I know we’re far from AGI, but the latest models are superb at handling long-running, complex work and coordinating and delivering it. I have Claude Opus 5.5 running on a job for 3 days straight right now, and it’s nailing it.
We are way past the point where we could just do it ourselves with tons of people and lots of human effort. I love it.
Sol 6.1 and Astra are equally good. Modding Assetto Corsa…
Another example that showed me we’ve passed the tipping point is asking AI to mod Assetto Corsa with car and track packs.
Dan and I race every Thursday evening, and he modded AC. It’s complicated and a lot of work. His AI sent me the build out a pack with all the mods.
I pointed OpenAI Astra at it on my PC and asked: “Add all these mods and test it until it’s working.”
It ran for four hours straight, opening AC, selecting the car, selecting each track, driving a test lap, tweaking the track and settings, and everything it needed to.
144 cars and tracks later, it was done.
While I’m sure a human could do this, I’m also not so sure they’d get it right. This would take months of effort and frustration.
I’m convinced we’ve passed the point where you can just throw humans at a problem and get close to the same results. The complexity it can manage over time in a multi-threaded way is off the charts.
If you’re only using AI to email and Google stuff, think bigger. You’re asking Einstein to unsubscribe from emails and buy groceries. Try harder - amazing power is in your hands.
And if you rolled your eyes, use the best model you can afford. Copilot and low-cost models don't reflect the power we wield right now.