Absolutely critical reading.

Inside tech, AI Slop has WAY, WAY, WAY up. Slop replying to slop. Deepslop.

Deepseek and GLM do slop just as well as Opus 4.8 and GPT 5.5 at a FRACTION of the cost.

Frontier AI still cannot do basic valuable things inside companies: * Write non-slop coherent sentences * Come up with 'novel' plans for features * See whether a design is 'good' * Define architecture from a systems perspective with constraints * Tell you what's important

They can absolutely execute mid-level code changes and reviews.

The reality is tech people are trying to use AI to do BOTH and it's costing them big right now.

AI is best seen as bureaucracy, at a personal and company scale. It's become EXPENSIVE bureaucracy (https://attention.substack.com/p/the-age-of-ungovernable-ai-...)

There are some companies whose work depends on efficiently optimized bureaucracy: financial markets will continue to pour millions if not billions into tokens, because the incremental intelligence on the datapoints is really worth it.

For 'glue' between previously hard to connect disciplines - this is also massive improvement. Privacy, data integrity, etc. "reviews" will become much better.

But for actually making the product better and organizations better, it's not clear that tokenmaxxing is an appropriate or even worthwhile strategy.

From experience it instead seems to amplify the native differences in ability, such that the bottom 80% waste tons of tokens to generate more and more output to seem like the top 20% of performers. It can be exhausting to separate the increased amount of wheat from the slop-generated chafe.

This is a losing proposition for the AI providers, and the companies. And will start to really punish the incremental high-priced tokens outside of a very small wedge of industries.

You really gotta hope the next 'versions' of models produce, or the economics are very, very cooked.

'We created a monster': companies rein in AI usage as costs strain budgets