This stands up as long as you take it for granted that agentic engineering leads to 'worse outcomes' and the opposite of 'better work', and that it never 'improves the system'. To say the least, this is begging the question. But to give it a fair shake out of respect, regardless:
1. AI is the most useful tool most of us have ever worked with, but a swarm of agents can produce more work in a single day than a person can review in our lifetime.
Not controversial, nor an argument for anything.
2. Slop is when we put less effort into producing something than we ask the consumer to spend understanding it.
This is completely arbitrary, and useless in practice anyway; the bar for how much organic human effort should go into the programming of the Waymo app has no relation to how hard it is to order a Waymo.
And if this is intended as a formal definition of the term used throughout the rest of the "manifesto", it's even more plainly insufficient and contradictary. Like, take the very next point, which defines it diffently:
3. Using AI doesn't necessarily make something slop. Slop happens when we pass AI output along without taking responsibility for the quality or the reader's experience.
Classic motte-and-bailey; standing up against people who publish software without taking responsibility for it or thinking about the user is not something that needs a manifesto.
4. AI cannot replace human judgement, it can only amplify our judgement (or lack thereof).
Huge conclusion offered without even a passing attempt at justification. Unless you believe in souls that are bestowed to humans and humans alone, we are machines, and our thinking (judgement is only 1/4th of thinking btw) is clearly replaceable. Also, "amplifying" a quantity is the same thing as "replacing" parts of that quantity, which doesn't help.
Here's the first point where I'll remind everyone that the devil's soulless calculators proved the Riemann Theorem yesterday.
5. Slop is asymmetric. The cost of producing a thousand-line PR or a ten-page doc is approaching zero, while the cost of reviewing it remains constant.
6. In code, even before AI, slop proliferates when we repeatedly work around problems instead of solving them correctly. It comes from tools that lean HEAVILY toward adding rather than subtracting.
Truisms, as stated; strawmen, as implied.
7. A lot of slop comes from pressure. Companies are pushing their people to redefine what is humanly possible.
Not an argument against anything other than Capitalism (which, hey: preach!).
8. Slop compounds. The more slop sits in our knowledge base, in chat, in code, the more of it the next agent reads and copies.
9. The more slop flying around a team, the more we lose the will to fight it. Communication and quality get worse over time, not better.
Yes, if you assume that there is a thing called "slop" that is definitionally worse, it is bad to have it around. Begging the question.
10. Human time and attention is incredibly valuable, so we optimize for using it wisely.
11. We fight slop because AI should make our work better, not just easier.
Truisms, as stated; strawmen, as implied.
12. We optimize systems, tools, and products for efficient elicitation of human intent and for efficient human understanding.
Openly contradicts the basic premise of engineering, which is a practical science dictated by practical utility. This is only a good point if you take it for granted that anything that's not optimized for human understanding is always worse, eventually.
13. We do our own thinking when working with a model. We explore resources, code, and prior art to truly understand the problem before the work starts.
Truism. (Also, the term 'a model' is the reddest of red flags for someone not understanding the basics of agentic engineering.)
14. We own the intent, taste, and consequences of what we ship, however much AI helps us make it.
Truism, as stated; strawman, as implied.
15. When an agent wrote something we're sending, we make that clear, and we check the parts that matter before it goes out.
The first real stance! The latter part is a truism (how closely do you check it? Are you checking it as closely as you possibly could? Since you're not, isn't your line arbitrary), but the former is a great prescription. Untagged artificial prose (or code, I guess, when that's even coherent to the end user) is socially dangerous.
16. When someone sends us slop, we tell them, kindly, and ask them to review it first. Most of us have needed that same note at some point.
Ok. Nice to see a point that 100% of the people signing this manifesto would ignore in practice -- it's an AGILE manifesto tradition!
17. There will always be alpha in reviewing **something**. As models improve, we may stop reading most of the code. We may still review a plan, design, or result, because if we don't, we just get the same output as everyone else.
Contradicts the baley, but yes, this is the motte I've been repeatedly arguing. Spoiler alert: the models improved.
18. Agent reviews, linters, and verifiers are useful but not sufficient. For now, we still review at least some subset of the outputs. What subset depends on the stakes.
