Trendoday

In 27,280 coding agent runs, 64% of inference spend went to runs that failed

2 comments

  • Points344
  • Comments120
–Rank now
#1Peak rank
2Days on list
165Score · 7d
Nov 17, 2026First seen
Rank · 7 days

On Hacker News

Points
344
Comments
120
Submitted by
Edward40
Posted
Nov 17, 2026
Site
runta.com

Story details from Hacker News ↗ · linked article on runta.com.

Trending history

In 27,280 coding agent runs, 64% of inference spend went to runs that failed was first seen on Hacker News · Front page on Nov 17, 2026. Since then it has appeared in 92 checks on 2 different days. Its best position was #1 of 30, reached on Nov 17, 16:43 UTC. It dropped off the list after Nov 18, 14:43 UTC.

Rank history

Position on Hacker News · Front page at every check (lower is better).

#1#11#20#30 15:5821:4003:2109:0214:43 Nov 17, 15:58 UTC · #7Nov 17, 16:13 UTC · #3Nov 17, 16:28 UTC · #2Nov 17, 16:43 UTC · #1Nov 17, 16:58 UTC · #1Nov 17, 17:13 UTC · #1Nov 17, 17:28 UTC · #1Nov 17, 17:43 UTC · #1Nov 17, 17:58 UTC · #2Nov 17, 18:13 UTC · #2Nov 17, 18:28 UTC · #2Nov 17, 18:43 UTC · #2Nov 17, 18:58 UTC · #3Nov 17, 19:13 UTC · #4Nov 17, 19:28 UTC · #4Nov 17, 19:43 UTC · #5Nov 17, 19:58 UTC · #7Nov 17, 20:13 UTC · #8Nov 17, 20:28 UTC · #10Nov 17, 20:43 UTC · #9Nov 17, 20:58 UTC · #9Nov 17, 21:13 UTC · #11Nov 17, 21:28 UTC · #11Nov 17, 21:43 UTC · #13Nov 17, 21:58 UTC · #13Nov 17, 22:13 UTC · #13Nov 17, 22:28 UTC · #13Nov 17, 22:43 UTC · #13Nov 17, 22:58 UTC · #13Nov 17, 23:13 UTC · #13Nov 17, 23:28 UTC · #13Nov 17, 23:43 UTC · #14Nov 17, 23:58 UTC · #15Nov 18, 00:13 UTC · #16Nov 18, 00:28 UTC · #16Nov 18, 00:43 UTC · #16Nov 18, 00:58 UTC · #16Nov 18, 01:13 UTC · #16Nov 18, 01:28 UTC · #16Nov 18, 01:43 UTC · #16Nov 18, 01:58 UTC · #18Nov 18, 02:13 UTC · #18Nov 18, 02:28 UTC · #18Nov 18, 02:43 UTC · #17Nov 18, 02:58 UTC · #19Nov 18, 03:13 UTC · #19Nov 18, 03:28 UTC · #18Nov 18, 03:43 UTC · #19Nov 18, 03:58 UTC · #19Nov 18, 04:13 UTC · #20Nov 18, 04:28 UTC · #20Nov 18, 04:43 UTC · #21Nov 18, 04:58 UTC · #21Nov 18, 05:13 UTC · #21Nov 18, 05:28 UTC · #21Nov 18, 05:43 UTC · #23Nov 18, 05:58 UTC · #24Nov 18, 06:13 UTC · #23Nov 18, 06:28 UTC · #24Nov 18, 06:43 UTC · #25Nov 18, 06:58 UTC · #25Nov 18, 07:13 UTC · #24Nov 18, 07:28 UTC · #25Nov 18, 07:43 UTC · #25Nov 18, 07:58 UTC · #25Nov 18, 08:13 UTC · #27Nov 18, 08:28 UTC · #26Nov 18, 08:43 UTC · #27Nov 18, 08:58 UTC · #28Nov 18, 09:13 UTC · #27Nov 18, 09:28 UTC · #26Nov 18, 09:43 UTC · #26Nov 18, 09:58 UTC · #25Nov 18, 10:13 UTC · #28Nov 18, 10:28 UTC · #28Nov 18, 10:43 UTC · #27Nov 18, 10:58 UTC · #27Nov 18, 11:13 UTC · #27Nov 18, 11:28 UTC · #27Nov 18, 11:43 UTC · #28Nov 18, 11:58 UTC · #29Nov 18, 12:13 UTC · #28Nov 18, 12:28 UTC · #29Nov 18, 12:43 UTC · #28Nov 18, 12:58 UTC · #27Nov 18, 13:13 UTC · #27Nov 18, 13:28 UTC · #27Nov 18, 13:43 UTC · #28Nov 18, 13:58 UTC · #29Nov 18, 14:13 UTC · #28Nov 18, 14:28 UTC · #28Nov 18, 14:43 UTC · #28

Last 14 days

Best rank per day (UTC).

Nov 7today

Metric history

Platform metric as reported by Hacker News.

