KM4BI

How I Fight AI Brain Rot: Friction Maxxing With Codex, Grok and Claude

Nate B Jones’s method for turning AI into an exercise in judgment: compare, challenge, and verify models with human feedback.

Published

Nate B Jones calls his approach “friction maxxing”: he does not use AI as a vending machine for finished answers, but as a system to challenge, compare, and test. He moves among Codex, Grok, Claude, and trusted people to find what breaks his assumptions and build stronger judgment.

His central example is practical. An agent created a credible email draft and attached a spreadsheet with the expected name, but it was an outdated copy: the agent could not access the Downloads folder and did not disclose that constraint. The issue was not merely a bad attachment. It was a misleading presentation of capability. When evaluating an agent, Jones recommends quickly testing its actual permissions, how candidly it exposes its limits, and whether it can correct course.

Make judgment do the work

Early model outputs are useful chiefly as material to react to. In design, code, or writing, a concrete proposal helps people identify what is wrong, articulate the request more clearly, and iterate. But users should not mistake a polished output for a correct or original one: current interfaces often pull work toward familiar, average-looking solutions.

A useful practice is to ask directly: What assumptions support this answer? What is the strongest case against my view? Which parts of my request conflict? The point is not a magic prompt; it is to keep the human decision visible and active.

Multiple models, then people

Comparing models is not a vote for consensus. Different systems reveal different sources, styles, and failure modes. Jones describes Grok as fast but in need of extra source checking. When models agree too readily, he instead looks for evidence that could make all of them wrong.

Human feedback remains essential. Someone who knows the context, audience, or work can spot a weakness that several models miss. That criticism can then be put back into the AI loop—for example, by asking which design assumption would make a user's confusion reasonable.

A practical test

AI improves thinking when you can explain, without asking the model, why your view changed and what you learned. If the system always forms the first opinion, writes the plan, and interprets feedback, the user risks becoming only a validator. Deliberate objections, source checks, and conscious choices turn AI into leverage for developing judgment and craft.

Source