The whole maxxing thing is pissing me off.

Publicis recently released what I originally though was a poorly execute parody.

Turns out it was a genuine, provocation pre-Cannes.

But it has been met with contempt and calls of hypocracy on our glorious slopfeed, linkedin.

I don’t think thats wholly because people disagree with the point - that there are too many overhyped AI promises and cut‑price pitches flying around.

I think it’s more that it feels like a poorly considered, perhaps slightly hypocritical ‘maxxing’ trend-jump, which unnecessarily associates the point with a pretty dark past.

‘Maxxing’ has been around for years.

Originating in game-theory.

Adopted by the ‘incel’ community (I wont promote them, but you know, the subject of adolescence and Louis Theroux’s inside the manosphere) to promote looks maxxing - the practice of maximizing physical attractiveness.

More recently, token maxxing was all over the AI industry news.

The belief that (AI / LLM) token consumption equates to productivity and thus can be used as a metric to monitor an employee's work.

It’s quite simple - as soon as you incentivise something, people will do their best to hit the incentivisation metric.

It led to employees choosing the most expensive model for the simplest task.

Employees actively prompting in the least efficient way.

Employees not having any understandable lineage back to the cost of their work.

Employees not caring about whether their work was attached to a business outcome.

This is great if you’re one of the big tech bros, trying to get as much revenue as possible. To recoup the vast amount of model training and compute costs racked up in the race to have the best arbitrary model metric.

Not so great if you’re an enterprise like Uber - burning your annual AI budget in 4 months…

Great if they managed unlock a tonne more demonstrable value, but I’m not sure theres anything to demonstrate that!

Sounds familiar, if you’ve been around cloud computing over the last decade.

Employees having no idea how much a query actually costs to run.

Employees being able to spin up a compute instance that burns £500k over a weekend.

Hyperscalers pushing consumption at seemingly any cost.

FinOps was the answer here.

Basically just good governance, visibility of consumption and an education piece on the price of consumption.

I’ve seen it cut 20% off annual cloud bills.

Save £4m annually, just by optimising the method of querying.

I think this is why the maxxing stuff grates on me.

Well, toxically maxxing the thing someone else wants you to maxx.

If someone pointed a weapons grade AI model at my head and forced me to maxx something…

I’d advocate for outcome maxxing.

Or budget maxxing.

But, I’d rather just be an anti maxxer.

Avoid pushing everything to the absolute maxx.

Ground the work we do in actual, tangible value.

Establish the business impact and value creation before going all in on big bang development.

Establish cost visibility / management metrics prior to kicking off.

Optimise the prompt / approach throughout delivery to optimise the output, not grow the volume of tokens consumed.

Measure the cost and impact at the end, to improve future work.