0134267405 15:5821:4003:2109:0214:43 Nov 17, 15:58 UTC · 13Nov 17, 16:13 UTC · 29Nov 17, 16:28 UTC · 57Nov 17, 16:43 UTC · 79Nov 17, 16:58 UTC · 100Nov 17, 17:13 UTC · 127Nov 17, 17:28 UTC · 139Nov 17, 17:43 UTC · 162Nov 17, 17:58 UTC · 182Nov 17, 18:13 UTC · 194Nov 17, 18:28 UTC · 205Nov 17, 18:43 UTC · 217Nov 17, 18:58 UTC · 231Nov 17, 19:13 UTC · 241Nov 17, 19:28 UTC · 250Nov 17, 19:43 UTC · 258Nov 17, 19:58 UTC · 267Nov 17, 20:13 UTC · 269Nov 17, 20:28 UTC · 272Nov 17, 20:43 UTC · 278Nov 17, 20:58 UTC · 283Nov 17, 21:13 UTC · 288Nov 17, 21:28 UTC · 291Nov 17, 21:43 UTC · 293Nov 17, 21:58 UTC · 296Nov 17, 22:13 UTC · 301Nov 17, 22:28 UTC · 303Nov 17, 22:43 UTC · 307Nov 17, 22:58 UTC · 311Nov 17, 23:13 UTC · 313Nov 17, 23:28 UTC · 318Nov 17, 23:43 UTC · 322Nov 17, 23:58 UTC · 323Nov 18, 00:13 UTC · 324Nov 18, 00:28 UTC · 326Nov 18, 00:43 UTC · 329Nov 18, 00:58 UTC · 332Nov 18, 01:13 UTC · 332Nov 18, 01:28 UTC · 335Nov 18, 01:43 UTC · 335Nov 18, 01:58 UTC · 337Nov 18, 02:13 UTC · 337Nov 18, 02:28 UTC · 338Nov 18, 02:43 UTC · 338Nov 18, 02:58 UTC · 340Nov 18, 03:13 UTC · 341Nov 18, 03:28 UTC · 343Nov 18, 03:43 UTC · 344Nov 18, 03:58 UTC · 344Nov 18, 04:13 UTC · 345Nov 18, 04:28 UTC · 346Nov 18, 04:43 UTC · 347Nov 18, 04:58 UTC · 350Nov 18, 05:13 UTC · 352Nov 18, 05:28 UTC · 352Nov 18, 05:43 UTC · 352Nov 18, 05:58 UTC · 356Nov 18, 06:13 UTC · 358Nov 18, 06:28 UTC · 358Nov 18, 06:43 UTC · 358Nov 18, 06:58 UTC · 360Nov 18, 07:13 UTC · 362Nov 18, 07:28 UTC · 362Nov 18, 07:43 UTC · 362Nov 18, 07:58 UTC · 363Nov 18, 08:13 UTC · 366Nov 18, 08:28 UTC · 368Nov 18, 08:43 UTC · 368Nov 18, 08:58 UTC · 369Nov 18, 09:13 UTC · 370Nov 18, 09:28 UTC · 371Nov 18, 09:43 UTC · 373Nov 18, 09:58 UTC · 373Nov 18, 10:13 UTC · 375Nov 18, 10:28 UTC · 377Nov 18, 10:43 UTC · 377Nov 18, 10:58 UTC · 379Nov 18, 11:13 UTC · 379Nov 18, 11:28 UTC · 382Nov 18, 11:43 UTC · 383Nov 18, 11:58 UTC · 385Nov 18, 12:13 UTC · 388Nov 18, 12:28 UTC · 389Nov 18, 12:43 UTC · 390Nov 18, 12:58 UTC · 390Nov 18, 13:13 UTC · 392Nov 18, 13:28 UTC · 394Nov 18, 13:43 UTC · 399Nov 18, 13:58 UTC · 400Nov 18, 14:13 UTC · 400Nov 18, 14:28 UTC · 405Nov 18, 14:43 UTC · 405

From the discussion

A few top-level comments from Hacker News, shortened. Read the full thread on Hacker News.

The headline here is confusing. They ran a bunch of harness and agent combinations on terminal bench tasks, and 64% of the spending was on tasks that didn’t pass terminal bench verifiers. It’s unclear whether that’s supposed to be surprising against the selected models for the benchmark. I suspect it’s not and the…

There's no IP in a harness, and they don't matter that much. Harnesses simply make different choices around certain context engineering things, what they name tools, how they compact, their system prompt, etc. So it's not surprising that they work "better" for some models than others, but I'd bet that if you made…

I find myself pulled in two directions by the performance of coding agents: First, keeping them on a short leash seems sensible. I don't get failed results very often. When I first started using coding agents, this seemed not just sensible but required in order to not deal with a lot of hallucinated weirdness.…

Questions

Is In 27,280 coding agent runs, 64% of inference spend went to runs that failed trending on Hacker News right now?

Not at the moment. It was last on the list on Nov 18, 14:43 UTC, after 2 days of trending.

What is the highest rank In 27,280 coding agent runs, 64% of inference spend went to runs that failed reached?

#1 on Hacker News · Front page, on Nov 17, 16:43 UTC. Trendoday checks this list every 15 minutes.

How long has In 27,280 coding agent runs, 64% of inference spend went to runs that failed been trending?

It was first seen on Nov 17, 2026 and has been on the list on 2 days (92 checks) since.

How much discussion did it get on Hacker News?

344 points and 120 comments; submitted by Edward40 on Nov 17, 2026.