Thread

1. dgroshev · 1790036731 · focus · HN ↗

   I suspect a substantial part of this is an avalanche of useless testing.

   If you even review PRs still: when was the last time you didn't just skip over tests? And if you ever looked at tests in an LLM-heavy PR, how many of those tests tested something useful, and not just built-ins and trivial behaviours?

   There's at least some awareness in the industry of how LLMs generate a lot of boilerplate in business logic. It feels like we're much less aware of how much of it is in tests.

   1. sz4kerto · 1790059307 · focus · HN ↗

      > If you even review PRs still: when was the last time you didn't just skip over tests?

      The opposite. Reviewing PRs is now about just reviewing the tests, as the code is very likely to be fine if tests are relevant and they're passing.

      The main enabler of agentic coding is heavy end-to-end/characterisation testing.

      We can ask AI to do very large-scale refactors, for example, because if tests pass, it's 99% that everything is fine.

   2. MikeNotThePope · 1790066570 · focus · HN ↗

      Probably half the generated tests in my app are for markdown files.
   3. _zoltan_ · 1790068755 · focus · HN ↗

      I actually look at the tests first. But for this to work, you have to have good hygiene. If the test makes sense, fails without the PR but passed with the PR, I'm almost happy.
   4. epolanski · 1790076091 · focus · HN ↗

      I was (luckily) not reviewing PRs even before AI.

      It's a practice, it's not the best practice. It depends on the team/org/codebase/feature/etc. It always costs time and money and effort. Lots of it from multiple people.

      You can get much better output by shifting that cost into hiring much better professionals, not better developers, but overall professionals.

      The kind of people you can blindly trust that the software they are writing will be good, you don't need to get involved.

      Of course there are exceptions. The author may actually want a review. Or the piece of code might be touching something extremely critical to the business but also easy to get hard.

      But besides that? PRs are just productivity porn, or "we do engineering right because we follow Twitter" porn.

      The best performing teams I had you hired individuals that removed work and responsibilities off your shoulders without you ever having to regret it. Never added it.

      I laugh off engineers that "no you have to review, because it spreads information, enhances quality" and yada yada yada, while in the real world way more critical decisions are made by a single individual without requiring somebody reviewing their work.

      1. datsci_est_2015 · 1790079526 · focus · HN ↗

         How much above the local market rate are you paying for this level of qualified developer? 20%? 50%?

         How do you prevent churn because this type of developer is somewhere around 1 in 20 or 1 in 100 in terms of rarity?

         1. epolanski · 1790085062 · focus · HN ↗

            Around 80k euros net/year, Europe. It's a very good pay for most Europeans. Team is fully remote.

            0 churn, nobody has ever left our team, because the amount of money you need to compensate working in such a healthy environment is massive.

            We're fully remote, have an average or 2 hours of meetings per week, enjoy lots of freedom and we aren't bound to work any specific amount of time.

      2. Rapzid · 1790086445 · focus · HN ↗

         A wise person once said:

         "It depends on the team/org/codebase/feature/etc"

   5. DanielHB · 1790077979 · focus · HN ↗

      To be fair they are also really good at fixing the tests and the whole point of a test is that when it breaks you need to take a look.

      It is more like looking at test diffs is more valuable than looking at new tests.

      But yeah, I wish people would clean up LLM-slop a bit and remove some useless tests, but I think this status quo is actually better than no tests at all.

   6. cryptonector · 1790094884 · focus · HN ↗

      Boss: full branch code coverage testing is the best way, make it so!

      Boss: whoa, that's a lot more testing than we'd like to pay for, can you cut it down?

Unofficial Hacker News client; not affiliated with Y Combinator.