Completely meaningless prescription.
19. What shipped, who's using it, and how it's helping them matters FAR MORE than lines of code, prs merged, tokens spent, or the number of agents running in parallel.
20. Unreviewed output isn't always necessarily bad. One-shot prompts and vibe coding are high leverage tools. "Slop" is bad when it undermines our outcome -- a useful product, a maintainable system, or saving ourselves and our teammates time and energy.
21. When a mistake shows up twice, we scale ourselves and our team by improving the system and its guardrails. Anything a linter, type checker, or automated check can catch, we let it catch.
Truisms, as stated; strawmen, as implied.
22. None of us has this figured out. We're each relearning what we will and won't accept, and there's no single right answer.
A meta-motte, but I appreciate the humility -- I am certainly far less capable of an engineer than many of the people on this list, and have very little 'figured out' definitively.
But what this last point really does is the reason I felt the need to reply to this post in the first place: it makes room for the vitriolic, vindictive, and cruel.
"Slop" is not a technical term; it's a trite move in a rapidly-intensifying culture war that's somewhat unique to the US and EU.
Accordingly, this manifesto isn't technical guidance, it's a flag; the contradictions and clear post-hoc nature of it all don't matter, because the signers aren't using it as an epistemic tool.
[dupe] https://news.ycombinator.com/item?id=50488769
This stands up as long as you take it for granted that agentic engineering leads to 'worse outcomes' and the opposite of 'better work', and that it never 'improves the system'. To say the least, this is begging the question. But to give it a fair shake out of respect, regardless:
Not controversial, nor an argument for anything.
This is completely arbitrary, and useless in practice anyway; the bar for how much organic human effort should go into the programming of the Waymo app has no relation to how hard it is to order a Waymo. And if this is intended as a formal definition of the term used throughout the rest of the "manifesto", it's even more plainly insufficient and contradictary. Like, take the very next point, which defines it diffently:
Classic motte-and-bailey; standing up against people who publish software without taking responsibility for it or thinking about the user is not something that needs a manifesto.
Huge conclusion offered without even a passing attempt at justification. Unless you believe in souls that are bestowed to humans and humans alone, we are machines, and our thinking (judgement is only 1/4th of thinking btw) is clearly replaceable. Also, "amplifying" a quantity is the same thing as "replacing" parts of that quantity, which doesn't help.
Here's the first point where I'll remind everyone that the devil's soulless calculators proved the Riemann Theorem yesterday.
Truisms, as stated; strawmen, as implied.
Not an argument against anything other than Capitalism (which, hey: preach!).
Yes, if you assume that there is a thing called "slop" that is definitionally worse, it is bad to have it around. Begging the question.
Truisms, as stated; strawmen, as implied.
Openly contradicts the basic premise of engineering, which is a practical science dictated by practical utility. This is only a good point if you take it for granted that anything that's not optimized for human understanding is always worse, eventually.
Truism. (Also, the term 'a model' is the reddest of red flags for someone not understanding the basics of agentic engineering.)
Truism, as stated; strawman, as implied.
The first real stance! The latter part is a truism (how closely do you check it? Are you checking it as closely as you possibly could? Since you're not, isn't your line arbitrary), but the former is a great prescription. Untagged artificial prose (or code, I guess, when that's even coherent to the end user) is socially dangerous.
Ok. Nice to see a point that 100% of the people signing this manifesto would ignore in practice -- it's an AGILE manifesto tradition!
Contradicts the baley, but yes, this is the motte I've been repeatedly arguing. Spoiler alert: the models improved.
Completely meaningless prescription.
Truisms, as stated; strawmen, as implied.
A meta-motte, but I appreciate the humility -- I am certainly far less capable of an engineer than many of the people on this list, and have very little 'figured out' definitively.
But what this last point really does is the reason I felt the need to reply to this post in the first place: it makes room for the vitriolic, vindictive, and cruel. "Slop" is not a technical term; it's a trite move in a rapidly-intensifying culture war that's somewhat unique to the US and EU. Accordingly, this manifesto isn't technical guidance, it's a flag; the contradictions and clear post-hoc nature of it all don't matter, because the signers aren't using it as an epistemic tool.
TL;DR: :(