Daniel Lemire

For years, my employer has required professors to fill out complicated Excel spreadsheets every year. After all, someone has to keep track of who teaches what and who is responsible for the students in each course. These Excel spreadsheets are incredibly complex. We use a wide range of metrics to measure each professor’s workload in the name of “fairness.” There are multiple spreadsheets, each containing several tabs. Why Excel? Probably because someone on the IT staff was (and still is) very good at it. Once we’ve all completed our forms, they are emailed to administrative staff, who then manually re-enter the data into the actual database. This year, we received the new forms on Thursday. By Friday morning, I had launched my AI with a simple prompt: “Take these spreadsheets and build a web app that does the same thing.” Within an hour, I had a fully functioning web application. It is far superior to the Excel version: it includes autosuggestions, autosave, the ability to synthesize data across all sections, and a much cleaner interface. That same day, I called a meeting with the department chair. He immediately agreed that this was the way forward. Over the weekend, I created a short video presentation and sent a link to the live, working app. Upper management seems genuinely pleased. Their only request was that the new system must still be able to generate the exact same Excel spreadsheets they are used to receiving. This morning, I asked my AI to add an export button that produces those identical files. A few minutes later, I had a robust function that generates spreadsheets virtually indistinguishable from the originals. Are there bugs? Of course. But historically, the Excel spreadsheets had their own bugs and quirks. None of the issues in the new app are critical or could cause real damage. In the worst-case scenario, data could be lost or corrupted—yet the old Excel process was hardly safer, given how easy it is to accidentally overwrite or break formulas. “But Daniel, aren’t you a senior professor? Why are you building web apps?” Because I want to see, empirically and in the real world, what AI can actually do. We have several meetings this week. I’ll let you know how it turns out.

How many productive hours can you actually work in a day? Time spent is almost always irrelevant, or at best secondary. Your brain is like a laptop constantly slipping into energy-saving mode. It’s easy to “work” for hours without ever fully engaging your mind in a productive way. Even when you do manage to focus, the output is highly uneven: you probably do your best work in short spikes of a few minutes. This means it’s entirely possible for someone to outwork you in a fraction of the time. Staying focused and energized is difficult. You can’t simply will it into existence. Does that mean you should cut your working hours? Not quite. The problem is that you can’t just sit around waiting for your brain to “turn on.” Nobody ever became a highly productive engineer by watching Netflix. What looks like unproductive time might simply be the necessary groundwork, the seeds, for later bursts of productivity. To answer my own question: you are probably only capable of being fully engaged for a few minutes per day, but it takes a lot of time and effort to create the conditions that allow those productive minutes to emerge.