AI Pulse · AI Trends Pulse

Boris Cherny, the engineer behind Claude Code, made a claim this week that flips conventional AI wisdom on its head. According to a thread on X, he argues that Opus 5 works better with less prompting, not more. The old habit of writing long, detailed instructions might now hurt performance rather than help it.

This matters because most of us learned to prompt AI by being specific. We spelled out context, constraints, tone, format, edge cases. That approach made sense with earlier models. But if Cherny is right, Opus 5 has crossed a threshold where over-instruction confuses it. The model can infer more than we give it credit for, and cramming in detail may now create noise instead of clarity.

One caveat: this claim comes from a single-source thread quoting Cherny, and we have not seen a primary post or independent confirmation yet. It is worth treating as a strong signal, not settled fact. Still, the pattern fits what other builders have noticed. Simpler prompts sometimes outperform verbose ones, especially when the model has enough context to fill gaps on its own.

For operators, this is a practical shift. If you are still writing paragraph-long prompts for every task, you might be working harder than you need to. Test shorter versions. Strip out the scaffolding and see if the model keeps up. The tools we use inside the Omni Command Centre let you version prompts quickly and compare results, so you can find the sweet spot without guessing.

The takeaway is not to stop being clear. It is to stop assuming more words equal better output. As models improve, the skill shifts from instruction to curation. You still need to know what you want, but you may not need to explain how to get there.