"Why didn't it work?"
I had been messing around for a few hours, and the application was functional. At least as far as I understood — back end was tight, all the front-end UI elements in place, the button press happened with the appropriately satisfying "click" — but nothing happened.
But then the realization hit me — a dim bulb light finally flickering to life after sputtering for what in retrospect felt like an eon...
"It did work."
The application isn't particularly important. I had been testing in my lab a user-facing application which was going to assess records, make a judgement call, trigger a workflow and then write up a response. My sample data was ~100 fictional records of enterprise customers who needed their telemetry and usage assessed, and action taken. The build is fairly benign — this was more about me learning multi-agent hand-offs than anything else — but it needed to show a real use case. It needed to work, though, and the instantaneous "nothing" was a shock. I immediately went to debug thought process...
Except the records had updated. The actions were taken. Each agent had fired, and the next actions were queued as designed. But this was unbelievable. As an operator, I was used to this kind of work being a multistage (probably multi-meeting) sequence of events — spreadsheets had to be updated, boxes checked, dashboards refreshed. It all just happened — and happened instantly. I built it, and yet I didn't believe it.
AI is inviting us to rethink many things, but one of them is going to be time as a measure of effort. The meetings and spreadsheets and slide decks are evidence of effort. The prettier slides, the deeper workflows — those show that the builders spent more time on them. More opportunity for revision and rethinking. This paradigm is shifting. AI agents aren't measuring time in the way we as humans think of it. The work is measured in tokens. My workflow used 4 agents across simple work on merely 100 records. This is nearly instantaneously complete to an AI. The effort in the work was in the pre-build... not in the work. The set up, not the execution.
This won't be the last time I see this lesson play out in real time. I was the builder, and I couldn't fathom how fast the execution would be. Getting others — users like salespeople or administrators or executives — to believe it will be a real challenge.
My solution? You'll laugh. I built a timer into the app. A record of the amount of time the AI thought about things and did its work. Not because the AI needed it to measure its effort — because the human interacting with it did.
If you've found this page — consider your own efforts to bring AI to your enterprise and ask yourself if the work being done by the AI is being discounted or distrusted not because it's wrong, but because of the very human judgement that the AI didn't spend enough time on the work to be good. The challenge isn't the AI — it's helping your users rewire that expectation. Show them that it completed, express the effort visually. Look at every major AI platform — Gemini, ChatGPT, Claude — they all have some variation of a spinning thinking wheel or token counter conveying to the user the vast amount of work happening in the background. Your users need to know the output you are giving them is quality — that effort was put in by the AI, even if it didn't take any time.