This is a reaction to Dan Koe's challenge of writing in a more authentic way in the age of AI.
I woke up yesterday.
First order of the day, bathroom. Some things never change.
Second, I didn't pull up Instagram. The first thing I do while seated on my throne is check what my AI agents have been doing.
It's 2026, and people are talking to agents instead of each other.
I open what I call the Agentic Command Center. I built it with agents. Four agents, designed on guidance from agents, after a discussion with my strategic agent.
Agents everywhere.
(Most of them aren't really agents. They're glorified chatboxes. I'm narrating this one to an agent too.)
Two years from now nobody will think twice about it. Today it still sounds odd to tell a friend you spent the evening chatting to artificial intelligence and had fun.
I wanted to write about stability in the AI age, and then I realised the stability metric is different for every single person.
So here is mine.
Here's how I imagine stability.
On the first of each month I swipe my card and buy a subscription from my AI provider of choice. I switched providers today, as a matter of fact.
The amount is predictable. I would love to keep it at $20. There will be better and more expensive tiers, and part of an intentional AI life is limiting AI to the tasks that actually matter.
I don't audit what my agents did overnight. I assume they did the job, and sometimes created a new problem in the process. That's inevitable.
Stability, for me, includes a little instability. I wouldn't be happy knowing every task gets completed with 100% precision. It needs to fail every now and then, for the thrill, and for the reminder that I'm alive.
Stability also means my life gets progressively better, and so does the life of the people I care about. Their metric can be different. For a lot of people a better life is a predictable day, a predictable salary, and ideally zero challenges. That's the majority, honestly.
So, the question:
Are we at a point where life surrounded by AI agents is stable?
Big fat nope.
The technology has been around for seven to ten years. It hasn't made insane breakthroughs lately.
What's been racing is the rate at which we productify it.
The only thing you can reliably predict is that the ecosystem changes every week. That's a confirmation, not an argument for stability.
We have the use cases figured out. We have the prototypes figured out. We do not have infrastructure, scaling, or a business model that makes AI companies profitable. Intentions and sustainability, not by a long shot.
It feels like we're throwing things at the wall to see what sticks. From a business perspective, not much is sticking.
Three variables that need to settle.
One: price. Most AI companies lose money daily. Some burn billions a year. Investors keep pouring money in expecting a 10x or 100x return, which looks unrealistic at the scale some of these operations run.
Two: infrastructure. When I look at infrastructure I see a series of overloaded clouds. Cloud is probably the real business today: let people run their AI projects on your machines, your utilization goes up, your margins follow. Congratulations if that's you. But the queue isn't endless.
Three: energy. Even with a record number of servers plugged in, we aren't producing close to enough power, and the grids were never designed for this throughput. If the energy showed up, the next move would be making it cheap, which means coal and a few riskier options like nuclear.
Model reliability isn't the problem. Models have become amazing compared to the early GPT-2 and GPT-3 tests that blew my mind at the time. Everything around them is the problem.
Here's what that instability looks like on a Tuesday.
I wake up, do my bathroom break, open my agentic center. Instead of a report of what got done, I see:
Codex is overloaded. Please try again in 177 minutes.
Fine. Overloaded. The 177 minutes pass, I try again manually:
Overloaded. Please try again in 213 minutes.
OK, something's real, people are cooking. Four hours later:
The models are temporarily rate limited. Please try again later.
Suddenly you don't know when later is. And later becomes a full day of watching a screen and restarting a process.
So, a conclusion.
Generally, life with AI agents is not stable. It is stable for two groups.
People with money and compute. A decent machine with a high-end GPU that runs local models. They'll keep upgrading rigs, running better models, and learning how the models actually work. They won't lose their edge.
People who don't care about AI. That one is going to change, dramatically. This is becoming a skill that pays bills. We keep saying you can always fall back to woodwork, and then I watch what robots powered by these models can already do. The technology isn't there yet. It wasn't there for the iPhone three years before the iPhone either.
I run local models myself, and local makes a few things predictable. I know where my data sits. I know what quality of answer the same model gives me every day. I know exactly when the model changes, because I'm the one who changes it.
For everyone on the big third-party providers, life gets harder. More dependency than today. Thinking, strategy, and even the way people talk starts to converge on the same shape.
None of this makes AI a bad thing.
It's one of the best things that happened to humanity since we discovered nuclear power, up until the point that got weaponized.
I'd like to keep living in the version where AI stays unweaponized, predictable, and really, truly mine to operate.
That unpredictability is exactly why I keep coming back to whether any of this is actually paying off, which I dug into in I Still Don't Have ROI on AI, But I Do Have a Subjective Win. And it's a big part of why I don't think the answer is grinding harder to keep up, a point I argue more directly in Do We Really Need the Toxic Hustle Culture in 2026?.