The Spec Is the Work
5 min read · 4 principles · apply it by lunch
Morning. That answer you rewrote three times yesterday wasn’t the model’s mistake.
The last time AI work came back wrong, could you point to the line in your own request that made the wrong answer permissible? There usually is one, and it’s what you’re actually correcting the fourth time you type ‘no, more like this’.
Barry Boehm joined TRW in 1973 as director of software research and technology, and spent the rest of the decade counting where the money in a large software project actually went. He set the results down in Software Engineering Economics in 1981, and the finding that outlived the book is the cost-to-fix curve. A defect caught while the requirements are still being written costs a fraction of the same defect caught in testing, and a small fraction of it caught after delivery; on the large systems Boehm studied the gap ran to two orders of magnitude. Boehm’s own smaller projects showed a much flatter curve, and newer projects flatter still, so take the shape as the lesson and hold the multiplier loosely. The mechanism isn’t mysterious: an ambiguity is cheap to settle while nothing has been built on it, and every hour that proceeds on the ambiguous reading is an hour that gets redone. The trap is that an ambiguous request doesn’t announce itself. It runs, on whichever reading the model picked, and you meet the ambiguity in the rewrite. Day 32 taught you to brief a model as you would brief a sharp new hire. This is the harder half: deciding what done means before anyone starts.
The Spec Is the Work is Day 121 of 365.
In the Daily, each of its four principles is explained in full, with a one-page template to apply it the same day. A new framework arrives every morning at 7am for a year, across all fifteen courses of the FutureMBA curriculum.
$29 a month · cancel any time · 7-day money-back guarantee