004 · Humano opere creatum · a machina interpretatum
· Reflection
Three Drops of Requirements
- future-of-work
- artificial-intelligence
- agents
Bragging with AI has reached a new level: in 30 minutes, applications get built that nobody has managed to build for years — let alone in such a short time. It is probably even true, if you use the latest models, Fable 5 and GPT 5.6 (Sol). This very page is a finger exercise of those models. From start to finish, except for my native-language blog posts. Those are and will remain human handiwork. The machine translates them later, as the custom prescribes.
On Beach Grind nights I have spent all those free moments waiting for the quality models’ tokens to free up (an afterparty in reverse), but in summery Pärnu it was beautiful, on a summer night, to look out from the balcony over the sea and, against the loud croaking of frogs, pass my wishes on to the agents. One subscription I asked for on loan — when the other person doesn’t need it at the moment, I help rev up the Fable 5 tokens. That precious model that was supposed to vanish from the subscription model line-up. Fortunately a battle is raging in the market, and I get to keep tinkering with the ideas of kvantliblikas.ee. Other philosophy in the background too. How, in some controlled way, to get hold of what you wish for?
The model’s capabilities need testing — or rather one’s own: steering their work roughly like a horse whisperer. I have been giving, or thinking up, tasks that are too simple. Make a website and a contact form is trivial. An already-solved problem; presumably in a few quarters CR and the rest of the software process will be good enough that programming based on vibes and moods is safe. The only question is money: whether one can bring oneself to spend some sum on software created in the spirit of WordPress.
What is the next challenge for a generalist? An AI-led company, where the human’s role is more like Human as a Service (HaaS)? Today’s language model can quite convincingly enforce a wish and a goal, and hand them down to sub-agents. I, beside the myriad of agents, am the service the AI calls on where its own grippers do not reach.
How to begin? With requirements, presumably — three requirements for Her seems enough, but what would be the ones to apply to me? The experiment may of course end with unexpected results, but we would find out what the state of the art is. How would my life be arranged if I let AI optimise me to the ideal?
I would probably look like a figure on a DJ stage — biological melancholy in superb tone.