The Two-Tool Rule (Why My Stack Stopped Growing at Six)

My AI stack stopped growing at six tools, and it has stayed there for over a year while the market around it multiplied. The discipline that froze it is a rule I call the two-tool rule, and it is almost embarrassingly simple: a new tool may only join if it either replaces an existing one outright, or does a job nothing in the stack currently does. Expansion requires displacement or a vacancy. Everything else is a no.

The rule exists because my stack once held fourteen tools, and the audit that followed was not pretty. The pattern in the bloat was consistent: each addition had been justified at purchase by a real-sounding capability, but the capabilities overlapped in practice, because tools in adjacent categories increasingly do chunks of each other's jobs. My writing assistant, my note tool, and my general assistant could all summarize documents; my image tool and my design tool overlapped on half my use cases; two automation platforms ran the same three workflows in parallel for months because I never decided. Nothing in the stack was missing. Everything in it was partially redundant.

Replace-or-vacancy forces clarity that feature lists obscure. When a shiny candidate arrives now, the first question is which slot it is interviewing for, and the second is whether the incumbent actually failed. The burden of proof sits on the newcomer, where it belongs: incumbents have accumulated context, integrations, and my own learned muscle, all invisible costs that comparison pages never price. Most candidates die at this stage, not because they are bad but because my vacancy list is empty, and an excellent solution to a problem I do not have is not an addition, it is clutter.

The displacement mechanism keeps the stack honest too. A challenger that beats an incumbent on my real work wins the slot, and the loser gets cancelled, not demoted to backup, because backup tools are how sprawl begins, and the honest version of my past bloat was a museum of almost-good-enough backups. Displacement also dates the stack naturally: tools leave when they stop being the best answer, which is the only reason a tool should ever leave.

The counterargument I hear is capability anxiety: what if I need the tool later, or what if the market makes one category essential. My answer is the vacancy clause, which handles the future honestly, if a genuinely new job appears, the stack admits a tool for it, and so far the jobs that actually appeared were served by tools I already had. The stack at six, fully audited, costs less than half of the fourteen-tool version, and the output difference is zero. The tools stopped being a collection and became a team, and like any good team, it stays small enough that everyone in it has a job.