Producing got cheap, and the interesting question changed with it. It used to be what we could afford to make; it is now what we are capable of making. A person working with agents reaches past their own craft, into work that used to need three more people or a budget nobody had, and the point of that is not a smaller invoice. Teams get smaller. What a team takes on gets bigger, and that is the trade we are interested in.
What became possible that was not
- A designer ships production code instead of handing over a mock-up for somebody else to rebuild. We publish that as designers who build, and it is this principle with evidence under it.
- A researcher tests five concepts where the budget used to buy one.
- Work that was never economic now is: an internal tool, a throwaway prototype for a single workshop, a bespoke artefact per client.
- The floor for a prototype rises. Real stack, real data shapes, accessible from the first commit, because that costs hours rather than weeks.
- One person carries a vertical slice end to end, spec to screens to deploy, without waiting on three handovers.
What we do with the time
The time saved does not disappear, it moves. It goes to more passes at the problem while being wrong is still cheap, more research reaching the spec, and more time with the thing running in front of people. Compression buys loops and the certainty they bring, not fewer days.
It also buys the things that used to get descoped first. The empty state, the error state, the assistive-technology pass: all of them were cut for time on projects where nobody wanted to cut them, and none of them takes the time it used to.
Where the time went on real projects
- Evidence arrives before the commitment. On a nine-week project in 2017, the first real customer reaction landed in week six, after five weeks of arguing about the direction. It now lands in week two, while changing course is still cheap.
- More cycles, not fewer weeks. Our largest current discovery is in its thirteenth month and has had more rounds of testing than any project in our history. Making stopped consuming the time, so the time buys iterations instead. Duration is a property of the problem.
- We test the product rather than a description of it. A concept in a deck earns an opinion about a deck. Twice on work for one drinks brand, the concept we were asked to test failed in front of customers, and the finding beside it became the product.
- The prototype is the specification. On one data product, prototypes built in production code went to the client's developers as the build specification. One handover went, and with it the translation loss where strategy usually dies.
What this asks of a person
Agents amplify somebody who knows what good is, and get somebody who does not to mediocrity faster. That makes taste and judgement the constraint, which is why the learning programme is the part of this we invest in hardest. Reaching further is a skill, and it is one we teach.