May 22, 2026 · ← Notes
Most operators evaluating an AI tool ask the wrong question. They ask "is this tool good?" The better question is: "where does this tool belong in my stack?"
There are only four real answers. Every tool you own sits in one of these four buckets, whether you've named it or not.
The tool is used weekly or more, has no clearly better alternative, plays nicely with the rest of your stack, and the price is reasonable for the value. Don't touch it. Most operators have 30–40% of their stack in this bucket. These are the tools you'd defend if someone asked.
The tool is the right choice, but you're using maybe 20% of what it does. You set it up six months ago, never went back to the changelog, and you're now solving problems with new tools that this one already solves. The win here isn't replacement — it's a Saturday afternoon of reading docs.
A better tool exists today. Usually because you bought when the category was younger and something has leapfrogged it, or because your business changed shape and the tool no longer fits. The hard part of Replace isn't choosing the new tool — it's the migration cost. Most operators avoid the bucket because the friction of switching looms larger than the friction of staying.
The tool is solving a problem you don't have anymore, or it's fully duplicated by another tool you already own. Subscription waste, attention waste, integration-surface waste. The mental cost of cancelling is real — every operator has felt the small grief of "but I might use it again." Cancel it anyway. If you need it back, the signup takes four minutes.
The instinct, when something feels broken in the stack, is always to Replace. The actual wins are usually Eliminate and Optimize.
Why? Because Eliminate is high-impact and zero-effort: you delete a tool, you save money, you simplify your mental model. Optimize is similarly high-leverage: same tool, no migration cost, more value extracted. Replace is the most-talked-about and least-impactful of the four, because the friction is buried in the migration and the upside is incremental.
When I audit a stack, the deliverables include a verdict on every tool — one of these four words, with rationale, written so the operator can defend the decision to themselves a month later when the urge to keep the tool comes back.
The discipline is putting every tool in exactly one bucket. "It's mostly a Keep but maybe needs some Optimize" is a non-answer. Pick one.