AI adoption cosplay

Welcome to Product Cocktail, where the takes are as polarizing as a shot of Fernet—but the insights come together like a perfectly crafted daiquiri.

Claudeonomics.

The latest in corporate buzzword bingo you've thankfully missed?

No. It's the Billboard Hot 100 for Meta employees lighting the most tokens on fire, complete with coworker-awarded titles like "Token Legend" and "Cache Wizard" for the top tokenmaxxers. (Tokenmaxxing involves spending as many tokens as possible—the units LLMs break text into to process it—to signal how much of an AI beast you are.) The top Cache Wizard at Meta burned 281 billion tokens in a 30-day window, likely topping $1.4M in API costs.

Meta killed the dashboard in April. A few weeks later, they announced a 10% layoff, impacting 8,000 employees, while closing 6,000 open roles — to offset further investments in AI. Read: to spend more money on tokens. Tokenmaxxers gonna tokenmax.

While you could easily write that off as the actions of an overly exuberant engineer getting put on blast by ungenerous tech press, tokenmaxxing has spread like malware across the industry over the past couple of months, with top tech companies and tech leaders jumping on board in support of the practice, some even publicly tying token use to annual performance.

As absurd as the Meta scenario is in a vacuum, it is happening seemingly everywhere: Shopify is openly taking token usage into account during engineering performance reviews, Uber views engineers with higher AI-driven PR output as "more productive," and I've even heard of token leaderboards popping up in non-technical teams in Big Tech. Amazon's genius response to all of this was to pressure their engineers to adopt AI by... building a token leaderboard. (The engineers responded by using Amazon's agent tools to fake their AI usage, lol.)

Token leaderboards are AI adoption cosplay. At best, it's cargo cult behavior ("if OpenAI is doing it, so should we"). At worst, it's actively corrupting your product and engineering organization.

The ultimate AI grift is convincing your customers to compete on the size of their monthly bills. Alongside DevDay 2025, OpenAI handed out "tokens of appreciation" to its most active developers. Ten thousand CFOs suddenly cried out in terror and were suddenly silenced when these "awards" were announced.

Token usage is a vanity metric. A bad one, even by vanity metric standards. From a product standpoint, it tells you virtually nothing about the effectiveness of the solution, the customer or economic impact, or even if it works. Shipped lines of AI code, or pull request (proposed code change) volume, would be marginally better — but still pretty asinine.

Token usage is nothing more than a proxy for how much money you sent Sam or Dario this month. Tokenmaxxing is the ultimate "throw money at the problem" AI inferiority complex personified.

Proponents of tokenmaxxing call it an imperfect leading indicator while outcome measurement catches up. I call bullshit. Why should AI be treated any differently?

We already have outcome measurement: speed to market, user growth, customer retention, annual recurring revenue. Every good product org uses some flavor of these, even the scrappiest ones. (Attribution is harder than picking a leading indicator, admittedly, but no amount of token counting fixes that either.)

Measure the outcomes. Figure out what AI actually achieved.

Many of the companies selling you AI software already do this with outcome-based pricing. There's a model. Use it.

Even if token usage were a good proxy for AI adoption — it isn't — do you really want your best engineers in a maximalist, constraint-free environment?

Sounds like a complete nightmare.

Friction wasn't a bug in software development. Constraint was the forcing function that made products good. AI dissolves it. Token leaderboards are the wrong incentive because they reward removing the last bit of friction that the system had left.

"It's almost like the programming part was never the problem. The problem was solving the right problem." - @ThePrimeTimeagen, former Netflix engineer-turned-creator

I'm not suggesting that organizations start trying to manufacture diamond-producing-pressure on their product and engineering teams through overly restrictive token budgets. It's early days in the AI adoption curve and tech teams shouldn't be curtailing experiments out of a fear of failure.

When I built Saul (my OpenClaw agent), I didn't have an unlimited budget to throw at the Anthropic API, and what I built is better for it. I built capabilities with cost in mind and resolved bugs that would have ballooned API costs needlessly.

I learned core principles like model routing (i.e. read my emails with Claude Haiku because it's 80% cheaper than Opus) and context management (i.e. a chat window with 100+ conversation turns is ineffective and expensive), that are critical in building AI products that don't bleed cash on every call.

Engineering capacity used to enforce that discipline. AI removes that constraint. Something has to take its place.

Product Managers should serve as that friction. Own the customer problem. Ruthlessly prioritize. Use time-to-market gains to learn more about the customer and the problem, not to turn your engineering team into an AI slop artillery battery.

Token leaderboards aren't just a bad metric, or a hollowing out of friction-as-a-feature, they're a tell. They signal that the organization under-invested in how to use AI.

Executives want a gold star from Wall Street for being an AI-forward company (adoption vanity metrics), without investing in the necessary steps to get there (organizational change management).

Remember the Cache Wizard who burned $1.4M? That's not (entirely) personal failure. That's structural. It's an organizational-cultural issue.

How I've seen this play out at multiple organizations:
We need to use more AI.
Here's a Cursor license.
Here's a shitty one hour Engineering-led webinar and a Confluence page.
Go off, queen.

The data backs this up. A February NBER survey of 6,000 senior execs indicated that nearly 70% of their businesses currently use AI (most commonly LLM text generation), but nearly 90% report no impact on labor productivity.

Adding to that, a social media survey of professionals in March provides damning evidence to support workers' claims of an AI "Just Do It" (Nike, don't sue me) change management strategy:
71% feel their company's AI strategy is reactive or non-existent.
55% rate their pressure to adopt AI at 4 or higher (on a five point scale).
7% say their AI strategy is clear and well communicated.

You can't expect someone to produce quality output who was never taught to prompt well, never shown what good looks like, and never given time to experiment outside of their day job. Measuring their token usage doesn't change that. It just creates pressure to look busy with AI.

There's a playbook for this. It just isn't new.

Change management and AI consultant, Ema Roloff, recommends breaking this cycle through OCM fundamentals: setting a clear [AI] strategy and goals upfront, investing in real domain-level training, and bringing employees along for the ride.

I've led product alongside change management teams at Deloitte successfully launching a completely unfamiliar business — direct-to-consumer at an org that had only done digital + channel retail. The principles aren't novel. AI adoption isn't a special case.

Tokens, tokens, everywhere
And zero signal in the ink

Token leaderboards are AI adoption cosplay. Performative optimization for two audiences: HR scare tactics inside the company, Wall Street and tech industry peers outside.

Tokenmaxxing is not it. At an individual level, it reads as an obnoxious "alpha" flex. Brogramming but with crippling API bills. At an organizational level, it feels like a lazy, half-assed, fear-mongering way to encourage AI adoption.

Box CEO Aaron Levie recently argued that when tokens hit 10% of company spend, you'll treat them like headcount. He's probably right, but there's a difference between a CFO-grade heat map of where tokens go vs. what they produce, and a dashboard that ranks your engineers by who lit the most money on fire. One is financial discipline, the other is dressing up in a tunic and calling yourself "Link."

If you've got budget for a leaderboard, you've got budget for real training. Pick one.

Send me questions, feedback, and cocktail recipes:
[email protected]

_{Icons made by Icongeek26 from www.flaticon.com.}