_______ __ _______
| | |.---.-..----.| |--..-----..----. | | |.-----..--.--.--..-----.
| || _ || __|| < | -__|| _| | || -__|| | | ||__ --|
|___|___||___._||____||__|__||_____||__| |__|____||_____||________||_____|
on Gopher (inofficial)
HTML Visit Hacker News on the Web
COMMENT PAGE FOR:
HTML Minions: Stripe's one-shot, end-to-end coding agents â Stripe Dot Dev Blog
iepathos wrote 6 hours 7 min ago:
"1000 PRs/week" with no breakdown of complexity or value is a vanity
metric. If these are mostly migrations, boilerplate, and bug fixes on
previous Minion PRs that were bug ridden, then you've just created 1000
code reviews/week to waste human time rubber-stamping. That's not
productivity, that's busywork with extra steps.
It's like measuring productivity by how many people you pull into
meetings each week. The CIA's Simple Sabotage Field Manual literally
recommends holding as many meetings as possible with as many people as
possible. The CIA should add "open as many PRs with AI as possible" to
their list. Bonus sabotage points if the PRs are made from ambiguous
"one-shot" attempts described in Slack with no follow up clarification.
yellow_lead wrote 6 hours 49 min ago:
> The Leverage team builds surprisingly delightful internal products
that Stripes can leverage to supercharge their productivity.
The Leverage team kind of sounds like the Department of Government
Efficiency
ericyd wrote 6 hours 56 min ago:
Thousands of PRs a week generated by AI and requiring human review
sounds like a ton, I wonder what their PR merge rate was before this?
ramon156 wrote 7 hours 4 min ago:
Stripe has become a weird company on my opinion. I'm glad Mollie is an
option that does not force me into certain technical choices.
alembic_fumes wrote 7 hours 11 min ago:
> Over a thousand pull requests merged each week at Stripe are
completely minion-produced, and while theyâre human-reviewed, they
contain no human-written code.
I pity the senior engineer, demoted from a helmsman into a human
breakwater, tasked to stand steady against an ever-swelling sea of AI
slop.
lelanthran wrote 7 hours 9 min ago:
>> Over a thousand pull requests merged each week at Stripe are
completely minion-produced, and while theyâre human-reviewed, they
contain no human-written code
> I pity the senior engineer, demoted from a helmsman into a human
breakwater, tasked to stand steady against an ever-swelling sea of AI
slop.
I'm skeptical that the human-in-the-loop, whose only task is to read
code, is going to be able to review at the rate that the AI can
produce.
It's Undefined Behaviour, now in every language.
oakpond wrote 7 hours 14 min ago:
Hardly anything substantial about how well this works in practice. It's
a hiring ad.
xnx wrote 6 hours 32 min ago:
> It's a hiring ad.
And also a project to pad someone's resume.
MarcLore wrote 7 hours 36 min ago:
The emphasis on one-shot execution is interesting. Most agent
frameworks still rely on iterative loops with human checkpoints, but
Stripe's approach of giving the agent a complete context dump upfront
and letting it run seems closer to how senior engineers actually work -
you read the whole PR/spec first, then write the code. The tricky part
is always the context window: once your codebase exceeds what fits in
context, one-shot falls apart and you're back to chunked reasoning.
Curious if they hit that wall and how they handle repo-scale tasks.
torginus wrote 6 hours 58 min ago:
My experiments using a more hands off approach of prompting claude
code have always resulted in a two-steps-forward-one-step-back play,
where the agent clearly did some good stuff, but did some other stuff
in a somewhat undesirable manner, which subsequently needed
correcting.
This usually results in A: creating commits where tons of code is
being constantly added and removed, B: due to Claude's somewhat
cavalier attitude to existing code, has steadily eroded my
familiarity with the code base.
I'm still not convinced that these longer loops are that beneficial,
compared to 1min prompts to 5-10min AI work.
OneMorePerson wrote 7 hours 16 min ago:
Doesn't delegating make this a lot more possible? You can fire off a
request to a sub-agent, they respond with some predictable status
that you can parse, and then you continue (you being the "main"
agent), so the context window can remain relatively small. Kinda like
how a human does it.
crimsonnoodle58 wrote 7 hours 37 min ago:
I've thought about implementing the same at our company. Something that
iterates through all our tickets, one shots them and creates PRs.
