1. Scene hook
Last night I was still at my computer, tweaking a plan line by line, and the more I changed, the messier it got.
It’s only in the past couple of years that I’ve slowly figured this out: a lot of getting stuck wasn’t because I wasn’t working hard enough. It was because I was stuck in manual trial and error, without leaving behind a reusable record of the process. I’ve made this mistake before too—finish one project, then start the next one from zero all over again.
2. What this is + who’s already using it
What stayed with me most here wasn’t some new model. It was the idea: Lila Sciences wants to run a lab like a data center, with robots and equipment running experiments 24 hours a day while AI directs the work and keeps accumulating validated data.
I could picture it right away: Andy Beam and Rafa Gómez-Bombarelli standing in a warehouse-like lab, watching sample plates move along tracks, like they’re looking at an assembly line that never clocks out. We obviously don’t need to copy that literally, but the idea hits hard for those of us doing side projects, consulting, or content services too: if nothing gets captured from each delivery, the next round will still be exhausting.
3. What it would cost to replicate today
If we really wanted to recreate this kind of AI + automated lab, the money would probably start at 1 million RMB, the timeline would be at least 6–12 months, and the technical barrier would be high. This isn’t something we solve by installing one piece of software—we’d have to connect equipment, workflows, and data into one system. If all we want right now is to borrow the idea, I’d start by opening the spreadsheet or project tool I use most and adding a new column: “What did this round of trial and error teach us?”
Not everyone needs a tool like this, and it’s fine if we’re not trying it now. For most small teams, what’s more worth copying is the method behind it: standardize repeated actions, save the results, and make the next round depend less on improvisation.
4. Advice by stage
If I were just getting started, I’d focus on one high-frequency action first—pricing, writing proposals, sending follow-ups—and write the steps down. If I already had 1–2 clients, I’d start organizing which messaging, workflows, and delivery templates save the most time. If I were scaling, I’d build my own data loop earlier: who does the work, how it gets recorded, and where things tend to slip through. We may not need to build a futuristic lab, but at the very least, we don’t have to keep starting from zero every single time.