[amazingman]

The dogmatism on both ends of this argument grows tiresome.

[port11]

Indeed. And so is reading “x is the real thing. Real y, real z.” Plus all the other filler crap.

If you can command a model, you can tell it to write without fluff.

The amount of effort to ask an LLM to do work is massively lower than the effort for a human to review it.

[amazingman]

I don't find this problem any different than with human writers. Agents are verbose, sure, but I mostly find them providing far more useful information (in far less time) than your average (P80, really) SWE.

[nevertoolate]

  def my_swe_percentile(best_agent_swe_percentile):
      return min(100, best_agent_swe_percentile * 1.25)