No AI slop: A manifesto

16 points by asteroid 19 hours ago on tildes | 6 comments

kmcgurty1 | 9 hours ago

Ironically, the AI engineer code summit website looks like slop https://ai.engineer/code/2026

shoelace | 9 hours ago

I'm happy to see any attempts to get to the heart of what is gained and lost though letting AI take over so much of the software development process. I find this really important to think about for the future of my own career, and this seems to me a solid and succinct* rundown of the bare minimum expectations for when we create software the modem way.

But it really is just a bare minimum set of conclusions – essentially just saying that we can't hope to review everything, so we should just focus on reviewing the overall design, the system that produces the software, and the output. That sounds sensible to me, and I'd expect most developers who have resigned themselves to an AI-heavy future would accept that this is what needs to happen. I really want to see best practices develop in terms of how to achieve that: what we should actually do to effectively plan a design in a way the LLM can't possibly screw up when we tell it to go nuts, what guardrails and linters work and to what extent the human should be in the loop for those, and what we as developers should be doing to review the output. A lot of talk online suggests that all three of these should also be pretty much done by agents! So I want to see more proposals on what a human in the loop should actually practically be doing with all three of these.

I want to spend my next years learning skills that will be useful for my career over the coming decades, but until I have a clear idea of what humans are even for in software development anymore, I can't really do that!

(*though other comments here suggest that just 22 bullet points are apparently too much for some people to bother reading, lol)

forked_bytes | an hour ago

  1. AI is the most useful tool most of us have ever worked with

No it isn't.

  1. Slop is when we put less effort into producing something than we ask the consumer to spend understanding it.

This is in the right direction, but slop doesn't necessarily require effort on the consumer's part.

  1. Unreviewed output isn't always necessarily bad.

Yes it is. Stop making excuses for slop.

  1. When a mistake shows up twice, we scale ourselves and our team by improving the system and its guardrails.

Why twice? Why not the first time?

i got bored half way through

Because it is boring. It also isn't going to change shit. And ultimately "outcomes over output" is the biggest nothingburger platitude to ever platitude. Everyone is trying to aim for the best outcome. No one would say that output is more important than outcome. We are just a little limited in our experimentation due to the monotonicity of time. The agile manifesto at least managed to do something useful and compare and contrast time spent on a project on conflicting ideas and prioritise the ones that they felt they had observed leading to better outcomes. Imagine if they'd lead with outcomes over planning. No shit Sherlock. If you asked anyone that question you know what the answer is. Moreover valuing outcomes over output doesn't suggest how to improve the output!

It is completely obvious for me in my workplace that we are doing more with less and that customers are better off because of coding harnesses powered by LLM, for now. Are the developers better off? Open question. Will it always be true on any time horizon that using LLMs beats not using LLMs? I have no idea and neither does anyone else so I am taking the short term as a proxy because it's the best I can do. The rest is a bunch of guff.

I summarized it with AI for us

The manifesto argues that AI should amplify human judgment rather than replace it, emphasizing quality, understanding, accountability, and meaningful outcomes over producing more work faster.

:)