Sprawl of the kind Evans describes accumulates the way filing accumulates. Each vertical SaaS application, each script, each 10 meg spreadsheet running a department arrived because someone had a problem and a budget, and stayed because retiring it was nobody's job [2]. Evans notes that after all that accumulation the company is still full of boring, repetitive tasks [4], which is the shape you would expect if the estate were a residue of past decisions rather than a solution to present work. Cheaper production acts on the wrong variable.
Put the essay's two numbers on one clock. The joke version has an engineer spending an hour to automate ten minutes of work, and the AI version makes the same tool in five minutes with no engineer and no code, a twelvefold cut in build time [5][1]. Against that sits eighteen months for the purchase decision a cross-departmental workflow requires [12]. Eighteen months is roughly 777,600 minutes, so the generation step is about six ten-thousandths of one percent of the cycle [2]. Compressing it to zero moves the total by nothing a CFO could measure.
One answer to the assent problem is the forward-deployed engineer: a builder who walks a law firm or an architecture office and sees the automatable task the professional does not [8]. That addresses a real gap, since the great matrimonial lawyer spends the day on cases and clients, not on what discovery software might do [7]. It works one floor at a time with a person in the room, but it leaves the assent problem untouched. It also understates discovery: most of what got automated in recent decades was not obvious, and the fix usually meant redefining or unbundling the problem, with half a dozen failed attempts standing behind each company that found the right one [9][10].
The board-deck version reads: AI compresses the estate, licence spend falls, headcount follows. It is incomplete for a bookkeeping reason. Firms frequently cannot say how much software they have, what is actually used, or what they are paying for [3], and you cannot decommission an inventory you have never taken. A five-minute tool built by someone who is not an engineer produces an asset with no owner and no line in any register [5]. Generation adds rows before it removes any.
The ten-year question stays open. Whether dynamic, generated software eventually displaces packaged applications is not settled here, and the supplied text stops mid-sentence as Evans begins laying out his spectrum from institutionalised to improvised, so how he resolves it is not on the record we have [13][14]. The decidable question this quarter is narrower: who owns a generated tool, and when it gets retired. Answering it now means a firm can still count its estate next year. Leave it unanswered, and the count arrives anyway, by way of a licence audit.
