 |
| AI-generated image in the style of the "this is fine" meme but instead of the dog, have a professional blissfully typing away at a laptop computer. |
I had another insight in my quest to realize material productivity gains from AI and not just fill the time with more work.
When you're looking at your to-do list and an LLM can knock out that Python script in 30 seconds, it's hard to resist. The code appears like magic, and it feels productive to generate it. Because LLMs work so quickly, it's easy to stop asking about whether the work is worth doing at all.
Working on the right thing is what matters. Speed doesn't count when the task doesn't matter. I sometimes find myself asking ChatGPT or Claude questions and then not bothering to read the responses thoroughly, like Amazon packages piling up inside your front door that you ordered but don't care enough to actually open. Before typing the next prompt, ask yourself:
- Can I clearly draw a line from this to our deliverable?
- Will this change a behavior or decision?
This is no different from how you should have operated before LLMs became powerful. You just need more discipline about prioritizing because it's tempting to knock out random tasks quickly that you would have de-prioritized before.
Another reason to catch yourself is solutions always take longer than you think because AI doesn't complete tasks to "done done". You still must take it across the finish line -- review it, discuss it, deploy it, test it, etc. As a consultant it's easy to underestimate the level of effort to do something if you are just thinking of the hands-on development time, and this is no different. The AI portion might represent a fraction of what's required to ship something useful. If you're not going to finish it, why start?
Every "quick" low-value task has an opportunity cost — that's time not spent on what actually moves the needle. The winners with AI won't be those who use it most, but those who use it on what matters.