Token burn: good or bad?

So which is it?

High token burn = good

or

High token burn = bad

Unless you are financially incentivized to burn hot (ie you are the token provider/reseller) it seems silly to say token burn rate is a useful signal, in isolation.

What does make sense is token burn rate relative to value creation. Most businesses would happily subsidize any token costs that produced more value than costs.

But how do you do that? How do you quantify value creation? If we are being honest, this has been, and continues to be, very difficult to quantify. It gets even more difficult when your token consumption is spread like peanut butter across departments in your org.

So enter the monthly plan. Shift the worry from token costs to monthly limits. Just don’t exceed your weekly/monthly limits and you have a predictable cost for relatively good token capacity. Better, but still not enough to know you are creating more value than costs. But better.

As we have already seen, cough Anthropic, these monthly plans are gateway drugs. They are intended to reduce the token anxiety to drive adoption. AI is so powerful that it is very likely to create a dependency that becomes difficult to imagine going without. Then the economics will shift. It is already happening.

Here is an analogy. Remember how amazing it was to “cut the cable” and stream your favorite shows? No ads. Lower costs. Way better. Until it wasn’t. All the reasons most people “cut the cable” have come back. You probably spend more $, in aggregate, than before to watch your shows. Ads everywhere. The honeymoon is coming to an end. It is just a matter of time. A few players still haven’t caved and just create great television for a flat monthly fee. But for how long?

There is a chance in the not-too-distant future that the AI monthly plans change. Less likely that they are removed, forcing users back to costly pay-as-you-go tokens. More likely the monthly plans start to look more like the costs of hiring a human. As AI continues to be capable of doing the work of an employee, I think we see the cost inch closer to the cost of hiring said human. It makes sense if you think about it. You can have the equivalent of an employee for less costs and there is no burden (ie pto, benefits, etc).

One possible positive side effect of this will be a meaningful push to run models (at home or in business) locally. The models will always lag behind cloud frontier, but they will get good enough. Imagine Opus 4.6 capable model running locally with zero token costs and unlimited use. It is going to happen in time.

So which is it? Good or bad to run hot? Well if you can answer that relative to value creation you have your answer. If you can’t then you need to be mindful on where and how you allow AI dependencies to root. Do not avoid AI adoption. Do be mindful of how it positions your company for when the economics shift.

Reach out! I would be happy to share more and even help you assess where you are in your AI journey…