2026 is the year large language models (LLMs) became useful. And so, the story of the last 6 months for me has been one of increasing and increasing token consumption. It has taken me to places I couldn't have visited alone. I've created more. Been more prolific.
But is that a good thing?
I love building things.
It has been one of the great pleasures of my life to use my skills with computers, such as they are, to build apps which are either useful, fun or just geeky pleasures.
Generally speaking, every year or two I spend a period of intense effort to learn some new aspect of computing. A new coding language. A new operating system. A new technology. A new communication protocol.
This year has been the year to lean into and go deep on LLMs.
It's been so fun, too.
Doors that were closed to me have opened.
But we mortals are finite, and our limits when it comes to building things are our:
time
knowledge
skills
talents
creativity
attention
We can use these precious resources for some things but not for everything.
As Oliver Burkman points out in Four Thousand Weeks, this isn't a problem really. Life isn't a to-do list which we must tick off all of the things we want. It's a menu, from which we can select what to have from the vast selection that is available to us.
And this year, from the menu, I chose to build things with LLMs. Lots of things. To go at it as hard as I could.
But there was of course a cost to that. Whilst I was applying much of my time and attention to this enterprise, other things suffered.
You might think that LLMs can be used to expand your capabilities at no cost - that you can command a fleet of agents to go off and build solutions, research things for you, do your tasks. That your productivity is now without limits. This is not true.
An agent running in the background also uses some of your attention. Dilutes your focus. It has a cost, and a high one.
There's a potential for getting addicted to this thing, too. When you start to feel that you have to continue feeding the machine with tasks, or you are wasting the opportunities it could afford you, then you have messed up. You've undergone an inversion of control of a sneaky sort, and now you are working for this LLM rather than it working for you.
My favourite film is The Big Lebowski. The Dude has little. The film opens on him buying a carton of milk, writing out a check for $0.60. He has few ambitions. He is content to spend time with his friends bowling and enjoying simple pleasures. He doesn't want lots of money, he just wants his rug back. His friend Walter is a different story. Seeing the opportunity to make a million dollars, he pushes The Dude into a series of stressful situations, which almost manage to unseat The Dude's abiding calm, and very nearly spill his beverage.
Just lately I've been reflecting a lot about what the goal is for me. What I want.
Emily Austin's Epicurean advice to have tranquility as your goal is really compelling:
“For Epicurus, the first problem is that we focus on success rather than satisfaction. Satisfaction is a feeling or attitude that expresses a limit or endpoint, that point at which we have “Enough”. We can always have more money, more prestige, or more cars. We cannot, however, have “more enough”. Enough is enough. Epicureanism encourages us to draw the line of “enough” at the edge of the necessary desires, seeing that as the limit that surrounds the ingredients of tranquility. Epicurus thinks we can readily satisfy necessary desires if we (along with our friends and community) make them a priority. Note that there are two requirements for satisfaction-having enough and appreciating it as enough. When we have enough, but we fail to appreciate that we have enough, we lack tranquility because we feel unsatisfied.”
- Emily Austin, Living for Pleasure p140, Chapter 13 on Ambition, Work and Success
When I'm reading about LLMs, and the great race to build more. Build faster. Use more resources. Win this race... It feels like the goal is wrong. I perhaps need to stop. To slow down.
I'm asking myself, about my use of LLMs, what can I truly build with quality? What can I use my limited time, creativity and attention to build which is really worth something?
My feeling is we don't need really millions of tokens of LLM a day to be professional and to deliver good outcomes. To build awesome things.
I've realised that if I try to burn the candle at both ends, and maximise what I can do, the returns are diminishing. The cost is high. And the outcomes are questionable too. How many things can you really keep track of, and keep up the quality of, and maintain with a stable and reliable care.
For me it's worth taking the time to chill out a little, to abide, take her a little more easy.
To ask 'how many tokens are enough'?
