The hidden cost of switching between tools is the answer stitched by hand.

At half past seven, the regional operations manager at a grocery chain gets asked why the chilled delivery missed three stores. The chain is made up, and the morning is an ordinary one. To answer, the manager opens the warehouse system for when the trailer was loaded, the transport tracker for where it is now, the store ordering screen to see what was actually ordered and, last of all, the drivers' WhatsApp group, which is usually where the reason turns up. Each of those is right about its own slice, and none of them was built to know the others exist.

Usually it is the switching that gets counted when people talk about this, a login here and a minute of finding your place there. That part is real and small. The hidden cost of switching between tools is in what the person assembles at the end, since they have become the only place where those four systems ever meet.

Somebody stitches the answer together

OCTYN found the expensive version of this while building Brain, the planning system it runs its own portfolio on. The Brain write up says a missing answer costs little and a blended one costs a great deal, two things merged into one reply and delivered fluently. For the grocery chain, the blend might be the loading time for store twelve read against the route for store fourteen.

Or it might be a driver's message from yesterday read as this morning's. The write up has a line about that too: a fact true in July and recalled in September with no sense of its age is worse than no memory at all. A message in a group chat rarely says which delivery it meant.

What waits in the system nobody opened

There is a second cost, and it never shows up as time spent. OCTYN runs around eighteen coding sessions at once across twenty one codebases, as its work page recorded on 14 September 2026, and finding the one that had stopped to ask a question meant checking every window. The Cockpit case study puts the cost as an hour of nothing happening in a place nobody was looking.

An operation with a dozen systems has quiet corners like that. Say the supplier portal marks Thursday's dairy order short on Monday afternoon, and nobody logs into the portal until Thursday.

Fewer places to look

A search over everything is the obvious fix, and the Brain write up calls it the wrong build, partly because loading the whole workspace to answer one question is the expensive habit the system is supposed to cure. Brain reads the cheapest source first and stops once it has enough. Cockpit was held to the rule that watching has to cost less than the thing being watched, which is a fair test for anything new the ops office gets asked to keep an eye on.

For the grocery chain, that means an operations system reading the systems the work already writes into and keeping one record per delivery, each line showing where it came from and when. The manager still decides whether to send a second van, and by then the record already says which trailer is which.

OCTYN has no figure for what switching costs in minutes, its own or anyone else's. Cockpit has no timing instrumentation yet, so even the waiting it exists to reduce is unmeasured.

In the made up version, it was a flat tyre, and the drivers knew by six.