Every automation tool makes the same promise: save time, do less, focus on what matters. And at the level of a single task, the promise usually holds. An agent drafts the blog post. A workflow files the receipt. A bot triages the inbox. Each one shaves minutes, sometimes hours, off a job that used to be manual. But zoom out, and something strange happens. The total amount of work doesn't shrink. It grows. You didn't buy back your afternoon. You filled it with new workflows, new dashboards, new agents to debug. The shape of the work changed, but the volume didn't. This isn't a failure of automation. It's a pattern, and it has a name.

The Jevons paradox, applied to your to-do list

In 1865, the economist William Stanley Jevons observed something counterintuitive about coal. As steam engines became more efficient, you'd expect coal consumption to drop. Instead, it surged. Cheaper energy per unit made it economical to use coal in more places, for more purposes, at greater scale. Efficiency didn't reduce demand. It unlocked it. The same dynamic plays out with productivity tools. When you automate a task, you reduce its cost, the time and effort per unit. But cheaper tasks don't disappear. They multiply. Automating blog publishing doesn't mean you publish the same number of posts with less effort. It means you publish more posts. Automating job applications doesn't mean you apply to the same roles faster. It means you apply to more roles. Automating reporting doesn't mean fewer reports. It means reports for everything. I run 13 Notion agents. I have a blog pipeline, a job applications pipeline, automated summaries, and templated workflows for half my recurring tasks. Each one saves time per unit. But the total surface area of work I manage has expanded well beyond what I handled before any of them existed. The per-task cost went down. The task count went up. Jevons would not be surprised.

The automation tax

There's a hidden cost to every automated workflow, a kind of tax you pay in perpetuity once the system is live. Every automation needs monitoring. Did the trigger fire? Did the agent produce the right output? Is the integration still connected? Every automated workflow needs error handling, because the edge cases that were easy to catch manually become silent failures in an automated system. And every automated workflow needs maintenance, because the tools change, the APIs update, the context shifts, and the thing that worked last month quietly breaks. This is the automation tax: the ongoing overhead of keeping your automations running. It's not a one-time setup cost. It's a recurring line item on your time budget, and most people don't account for it when they decide to automate something. A 2026 ActivTrak study found that after adopting AI tools, workers spent more time on nearly every category of work, not less. Time spent on email increased by 104%. Messaging climbed by 145%. Business management tool usage rose by 94%. The tools were working. The workload was growing. Harvard Business Review captured it well: "AI doesn't reduce work, it intensifies it." The promise is that AI handles the grunt work so you can focus on high-value tasks. The reality is that the grunt work expands to fill the capacity you freed up, and now you're also managing the AI.

Each layer adds

Email was supposed to reduce meetings. It didn't. It created a new channel that generated its own volume, its own norms, its own overhead. Then Slack was supposed to reduce email. It didn't. It created yet another layer, faster and more fragmented. Now AI is supposed to reduce Slack. Early signs suggest it's adding to the pile instead. Each new communication or productivity tool promises to replace the one before it. But replacement almost never happens. What happens is accumulation. You end up monitoring email, Slack, AI summaries, dashboards, and notification feeds, each one demanding a slice of your attention. The meta-work of managing these layers becomes a job in itself. The pattern holds beyond communication. Project management tools were supposed to bring clarity. Instead, many teams now manage the project management tool as its own project. Automation platforms were supposed to eliminate repetitive work. Instead, teams hire automation specialists to manage the automations. The abstraction rises, but the work follows it up.

Automation moves the bottleneck

Systems thinking offers a useful frame here. The Theory of Constraints, developed by Eliyahu Goldratt, says that every system has exactly one bottleneck that limits its throughput. Fix that bottleneck, and the constraint doesn't vanish. It moves somewhere else. Automation does the same thing. When you automate the writing of a report, the bottleneck shifts from drafting to reviewing. When you automate data collection, the bottleneck shifts from gathering to interpreting. When you automate deployment, the bottleneck shifts from shipping to monitoring. The work doesn't leave the system. It migrates to the next weakest point. This is why automating a single task often feels productive while automating an entire workflow often feels like trading one set of problems for another. You solved the visible bottleneck and surfaced a hidden one. The throughput of the system didn't change as much as you expected, because the constraint was never just the task you automated. Federal Reserve research found that workers spend about 5.7% of their time using AI but save only 1.6% of their total work time, a net efficiency of roughly 28 cents on the dollar. Those savings don't accumulate into free hours. They get absorbed by expanded scope, additional review cycles, and the unchanged processes surrounding the automated task. If an analyst can draft a report in two hours instead of eight, but the approval chain still takes three weeks, those six saved hours are invisible.

The case for deletion over automation

If automation moves work around without reducing it, the real leverage isn't in making tasks cheaper. It's in eliminating them entirely. Deletion is underrated as a productivity strategy. Most people, when faced with a tedious task, ask "how can I automate this?" Fewer ask "what would happen if I just stopped doing this?" The second question is harder because it forces you to confront whether the task matters at all. But it's the only question that can actually reduce your total workload. Automating a weekly status report saves you 30 minutes a week. Deciding that nobody reads the weekly status report and killing it saves you 30 minutes a week and eliminates the automation tax. The first approach optimizes. The second approach simplifies. Over time, simplification compounds in a way that optimization doesn't. This isn't an argument against automation. That would be absurd, especially from someone running a fleet of agents to manage a blog. But it's an argument for being honest about the tradeoff. Automation doesn't give you time back for free. It gives you time back on one thing and quietly creates obligations on others. The net gain is often smaller than the per-task saving suggests.

What to do about it

Before automating, ask: does this task need to exist? If the answer is no, delete it. If the answer is yes, ask: what new work will this automation create? Account for the monitoring, the error handling, the maintenance, the expanded scope. If the total cost is still lower, automate. Periodically audit your automations the way you'd audit your subscriptions. Which ones are still earning their keep? Which ones have you been maintaining out of inertia? Which ones created more work than they saved? And resist the pull of automating everything just because you can. The goal isn't to have the most automated workflow. The goal is to have the least work. Sometimes those are the same thing. Often, they're not. We didn't automate ourselves into more free time. We automated ourselves into more work. The tools aren't broken. The incentives are just designed that way, and the only real countermove is to want less, not to optimize more.

References