But humans are still left to review the code in the end, and as a
developer, code reviewing is one of my least favourite things..
I'm not sure I could spend the rest of my career just reviewing code,
and never writing it. And I'm not sure my team would either. They would
go insane.
As developers, by nature, we are creative. We like to solve problems.
Thats why we do what we do each day. We get a thrill when we solve the
problem, test it and it actually works. When we see it in production
and users enjoying it. When we see the CPU usage go from 99% to 5%.
I fear we are soon becoming nothing more than the last validation step
between AI and reality. And once AI becomes reality, which is very
soon, the days of development as we knew it will be over.
caseyohara wrote 6 hours 41 min ago:
I reckon the developers most excited about AI & agents never got the
same thrill or satisfaction that you do. Those developers are plainly
motivated by different things, and thatâs okay.
throwatdem12311 wrote 6 hours 48 min ago:
One thing I donât see developers talking about much is that if your
job is to only read code instead of writing it, how do you expect to
stay good at reviewing code if you never write it?
I only speak for me but when I review code I need to dig into my own
experience writing and and remember what works and what doesnât
that Iâve internalized over years of writing and manually debugging
code. Take that out of the equation and I wouldnât be good at
reviewing code for long.
I used to write a lot of C++ back in the day, and I can still read it
and understand it for the most part but I would never be able to
effectively review anything non-trivial. I just donât have enough
recent experience writing it myself to have internalized all of the
obscure pitfalls and gotchas. And just vommitting out some C++ from
a bot and just having it redo things until it has the appearance of
working correctly isnât gonna help me with that.
âMy job now is just reviewing codeâ is such an extremey
short-sighted view Iâm terrified for the future where nobody
understands anything anymore. Iâm sure OpenAI and Anthropic would
love this though.
And yeah, reviewing code is one of the more tedious and unfun parts
of the job why would I want this?
One of the most annoying parts of my job is my supervisor who used to
be a dev but became a manager years ago. He doesnât really
understand the codebase enough anymore and I spend so much explaining
basic things to him now it actually hinders our productivity when he
wants to âcontributeâ. And let me just say that getting a Claude
sub for the whole team hasnât helped this at all.
And one last thing - every single engineer I know that needs to
maintain a Stripe integration hates them with the power of a million
suns.
iLoveOncall wrote 7 hours 38 min ago:
I can't think of a less ergonomic way to submit a task than to write a
huge Slack message with links and references everywhere.
This really puts the final nail in the coffin that was the legend that
Slack developers trigger a minion from their phone during their
commute.
It's also funny that they mention they used goose [1] as a starting
point. I discovered them at a conference, and quickly realized that
nobody was using that crap, to the point that literally every testimony
on their website is from their own team.
HTML [1]: https://github.com/block/goose
dist-epoch wrote 6 hours 42 min ago:
The best camera is the camera you have on you.
Smartphones have terrible camera ergonomics, yet they killed the
compact dedicated camera.
maximinus_thrax wrote 7 hours 39 min ago:
> The Leverage team builds surprisingly delightful internal products
that Stripes can leverage to supercharge their productivity.
Why does this sound so insufferable?
zdragnar wrote 6 hours 14 min ago:
None of the adjectives are literal, aside from "internal".
The whole thing is meant to play on your emotions, not convince your
mind.
gassi wrote 7 hours 14 min ago:
Sparkling DevEx
amelius wrote 7 hours 42 min ago:
Who came up with the idea to slowly change the color of selected text?
A minion?
ColinEberhardt wrote 7 hours 18 min ago:
The same minion that came up with the cute effect that covers your
screen with the word DEVELOPERS, when you scroll to the end of an
article?
amelius wrote 7 hours 7 min ago:
I didn't read that far. Reminds me of Steve Ballmer: [1] Was this
video part of the training set?
