I used to have a beautiful picture of the AI era: machines would do our laundry, cook our meals, write our reports. Humans would only need to focus on the most creative things, and life would get easier and easier.
Then I actually walked into it — and found the opposite. We’re more busy than before, not less.
I’ve been thinking about why. Here are the four reasons I keep coming back to.
1. The world has shifted from linear progress to exponential acceleration
The most direct thing I’m feeling lately: the pace of the technical and business world has fundamentally changed.
What really made this click for me was the release of MiniMax 2.7. It pushed model training into a new phase — “using models to train models.” Whether an Agent succeeds or fails inside its harness (the task-execution framework), the result is converted into a Reward Signal that gets fed back into the model for reinforcement learning. The whole system forms a self-iterating loop: model → Agent → harness → model.
We used to need humans to train models and tune parameters. Now the Agent framework itself is the model’s training ground. Every task execution is feeding back into the model. This “stepping on your own foot to lift yourself up” kind of self-reinforcement has pushed development speed straight from linear into exponential.
It’s also from this version onward that I started paying serious attention to harness engineering — it’s not just a tooling layer, it’s the engine of model capability growth.
The world is splitting into two paces: on one side, the slow GDP growth of the physical world; on the other, AI capability jumps measured in weeks. If you’re living in the second world, “continuous learning” stops being an optional career investment and becomes a basic survival skill. Fall behind by a week and you might not catch up. There’s a meme going around: if you stop learning AI for a stretch, you might as well stop entirely — because you can’t catch up anymore.
2. Decision cycles are compressing fast — “think it through first” no longer holds
Nowhere is this more obvious than in product delivery.
It used to take months to go from requirements to launch. Now, with AI, that whole flow can happen in weeks — sometimes a single week. For small features, pushing out an early version in two or three days to get real feedback is just normal.
The iteration loop keeps compressing. We’re losing even the gaps where we’d stop to reflect. The pace just pushes you forward.
More importantly: when the cost of trying something becomes very low, not trying becomes the biggest cost.
The old logic was: think it through first, then act, to avoid waste. The new logic is the opposite: ship something fast and let the market tell you the direction. The “one perfect solution” doesn’t exist anymore. Every version is a temporary fix you adjust as feedback comes in.
3. The boundaries of capability are dissolving — everyone can do more
The clearest view I have on this is from product management.
In the internet and mobile-internet era, the core work of a Product Manager was defining rules — clearly specifying business processes, interaction logic, data rules — and handing the rest off to engineering, design, and QA. The output was mostly demos and documents.
In the AI era, the core work of a PM has become mining intelligence + building an engineering loop:
- Mining intelligence — understanding what the model can and can’t do, and finding the intersection between model capability and real scenarios.
- Building an engineering loop — wiring model + data + tools + feedback mechanisms into a system that can iterate on itself.
That means a PM isn’t just making a high-fidelity prototype anymore. They need to ship a working system that’s demoable and keeps improving with use. It sounds like the role has become omnipotent — but really, the bar has been raised dramatically. You have to understand business, users, model boundaries, evaluation, and the engineering loop, all at once.
It’s not just product. Designers are writing code, engineers are doing design, ops folks are running data analysis. AI is turning work that used to require cross-functional collaboration into things a single person can drive end-to-end. The clean role boundaries are dissolving — but the number of capability dimensions each person has to master keeps going up.
More options, more landable things, more ambition to do more. Naturally we get busier — not because anyone is forcing us, but because we keep seeing things we couldn’t do before but suddenly can. And every “thing I can suddenly do” comes with the cost of learning a new skill behind it.
4. Welcome to the Agent era — now we have to manage “digital workers”
The recent rise of Agents has added a whole new homework assignment.
As more digital workers move into our workflows, we’re not just managing human-to-human collaboration anymore. We need to learn how to manage these AI helpers:
- How do you give an Agent the right task with the right context?
- How do you verify its output and spot where it went wrong?
- How do you coordinate human employees and Agents so neither side drops the ball?
None of this was in the playbook of decades of professional training. For most people, “managing digital workers” is like learning to manage a team for the first time — a skill you have to figure out from scratch.
And it’s harder than managing humans in one specific way: human employees you can develop slowly. Digital workers’ capabilities change every month. The workflow you tuned last month might need to be rebuilt next month after a model upgrade.
That’s another tax on our attention. Naturally, it gets harder to stop.
Closing thoughts
Looked at from another angle, this “getting busier” is essentially a new demand that technological progress is placing on us.
AI has freed us from repetitive physical labor — but it has also placed a higher-dimension challenge in front of us: an exponentially accelerating world, compressed decision cycles, dissolving capability boundaries, and a team of digital workers that needs managing.
The demands AI is placing on humans have shifted: from doing the work, to defining the standards, to building the systems, to defining the principles, to managing teams of digital workers.
Weekend musings. Just sharing.
Way to data scientist 
Leave a comment