Tool Churn and Switching Cost

A new tool arrives every few weeks and each one is plausibly better than what you have. The switching cost looks like an afternoon of setup, which makes trying it feel obviously worthwhile.

The setup is not the cost. Four other things are, and none of them appears on the day you switch. A related workplace concept is covered in this article, which is useful context for this argument.

What switching actually costs

Recalibration. Whatever sense you had of what this tool is good at, and what it gets wrong, was built over months and does not transfer. You are back to not knowing which tasks to hand it, which is the knowledge that determines whether it helps at all.

Your reference class. Estimating runs on comparison with your own past. Change the workflow and the recent past stops being a guide. Switch every few months and you never accumulate enough comparable history to estimate from, which is a quiet and permanent tax.

Team divergence. Two people on different tools produce differently shaped work, review each other's output with different expectations, and can no longer compare estimates meaningfully. On a team of three this is significant; the cost lands on everyone, while the decision was made by one person. Harvard Business Review provides management context for productivity and workplace behavior See Harvard Business Review.

The evaluation itself. You cannot tell whether the new tool is better. Your sense of a good day is unreliable, the trial period is short, and novelty makes everything feel better for about two weeks. So the switch is usually decided by a feeling that is known to be a poor instrument.

Why it keeps happening anyway

Not irrationality. Three real pressures.

The improvements are genuine. These tools have got substantially better, and staying still is a real cost too. This is not a case where the conservative choice is obviously right.

Trying things is enjoyable, and enjoyable is a legitimate consideration in work you do every day.

The fear of being left behind is strong, well-marketed, and occasionally correct.

The answer is not to stop switching. It is to switch deliberately and to notice the cost, because the cost is currently invisible.

What deliberate looks like

Set a floor. Six months on a primary tool unless something is actually broken. Long enough to build calibration and accumulate comparable history.

Trial in parallel, not in place. Use the new thing on a slice of work while the main workflow continues. Preserves the reference class and gives you something to compare against.

Decide in advance what would justify switching, and make it specific. If it handles our test generation noticeably better across ten tasks. Written down before the trial, for the same reason as any other measurement: deciding afterwards is a different exercise.

Switch as a team or not at all, for the primary tool. Individual editors are personal preference; the thing that shapes the output is not.

Write down what you learned about the old one. Which tasks it was good at, where it failed. That knowledge is expensive and it is discarded by default.

The version that actually matters

Underneath the tool question is a more important one.

The practices that make verification cheap do not churn. Tests, types, small changes, reversibility, documented intent. They were valuable before any of this, they are more valuable now, and they transfer across every tool you will ever use.

Time spent on those compounds. Time spent evaluating tools mostly does not — it produces a decision that will be revisited in four months.

So if the honest question is where to put the next few hours, the answer is usually not on tooling, even though tooling is where the interesting news is.

What I do

Six months, one primary assistant, parallel trials on side work, and a note of what each one turned out to be good and bad at.

Not because I am confident it is optimal. Because it is the arrangement under which I can still estimate, and being able to estimate is worth more to me than any plausible gain from being on the newest thing.

The short version

• Setup is not the switching cost; recalibration, a broken reference class, team divergence and unreliable evaluation are
• Frequent switching means never accumulating enough comparable history to estimate from
• Switching decisions are usually made on a feeling known to be a poor instrument
• The pressures to switch are legitimate — the tools genuinely improve and standing still costs too
• Deliberate: a six-month floor, parallel trials, criteria written before the trial, switch as a team
• The practices that make verification cheap do not churn, and time spent on them compounds