HTML [1]: https://www.youtube.com/watch?v=Vhh_GeBPOhs
mangoman wrote 7 hours 45 min ago:
Thereâs something off-putting about making a blog post about some
splashy tech thatâs is a fork of an open source project, and that
tech not also being open source? It reads to me like âHey, we thought
the open source goose project was just okay, so we forked it to do it
better. But weâre not going to contribute it back to and instead
rename it.â
I think it probably wouldnât be as weird if the project were a
meaningfully different fork of it, but it sounds like itâs trying to
accomplish the same goals as the open source project which I feel
should probably be ported back? and renaming it seems sorta
ungrateful? Kinda like that âyou made this? I made thisâ meme.
Maybe I just donât have an understanding of how different the
projects are thoughâ¦
surajrmal wrote 5 hours 47 min ago:
They seem to have just optimized its integration with their existing
tooling and workflows. That doesn't sound largely useful to the
broader community. It's also probably different enough from goose at
this point that rebranding it makes sense. I do think such
integrations are hugely important for productivity and usefulness of
this sort of tool. It seems like the post is advocating for doing
deep 1p integration to further improve the utility of coding agents.
hombre_fatal wrote 5 hours 51 min ago:
How would they contribute back when their fork is a customization of
how it works?
> Weâve customized the orchestration flow in an opinionated way to
interleave agent loops and deterministic code
Is goose in such disrepair that you can just drop code changes into
it and the smol developer auto-accepts it, happy that anyone is doing
the work?
Or is goose actually it's own project with 250 issues and 74 PRs and
might have its own ideas about how it's built?
nigger238 wrote 6 hours 47 min ago:
You should always expect this when you make MIT. When you give every
company permission to fork your software, make it proprietary, and
sue you for copying or reverse engineering it, don't be surprised
when they do exactly that. It's in their best interest, after all.
jcims wrote 6 hours 47 min ago:
I donât know enough about either but if their approach was to make
it substantially more opinionated, which is likely in the case of an
org thatâs subject to audits, it would make sense to keep it
separate.
nkohari wrote 6 hours 55 min ago:
I don't have specific information about Minions, but I do know about
Stripe's architecture and internal tooling.
The article isn't really talking about changes they made to goose,
it's describing how they went about integrating goose with the rest
of their developer infrastructure (ie. the AWS-based remote devbox
system, Toolshed, etc).
Hasnep wrote 7 hours 8 min ago:
Welcome to the free software movement!
oytis wrote 6 hours 53 min ago:
Copyleft license would not help if they are only using it
internally
post-it wrote 6 hours 47 min ago:
AGPL would
candiddevmike wrote 6 hours 44 min ago:
Do you have access to Stripes minion service so you can demand
the source code?
No copy left license requires contributing your changes.
unfunco wrote 7 hours 42 min ago:
â¦and you can get almost identical features by simply installing the
GitHub app inside Slack, and then asking Copilot to work on
something, this should take < 5m to set up for any organisation using
Slack and GitHub.
firtoz wrote 7 hours 42 min ago:
If they didn't violate the licence agreement then I'm struggling to
understand why it's off putting
dkersten wrote 7 hours 0 min ago:
Just because itâs legal and allowed doesnât mean itâs not off
putting.
Personally, I have no issue with them making their own internal
fork, but then blogging about their thing without contributing it
back leaves a little bad taste. If itâs so good, then contribute
it back, since they benefited from the volunteers.
dist-epoch wrote 6 hours 45 min ago:
You can't have it both ways. As a library author choose MIT to
encourage commercial usage because companies are afraid of GPL,
but then complain that companies are actually using it in a MIT
license way without contributing back.
thn-gap wrote 6 hours 36 min ago:
License it GPL, and it will be fed to a model as training data
to recreate it copyright free anyways.
orangecoffee wrote 6 hours 26 min ago:
Training falls outside of copyright concerns because of fair
use, so proprietary or free is orthogonal. This is how the
world is currently trending.
citizenkeen wrote 7 hours 2 min ago:
You donât have to agree that itâs off-putting, but if youâre
âstruggling to understand whyâ that demonstrates a serious lack
of empathy and awareness of social dynamics.
jMyles wrote 7 hours 22 min ago:
> If they didn't violate the licence agreement then I'm struggling
to understand why it's off putting
What? Who cares about the license agreement? Lawyers and
bureaucrats maybe. The real issue with _any_ software project is
whether it is meant to be a step toward a more livable and peaceful
world or not. Sure, some people make guided missile software to
murder people for profit, but that's just obviously antisocial
behavior, regardless of how well it complies with license
agreements.
