DEV Community

Brian Davies
Brian Davies

Posted on

What I Stopped Doing to Get Better AI Results

Improving my AI results didn’t come from learning more tricks.

It came from unlearning habits that felt productive but quietly degraded quality. Once I stopped doing a few specific things, outputs improved almost immediately — without changing tools or prompts.

Here’s what I stopped doing.


I stopped prompting before I understood the task

I used to rush straight to AI.

If a task felt vague or uncomfortable, I’d offload it immediately. AI would respond — and I’d mistake movement for progress.

Now, I pause first.

I clarify the goal, the constraints, and what a “bad result” would look like. When I do that, prompts get shorter and outputs get sharper. The problem was never the wording. It was the thinking.


I stopped regenerating instead of fixing

When outputs missed the mark, I used to regenerate.

Again. And again. And again.

That felt efficient, but it avoided responsibility. Regeneration hides the reason something failed. Editing reveals it.

Once I started fixing assumptions, clarifying constraints, and correcting direction manually, results improved faster — and stuck.


I stopped trusting fluency as a quality signal

Polished language is not proof of correctness.

I stopped letting structure, tone, or confidence convince me an output was good. Now I read for logic first:

  • Does this actually answer the question?
  • What claim is being made?
  • What’s missing or overstated?

The moment fluency stopped being persuasive, accuracy improved.


I stopped optimizing for speed first

Speed used to be the metric.

Faster outputs. Faster decisions. Faster delivery.

That optimization quietly removed verification and judgment from the workflow. I reintroduced pauses — not everywhere, but where errors would matter most.

Accuracy went up. Rework went down. Net speed improved.


I stopped collecting prompts and tools

Prompt libraries and new tools felt like progress.

In reality, they fragmented learning and prevented skills from transferring. I limited my environment and forced myself to understand why things worked instead of escaping to something new.

Fewer tools created more skill.


I stopped letting AI decide by default

AI suggestions are useful — not authoritative.

I stopped accepting structures, conclusions, or priorities just because they were presented confidently. I forced myself to choose, justify, and own outcomes.

Once decision-making stayed human, results stabilized.


What replaced those habits

Stopping those behaviors made space for better ones:

  • Clear task framing
  • Intentional evaluation
  • Editing over regeneration
  • Judgment over automation

This is why learning approaches like those emphasized by Coursiv focus less on hacks and more on control — helping learners build workflows that improve results consistently, not accidentally.

Because better AI results don’t come from doing more.

They come from doing less of what quietly undermines quality — and more of what keeps thinking in the loop.

Top comments (0)