Feedback Loop Speed as Technology Selection Pressure
Updated
Knowledge on this page was mainly distilled from A Stranger Explained the Future of Programming to Me in 1981.
The technology layer with the shortest gap between having an intention and seeing a result is the one people end up living in. This is not a design principle anyone planned. It is a selection pressure that has operated consistently since at least the late 1970s, and it explains platform shifts that are otherwise hard to account for on purely technical grounds.
The Pattern Across Eras
In the early 1980s, building something out of electronic components meant an afternoon per revision: finding parts, laying them out, soldering, discovering a wrong connection, and undoing physical work to change a single decision. Programming offered a file and a run command. Change a decision, run it again, ten seconds, and the machine tells you where you stand. The gap between intention and result dropped by orders of magnitude, and people migrated accordingly.
VisiCalc appeared on the Apple II in 1979 and turned a generation of accountants into programmers who never used the word. Changing one number recalculated every dependent number instantly. The spreadsheet's feedback loop was so tight that people who would never have called themselves technical adopted it without hesitation.
The interactive prompt is older than the spreadsheet: you type one line and the language answers before you have finished the thought, instead of requiring you to compile a whole program to discover you were wrong. Hot reload and the browser console continued the same compression, letting developers edit a live page and watch it rearrange under their hands.
Bret Victor built a talk in 2012 around the principle that creators need an immediate connection to what they are making, and demonstrated what a coding environment feels like when the feedback gap goes to zero.
Where the Line Currently Ends
Vibe coding, named by Andrej Karpathy in early 2025, is the current terminus. You describe what you want in a few sentences, wait roughly ninety seconds, and a working interface appears with routing, a database, and a login form you never asked for but would have needed. The feedback loop is now so short that the gap between intention and result has become imperceptible.
The pattern is consistent but the implication is not purely optimistic. Each jump shortens the distance to a visible result without necessarily shortening the distance to a correct one. The selection pressure rewards speed of feedback, not accuracy of feedback, which means the layer people migrate to is not always the layer that tells them the truth.
Q&A
Who first articulated this pattern?
In the article's telling, an unnamed stranger in a cinema in 1981 captured it in one sentence: 'I like programming better because you get immediate feedback.' He thought he was defending a hobby, but he was describing the selection pressure that determines which technology layer people adopt.
Does this mean slower-feedback tools die out?
No. The underlying layers persist because someone still has to maintain them. Electrical engineering did not vanish when programming became dominant. What changes is who has to think about the lower layer. It becomes invisible infrastructure maintained by specialists rather than the default surface everyone works on.
Is vibe coding the final step in this progression?
There is no structural reason to expect a final step. Every previous layer seemed like the last compression until a new one arrived. The pattern predicts that if something offers an even shorter feedback loop than natural-language prompting, people will migrate to it.
How does VisiCalc illustrate the principle?
VisiCalc gave non-programmers an instant feedback loop: change a number, see every dependent number update immediately. The loop was tight enough that accountants adopted it without needing to learn programming concepts. The tool's feedback speed, not its technical sophistication, drove adoption.
What is the downside of optimizing for feedback speed?
The selection pressure rewards seeing a result quickly, not verifying the result is correct. Each layer that wins on feedback speed tends to widen the gap between how fast you can see something and how fast you can confirm it works. The validation problem does not migrate upward with the abstraction.