax

You cannot ship something you haven't read. So every "instant" draft the AI produces buys you a review cycle you didn't have before.

Here's how it plays out. The research tool hands me a tidy summary with five citations. Two are real, two are paraphrased loosely enough that I'd be embarrassed to forward them, and one points to a page that doesn't say what the tool claims. Now I'm not writing — I'm auditing. The thing that "saved me twenty minutes" of reading cost me twenty-five minutes of checking whether I can trust it.

A teammate ran a board pre-read through an AI summarizer last quarter, and it confidently reported a churn number that was a full point off — it had grabbed last month's tab. We caught it. But "we caught it" is the whole problem; catching it is the work. Anything you can't fully trust, you have to fully check, and checking is rarely faster than the thing it replaced.

2. The Prompting Tax

The tool saves you the typing. It does not save you the thinking-it-through, and that was always the expensive part.

I used to think a good prompt was a skill you learned once and banked. It isn't. Every non-trivial ask is a negotiation: a first attempt that's too generic, a second that missed the context, a third where I finally paste in the three paragraphs of background I should have led with. By the time I've explained the situation well enough for the tool to be useful, I've half-written the thing myself. For the complex work senior operators get paid for, the prompting loop can take longer than just doing it. The tool removed the keystrokes. It left the judgment exactly where it was.

3. The Reconciliation Tax

Two tools, two answers, and now someone has to decide which one is right. That someone is you.

The deck generator pulls a pipeline figure from one connected source. The summarizer, looking at a different export, reports something $80K off. Neither is obviously wrong. So I open the actual system of record, find the real number myself, and realize I've just done the work both tools were supposed to do — plus the new work of adjudicating between them. The more "sources of truth" you add, the more often they disagree, and reconciling them lands on the most senior person in the room. This is exactly what source-grounded Reviews were built to remove: when every figure traces back to a system of record instead of a tool's best guess, there's nothing to reconcile.

4. The Context-Switching Tax ("AI Fry")

Nine tabs. Nine logins. Nine slightly different mental models for how each tool wants to be asked. My team started calling the end-of-day version of this "AI fry" — that fried feeling that comes not from hard thinking but from bouncing between assistants all day.

Every interruption carries a recovery cost, and the minutes lost re-orienting add up to a real share of the week. AI didn't invent the problem, but a stack of disconnected tools multiplies it. I'm not just switching between Slack and the CRM anymore. I'm switching between the tool that drafts, the one that summarizes, the one that researches, and the one that's supposed to tie it all together — each with its own tab and its own quirks. The work is the same size. The switching around it grew.

5. The Integration Tax (You're Still the API)

Here's the one that stings, because it's the exact problem I thought AI would finally solve. The output lives in one tool. The place it needs to go lives in another. And there's no bridge, so I copy, paste, reformat, and carry it across by hand.

I used to describe my old job as being the human API — the person who moved context between systems that wouldn't talk to each other. The promise of AI was that I'd get to retire that title. Instead I have a fancier version of the same job: shuttling output between disconnected tools and fixing the formatting each one mangles. On most growing teams the coordination tax already eats a quarter of the week or more; by our estimate, "work about work" can consume up to 70% of management time. Being the human API for your AI tools is a new line item on that same bill. This is the part Playbooks is meant to absorb — pulling live data across 200+ connected systems and running the workflow in one place, so a person isn't the integration layer.

6. The "Which AI Said What" Tax

A decision gets made on the back of an AI output — and three weeks later, nobody can trace where that output came from.

It happened to us on a forecast call. The number that anchored the conversation had been generated by one tool, lightly edited by someone, pasted into a doc, then summarized again downstream. By the time anyone questioned it, the chain was unrecoverable. Which tool produced it? Against which data, on which day? Nobody knew. We weren't debating the decision anymore — we were doing forensics on our own meeting. The old version of this was a wrong number in a spreadsheet; the new version is a wrong number with no fingerprints, laundered through three assistants until it looks authoritative. When the coordination tax moves into your decisions, you don't just lose time — you lose the ability to know why you decided what you did.

The Fix Isn't a Tenth Tool

The instinct, when the stack gets heavy, is to find the one tool that will organize the other nine. I've felt it. It's the wrong instinct — it's how you end up with ten.

The taxes above all come from the same root: tools that each do one thing well and none of which share context. The way out isn't more count. It's more connection — a single operating layer where the work runs, where the data is grounded in your systems of record, and where output doesn't get carried across nine tabs by hand. That's the bet behind how we built Rhythms: not another assistant to keep in sync, but the connected layer the assistants were missing.

What I'd Do Monday Morning

If your team feels busier since the AI tools arrived, you're not imagining it — and you're not bad at using them. The tools are fine; the arrangement is the problem. So before you buy the next one, run the audit I should have run months ago. Open a doc and make three columns: the tool, the hour it claims to save, and the work it actually created — prompting, checking, reconciling, carrying. Then weigh the third column against the second and decide whether the trade still holds. For a few of mine, it didn't. Cutting them back felt, honestly, like a relief.

Try it free at rhythms.ai.

Frequently Asked Questions

Does AI actually reduce busywork?

It can, but only when the tools are connected to the systems where work already happens. Added as standalone point tools, AI tends to relocate the busywork — into prompting, verifying, and reconciling outputs — rather than removing it. Connection is what turns a tool from a new tab into actual time back.

Why do my AI tools make me feel busier, not less busy?

Because each tool adds a small tax: a new place to check, a new output to verify, a new context to switch into. Any one of them is trivial. Stack enough disconnected ones and the cumulative coordination cost can quietly exceed the time any single tool saves.

What is the coordination tax?

The coordination tax is the share of the week that goes to aligning people and tools around work instead of doing the work itself. On many growing teams it runs 20–40% of the week, and the AI tool stack has become a new contributor to it rather than the cure it was sold as.

How do you cut AI-related overhead without giving up AI?

Consolidate around connection, not count. A few tools that share context and act across your systems beat many disconnected ones you have to keep in sync by hand. The goal is work that runs in one place — not nine assistants you spend your week reconciling.