Hasnep wrote 7 hours 2 min ago:
If you put up a sign on your house saying "businesses, feel free
to come use my driveway for whatever you want" and McDonald's
sets up a restaurant there then you won't have much sympathy from
me.
toyg wrote 7 hours 29 min ago:
Law, spirit of the law, common decency. Rare currency these days, I
know...
jongjong wrote 7 hours 52 min ago:
Unfortunately, I don't use Stripe products because they discriminated
against me by blocking my account because my project used a Blockchain
(which I built myself) as an authentication mechanism.
It's discrimination because Blockchain tech is part of my religious
beliefs... Why is it so that less intelligent people who believe that
there is a man in the sky watching over them have protection against
discrimination but I don't? Yet my beliefs are grounded in science and
an actual understanding of our socio-economic system. I deserve more
protection, not less!
Does the law require that one's beliefs be irrational in order to
benefit from discrimination protections?
greazy wrote 7 hours 30 min ago:
> Blockchain tech is part of my religious beliefs.
What are your religious beliefs? I'm intrigued to hear more.
jongjong wrote 7 hours 18 min ago:
I believe the monetary system is broken and creates asymmetric
monetary playing fields based on distance from monetary injection
points (banks and governments). The tax system makes it hard for
each unit of currency to travel far from a 'money printer'. After
just 6 hops, a dollar is taxed down to about 10 cents; so people
who are more than 6 hops from a money printer live in a much more
scarce monetary environment than people who are in the front row.
It's Cantillon effects on steroids. It means that the entire
economy has become a kind of social climbing game to get closer to
the money printers. I feel that this game is immoral and people
shouldn't be forced to participate. Private currencies should be
protected by law.
I essentially believe that the economy is fake. That people get
money due to mostly social factors and then make up plausible
narratives to explain their success in a way which omits all the
critical social elements... And these explanations sound plausible
to people in their social circle who are at a similar distance from
a money printer so the false beliefs and perceptive distortions are
socially validated.
I also believe I'm being persecuted and algorithms are suppressing
me for seeing through the scheme and for my ability to explain
complex issues simply.
ericyd wrote 6 hours 52 min ago:
This is different than a belief system protected by anti
discrimination laws
yieldcrv wrote 7 hours 39 min ago:
âBanks are afraid of only two things: regulators and their wives,
and theyâre more afraid of their wives than the regulatorsâ
I would retroactively make that quote gender neutral but they're
really not afraid of their husbands.
Financial institutions feel like blockchains donât have a clear
chronology of KYC/AML, they dont care about KYC/AML they care about
violating it for their relationship with the regulator.
jongjong wrote 7 hours 27 min ago:
I switched to a different, smaller payment provider. It was pretty
easy to switch. No problems at all there. I wonder why I even
wanted to use Stripe in the first place. You'd think with their
size they wouldn't have to fear regulators. These big companies
usually have all the regulators in their pocket.
yieldcrv wrote 7 hours 14 min ago:
Stripe is right in the middle
Smaller institutions take risks with a niche, and gun for
exceptions with the regulators that bigger institutions dont find
worthwhile to bother with
And the biggest institutions dgaf because their relationship with
the national government will never be broken
ohyoutravel wrote 7 hours 43 min ago:
Based on the above article about thousands of AI written PRs
littering their code base, you might replace âunfortunatelyâ with
âfortunately.â
vbs_redlof wrote 7 hours 48 min ago:
You were building on Tempo, and they deplatformed you?
jongjong wrote 7 hours 38 min ago:
No. My Blockchain had nothing to do with payments. The Blockchain
is for authentication (protection from fake accounts) and tokens
represented credits and licensing rights within the platform. The
token wasn't even listed on any marketplace.
embedding-shape wrote 7 hours 54 min ago:
Submitted five times so far: [1] Once with substantial discussions: [2]
(127 points | 2 days ago | 65 comments)
HTML [1]: https://hn.algolia.com/?dateRange=pastWeek&page=0&prefix=true&...
HTML [2]: https://news.ycombinator.com/item?id=47086557
DIR <- back to front page