If you follow AI (and it's hard not to), you are probably signed up to a few newsletters. Anthropic, OpenAI, Google. The progression of features and capabilities in those emails is staggering. Not a tweak or a patch note. A new model, a new capability, something that was not possible up until not too long ago.
It is easy to lose your sense of scale inside that. So it helps to step back.
While Moore's law was an observation about transistors, the shape of it, the same interval producing twice the result, turns up in places Gordon Moore never meant it to. Look at how long human eras used to last:
| Time period | Phase | Duration |
|---|---|---|
| ~300,000 – 10,000 BCE | Pre-civilization (stone tools, nomadic) | ~290,000 years |
| ~10,000 – 500 BCE | Agriculture and early metals (farming to Iron Age) | ~9,500 years |
| ~500 BCE – 1500 CE | Pre-industrial empires (Classical to Medieval) | ~2,000 years |
| 1500 – 1800 CE | Scientific and global expansion (Early Modern) | ~300 years |
| 1800 – 1950 CE | Industrial revolution (mechanization, mass production) | ~150 years |
Every phase is shorter than the one before it, and not by a little. Now zoom into the stretch we are actually standing in:
| Time period | Phase | Duration |
|---|---|---|
| 1950 – 1979 | Institutional computing (mainframes, early silicon) | 30 years |
| 1980 – 1999 | Personal computing and the web | 20 years |
| 2000 – 2015 | Mobile and cloud | 16 years |
| 2016 – present | AI and accelerated computing (specialized silicon) | 11 years, so far |
You could keep breaking that down, and the direction remains the same.
On 22 August, at the World Humanoid Robot Games in Beijing, a humanoid robot ran the 100 meters in 9.39 seconds, beating Usain Bolt's world record of 9.58s. Elsewhere in China, robots are running synchronized dance routines. Not long ago, the impressive part was that they could cross a stage without falling over.
What this does to anyone trying to build something
When I started my own AI journey I did not have much of a plan, just the feeling that I wanted to make something. What happened next is a story I wrote about here: I ended up spending more time managing agents than building.
Here is the part I did not understand while it was happening. Me being overwhelmed was not a discipline problem, and it was not a me problem. It was a direct consequence of the curve above.
Before our current AI era, the main constraint for building something was capability. Time to market was long. You had to work out whether the thing could be built at all, and then whether the tools you had could actually do it. That constraint has now dissolved. Getting from a rough idea at point A to a working product at point Z now depends on how fast you can think, what your subscription cost is, and how many tokens you can afford to spend on it.
So the bottleneck moved. It is not in the tool any more. It is in you: your attention, your memory, your ability to hold five parallel threads without dropping one of them on the floor. We have to become better at managing the resources standing at our disposal. Sometimes you need to spend more time creating the tools than the actual end product.
The tools got faster than the person holding them.
So I built the thing that holds the threads
Clayrune came out of that. I am not going to hand you a feature list, they are dull and there is one on the site if you want it. What is more useful is what I actually needed, because that is why each of these is in there:
- Every project on one screen, agents running side by side, so I stop losing track of what is where.
- Control from my phone, over a tunnel to my own machine, because the work does not stop when I walk away from the desk.
- More than one conversation inside a project, because real work is never a single thread.
- Memory that survives the end of a session, so I stop explaining the same decision for the third time.
- Agents that run on a schedule, or unattended against a standing brief, because not everything needs me watching it.
- A counter that tells me what each session cost. It runs on my Claude subscription rather than the API, so nothing is metered and nothing is routed through anyone else. I just get to see where the tokens went.
When I went looking for a tool that did this, I could not find one. That is the entire reason it exists.
All this existed in CR almost from day one.
The honest version
Can you work without it? Yes. Plenty of people do, and they do fine.
But the rate is not slowing down. Every month there is more to keep track of, more parallel work you could plausibly be running, more surface for something to quietly fall through. The tools will keep getting faster. Your attention will not.
Clayrune is open source and it runs on your own machine. If it is missing something you want, you can ask your agent inside it to go build that thing, and it will. A fair amount of what is in there today arrived exactly that way.
I do not know what next year looks like either. I would just rather not meet it with a screen full of terminal windows.
Ron