Nick Saraev: Count the Upkeep Before Automating

The Lesson

An AI-assisted workflow can produce a draft in moments. That does not tell you how long it takes to prepare the inputs, review the result, fix exceptions or keep the workflow working after your process changes.

In his March 2021 essay The small business automation problem, Nick Saraev argues that automation shifts work into setup and continuing maintenance. He asks readers to consider whether a process may change before the initial effort is recovered. His About page identifies his work as a developer and AI educator. The essay predates today’s agent products; its general effort comparison can help us ask better questions about them.

For this practice, count time rather than promising a return. The small calculation below is an original SelfGrowthVideos example, not Saraev’s example or a prediction about a particular tool.

Suppose a fictional volunteer group prepares an activity recap. Doing it manually takes 15 minutes. An assisted version takes 6 minutes to prepare and review each recap. The possible reduction is 9 minutes per run. Over 10 runs, that is 90 minutes.

Now include 120 minutes of setup and 20 minutes of upkeep during the same period. The assisted route would use 200 minutes: 120 for setup, 60 for the ten reviewed recaps and 20 for upkeep. The manual route would use 150 minutes. Under these invented assumptions, the assisted route takes 50 minutes more during the period you chose.

That does not make the project useless. You might choose to build it to learn, improve consistency or prepare for a different workload. It does mean that “saves time” needs a defined period and an account of the work still required.

Try changing one assumption at a time. What happens if the group only needs four recaps? What if review takes longer because the source notes are incomplete? What if a template change requires rebuilding part of the workflow? Record the uncertainty rather than choosing the most flattering estimate.

Use observations when you can. Time a manual example from start to finish. Time an assisted example including input preparation and checking. Keep learning time visible instead of treating it as free. If you have not built anything, label every duration as an estimate.

Review quality alongside effort. A shorter process that loses an important fact or creates an unsupported commitment has not met the task’s purpose. Define what a usable recap must include before using the timing comparison.

The next decision can be modest: simplify the manual template, test one assisted step, postpone the build or continue as a learning project. You do not have to connect an entire business process to learn which part creates extra work.

Reflection

• Have I counted input preparation, review and exception handling?
• How many times will I actually use this workflow during the period I chose?
• Which part of the process may change?
• Is my purpose saved time, clearer output or learning—and how will I observe it?

Practice

Original SelfGrowthVideos exercise: compare two routes over one period. This is our planning exercise, not Nick Saraev’s named method.

1. Choose a low-stakes repeated task and a specific period, such as the next ten fictional practice runs.
2. List the steps in the manual route. Record observed minutes where available; label estimates.
3. List the assisted route’s setup, input preparation, review, corrections and upkeep.
4. Calculate manual total time: number of runs multiplied by manual minutes per run.
5. Calculate assisted total time: setup plus number of runs multiplied by assisted minutes per run, plus upkeep during that period. Include any review time inside the assisted per-run estimate.
6. Check the fictional recap example above: 150 manual minutes and 200 assisted minutes. Explain where each number came from.
7. Change one uncertain assumption and recalculate. Identify what you need to observe before deciding to build.

Keep a separate quality requirement, such as preserving the stated facts and identifying missing information. This time comparison does not measure every benefit or cost.

Review

After a small trial, replace one estimate with an observation. Recalculate for your chosen period and record what changed.

If the workflow creates more work than expected, investigate the specific source: unclear inputs, frequent exceptions, long review or unstable requirements. Use that finding to choose the next adjustment.

Go Deeper

• Nick Saraev’s creator profile links to his original automation teaching.
• AI Agents & Automation focuses on bounded work and review.
• AI for Business & Agency connects automation with real processes.
• Nate Herk: prove a task before delegating it helps define a checkable result before you build.