Winter Is Coming. Time to Make Token Kimchi.

Tools like OpenAI Codex and Anthropic Claude Code give us access to an enormous amount of AI capability through relatively affordable, subsidized subscriptions.

If you have a Max plan, you can burn through a lot of tokens without worrying much about the cost.

And honestly, it feels amazing.

But I think many of us are looking at this the wrong way.

We are in the summer of AI tokens

Right now is the summer of AI tokens.

Resources feel cheap and abundant. If your company gave you a generous token budget, it might even feel endless.

You can lean on agents freely, generate plenty of code, analyze large contexts, and automate things that would have been way too expensive not long ago.

But summers do not last forever.

At some point, these AI companies will tighten the economics. Subscriptions may become more restrictive. Usage pricing may become more aggressive. Tokens may eventually become something companies track very carefully instead of casually burning.

One day, winter may come.

So what should we do during the summer?

When there is too much cabbage in the summer, Koreans make kimchi.

You preserve abundance so it survives the long Korean winter.

I think we should treat AI tokens the same way.

Instead of using today’s cheap tokens only for temporary outputs, we should use them to build systems that permanently reduce future token usage.

I call this Token Kimchi.

Use today’s cheap tokens to reduce tomorrow’s expensive tokens.

Not everything should stay an AI task forever

A lot of people are using AI agents for operational workflows now.

Writing reports. Managing content. Processing data. Reviewing code. Running repetitive tasks.

And that makes sense.

But after repeating a workflow enough times, patterns appear.

The inputs are similar. The decisions are similar. The outputs are similar.

At some point, the AI should probably stop “thinking” every single time.

Parts of the workflow should become deterministic.

Instead of repeatedly asking agents to rediscover the same answers, we should turn those patterns into:

• tools
• workflows
• interfaces
• reusable logic
• structured systems

That means fewer tokens, less reasoning, and less waste.

Turn tokens into assets

Most people use AI in a very consumptive way.

Prompt in. Output out. Done.

But the smarter move is turning token usage into infrastructure.

Use AI while tokens are cheap to:

• build reusable workflows
• optimize prompts
• reduce unnecessary reasoning
• improve context systems
• create better interfaces
• structure knowledge properly

These things compound over time.

A good workflow can save thousands or millions of tokens later.

Token efficiency will be the hot topic

I think token efficiency will become one of the biggest topics in AI over the next few years.

As companies adopt AI more seriously, they will start asking harder questions about token ROI.

How many tokens are we spending? What are we actually getting back?

That is why tokenmaxxing is not the right answer.

Using more tokens does not automatically create more value. Bigger contexts, longer prompts, more agents, and endless reasoning loops can become wasteful very quickly.

The real advantage will belong to teams that can get the same or better results with:

• fewer tokens
• smaller contexts
• tighter workflows
• better tooling
• more deterministic systems

The goal is not just smarter AI.

The goal is less wasteful AI.

Build while the tokens are cheap

Right now is probably one of the cheapest moments in history to experiment with large-scale AI workflows.

That is a rare opportunity.

This is the time to build systems.

This is the time to optimize workflows.

This is the time to turn temporary token abundance into permanent infrastructure.

Make token kimchi while the cabbages are cheap.