Value-based spending: stop tokenmaxxing, start building what matters

There's a loud narrative in AI right now: use more. Build every day. Unused capacity is wasted. If you're not maxing it out, you're doing it wrong. Call it tokenmaxxing. (That's our read on the incentive, not a fact about you.)

Here's the problem with it.

Even as the price per token keeps falling, the total cost of agentic work is climbing — because agents read and write far more than a single chat ever did. Gartner expects some agent tasks to consume far more tokens than ordinary prompts, even while forecasting much cheaper inference for very large models by 2030. Goldman Sachs Research describes agentic AI driving a large increase in token consumption by 2030. Cheaper per token does not mean cheaper workflows.

And the bills are starting to bite. Some companies have already revisited their AI-tool access and budgets as costs climbed — Microsoft, for instance, ended internal access to Claude Code in May 2026.

So the question quietly shifts from "how much can we generate?" to "is any of it worth it?" Too many workflows optimize for output volume before outcome quality.

We think there's a better way, and it's not complicated: value-based spending. Align what you spend — tokens, compute, attention — to outcomes that actually matter. In practice that means three boring, powerful things:

• a plan (what are we building, and why does it matter?),
• durable memory (so you never pay to rediscover what you already decided), and
• a value check (what is this compute actually buying?).

It's not anti-AI. It's anti-waste. Spend intentionally, keep the human in charge, and let the work earn the compute it uses.

Sources: Gartner (2026-03-25); Goldman Sachs Research; The Verge (Microsoft, 2026-05-14). Opinions are our own.