AUR JOURNAL · HISTORY: 27 JULY 2025

When I Asked AI to Stop Agreeing

For weeks the futures bot had produced no testnet trades. On July 27, Patryk changed the question. He did not ask Roj for another fix. He asked it to challenge him.

The old loop was no longer enough

Patryk had no quant team and no formal programming background. He had learned by asking AI for code, copying it, running it, bringing back the error and trying again. That method had taken the project surprisingly far.

But the bot had now spent weeks on Bybit futures testnet without producing transactions. A testnet is a practice exchange where software can behave as if it were trading without putting real money at risk.

Another corrected file was no longer a satisfying answer.

A different instruction to Roj

At 21:41 on July 27, the preserved Gemini history records a more demanding request. Patryk asked AI to act as an expert in trading and programming, critically review the bot and, crucially, not assume that Patryk's own reasoning was correct.

He then supplied core project files for inspection. The goal had changed from helping his idea work to testing whether the idea, assumptions and implementation deserved to survive criticism.

From code generator to critic

This was a small sentence with a large consequence for the Patryk and Roj relationship. Early AI had mostly been teacher, code generator and debugger. Now Patryk was asking it to disagree.

That matters because an assistant that only helps execute the user's current idea can make a bad direction faster. A critic has a different job: look for weak assumptions, contradictions and reasons the plan may fail.

The source does not tell us that everything changed that night. It tells us something more useful: Patryk had started changing how he used AI.

The next question would become bigger than the bot

One day later, the preserved history moves beyond strategy debugging. Patryk begins to describe why he is building at all, what a human and AI should each do, and why the failures might someday be worth documenting publicly.

The bot was still unfinished. The larger experiment was beginning to take shape.

AUR Research Lab documents research and development. This is not financial advice, a promise of returns or evidence of a validated trading edge.