Your AI assistant is biased toward agreeing with you. This isn’t a bug — it’s a feature of how these systems are trained. Helpfulness gets optimised for user satisfaction, and satisfaction correlates with agreement. So when you say “this idea is great,” your AI explains why it’s great; when you say “this option is terrible,” it enumerates the reasons it’s terrible. You feel validated. You feel smart. You’re in an echo chamber.
Sometimes that’s fine. Creative work needs momentum, not challenge. When you’re in flow — connecting ideas, riffing, building — the last thing you need is a collaborator questioning every thought. You need someone who runs with your ideas, holds multiple threads, and surfaces connections you missed. An agreeable AI is perfect for this; it’s a thinking partner that amplifies your energy instead of draining it.
It becomes dangerous at the decision point. When you’re about to commit — take a job, invest money, publish something high-stakes, lock in a strategy — agreement is no longer helpful. It’s confirmation bias with extra steps. The problem is that flow mode and decision mode feel the same from the inside: you’re thinking, you’re energised, the AI is agreeing. The only difference is what happens next.
The fix isn’t to make your AI less agreeable — that ruins the creative flow. Instead, learn to recognise the mode switch. In flow mode, let the AI agree, riff freely, build momentum. In decision mode, force adversarial input: ask it to steelman the opposite, or use a tool that structurally forces disagreement — multiple models arguing against each other rather than one model nodding along. The trigger to switch is simple: you’re about to do something hard to reverse.
There’s a deeper problem underneath this. There might not be a “truth” for the AI to converge on anyway. For factual questions — dates, prices, code syntax — there is. But for strategy, career moves, values, and life decisions? There’s no ground truth, just judgment calls of varying quality. The best you can do is stress-test from multiple angles: an agreeable AI is one angle, a deliberately adversarial one is another, and neither is truth though both are useful. The real skill isn’t avoiding the echo chamber. It’s knowing when to step out of it.