Tim Ferriss: why AI can create more work than it removes

Meeting notes take seconds to produce, then land in ten inboxes and go unused. A weekly report gets longer, but still does not support a decision. An AI-generated content batch arrives quickly, then the team rewrites it. These are illustrative situations, not reported incidents. They raise a practical question: who saves time, and who inherits the work?

In a Q&A published on October 21, 2026, Tim Ferriss, author of The 4-Hour Workweek, challenges the assumption that faster production necessarily means less work. AI accelerates the tasks we give it. If a task should not exist, making it faster can simply create more reading, checking and requests for everyone else.

Productivity is not the speed at which a tool generates an output. The useful measure is work removed from the whole process, including corrections and coordination, at a comparable standard of quality. Before automating a task, establish why it matters, remove unnecessary steps and then test what AI improves.

Tim Ferriss’s warning: elimination comes before automation

In the official transcript of podcast episode 887, Ferriss returns to defining priorities and eliminating unnecessary activity before automating it. His wording is blunt: AI and so on simply multiplies the useless.

He describes people doing things very quickly that probably should not be done at all, using AI to create more work rather than reduce it. His observation concerns use cases he has personally seen. It is not a representative business statistic or evidence that AI reduces productivity everywhere.

That distinction makes his argument more useful than a blanket verdict on AI. It gives a project owner a specific question to ask: if we stopped producing this output, which decision, service or obligation would suffer?

Where the extra work comes from

The author’s time saving is only one part of the calculation. A recipient still has to read, interpret, verify and sometimes redo the result. If a tool makes it easy to send five alternatives instead of one considered recommendation, someone else inherits the selection work.

Two articles in Harvard Business Review examine different risks. AI Doesn’t Reduce Work—It Intensifies It, published in February 2026, describes work intensification in a qualitative study at a US technology company. Being able to do more does not automatically mean working less. Findings from that setting do not establish an effect across all organisations.

AI-Generated “Workslop” Is Destroying Productivity, published in September 2025, considers outputs that look finished but leave recipients to repair the substance. Its survey draws on US employees’ self-reported experiences. It is not an experiment measuring the net benefit of every use of AI.

Neither article establishes that all AI-generated work is useless. They point to two costs worth testing locally: additional demand created because producing more is easy, and repair work shifted onto colleagues. A fast first draft can hide both.

Three ways output volume can be mistaken for value

Meeting notes without a next action

Imagine a sales meeting followed by a detailed, accurate summary. If the notes do not identify the next commitment, its owner and its deadline, someone still has to work out what to do. A better summary may not be the answer. One confirmed action in the CRM may be enough.

The practical test in turning meeting notes into a next sales action with Twenty CRM is whether the record improves follow-through. A document that merely joins the archive has not solved that problem.

A report with no decision attached

Consider another illustrative situation: a manager receives a twenty-page report every Friday. AI adds commentary and charts. But nobody has defined which changes require action, so the report grows without making management easier.

First establish who uses it, what they decide and when they need the information. A short exception alert may then be more useful than a comprehensive report. A regulatory report is different: it may need to exist even when it rarely triggers a decision. Meeting an obligation is a legitimate purpose.

Content that is quick to generate and slow to fix

A team can increase publishing volume while spending more time checking sources, correcting claims and restoring a consistent voice. If that work exceeds the writing time saved, the advertised acceleration has not reduced the team’s workload.

Compare publishable work at a similar quality level, not an AI draft against a finished human piece. And if the content has no identified reader or purpose, not producing it remains an option worth testing.

Remove, simplify, then automate what remains

For each candidate task, the business owner can work through four decisions in order:

1. Name the purpose. Who uses the output? Which decision, service or obligation depends on it? If there is no clear answer, test removing it reversibly after checking obligations and downstream dependencies.
2. Shrink the output. Could one decision, an alert or an updated field replace a report? Remove duplicate approvals without removing essential checks.
3. Choose the appropriate tool. A simple rule or a connection between existing systems may be sufficient. Use AI where interpreting text, classifying information or preparing a response adds real value.
4. Run a bounded test. Decide who validates results, which errors are unacceptable and what would stop the test before expanding deployment.

An agent can carry out a poorly chosen task consistently. That does not make its objective useful. The Hermes Agent guide for businesses explains the distinction between what an agent can prepare or execute and what the business must define and approve.

Measure net time saved across the whole process

Before the trial, record the time spent on the existing process, its volume and the required quality level. During the trial, include preparation, production, review, corrections, exceptions and maintenance. Count the recipients’ time as well as the operator’s.

Net time saved is the total baseline time minus the total time required by the new process. Track subscriptions and integration costs separately when assessing financial value. Compare equivalent workloads at a similar quality level. If a faster report needs an additional accounting check, include that check in the calculation.

Automation can also make a useful activity more frequent: processing more customer requests, for instance. That may be commercially valuable even if working hours stay the same. Track correctly completed requests and response times in that case. The benefit is increased capacity, not necessarily time released.

At Say Digital, that distinction belongs in the brief before development starts. Identify the work to remove, prototype a limited change, check the result with users, then deploy with tests, controlled releases and ongoing measurement. If the benefit disappears once review time is included, revise the scope rather than increasing output.

Decide what should disappear before adding AI

Tim Ferriss’s point is not that businesses should abandon automation. It is that an existing task does not automatically deserve to be automated. The first useful result of an AI project may be a task removed, a shorter report or a request that no longer reaches the team.

Does your process produce plenty of material but little relief? Talk to Say Digital about what to remove, simplify or automate, and how to demonstrate a net benefit before funding a wider rollout.

Sources

• Tim Ferriss — official transcript of episode 887, October 21, 2026. The author’s observations and argument, not a representative survey.
• Aruna Ranganathan and Xingqi Maggie Ye — AI Doesn’t Reduce Work—It Intensifies It, Harvard Business Review, February 9, 2026. Qualitative research in one company, with limits to generalisation.
• Kate Niederhoffer and coauthors — AI-Generated “Workslop” Is Destroying Productivity, Harvard Business Review, September 2025. Self-reported experiences of the burden created by insufficiently developed AI outputs.