当AI成为商业规则制定者:从争议处理到支付系统的权力重构
当一个独立开发者用AI写出的争议回复系统开始赢回被银行扣下的款项,当支付平台开始用AI自动处理合规审查,我们看到的不仅是效率提升,而是商业权力结构的一次静默转移。
核心观点:AI正在从提高效率的工具演变为商业系统中实际执行规则、重构权力关系的操盘手,这一转变对中小企业和平台生态的影响远比表面的功能升级更为深刻。
如果问一个普通创业者,AI改变了他业务的哪一部分,他可能会说“写文案”“做设计”或者“写代码”。这些回答都没错,但它们只触及了表面。真正深刻的变化正在商业系统的底层发生——AI开始扮演规则执行者、权力仲裁者甚至政策制定者的角色。这不是科幻,而是正在发生的现实。
一个引人注目的案例来自独立开发者Pieter Levels。他在Twitter上分享了一个让他自己都感到震惊的成果:用“vibe coding”方式构建的争议响应系统,让他十年后第一次赢回了Stripe上的大额争议——一笔1199美元的款项。这个系统的运作方式听起来简单但意义深远:当客户发起争议时,网站收到Stripe的webhook通知,然后自动收集证据,生成包含用户全部操作记录的PDF文件,包括他们何时注册、在应用中做了什么。在最近一次案例中,用户使用了产品几个月、生成了数千张照片,然后试图通过银行拿回款项。
这个案例的魅力不在于技术复杂度——它本质上是一个自动化证据收集和生成系统——而在于它揭示了一个权力反转的契机。长期以来,争议处理一直是小型在线卖家的噩梦。争议不仅意味着退款损失,还有每次30美元的争议费——而且这笔费用只有在卖家赢回争议时才返还。更致命的是,如果争议率超过1%,卖家可能会被Stripe、Visa和Mastercard永久封禁,不仅是当前业务,甚至是个人的名字。这套系统本质上是一个对卖家极为不利的规则体系,因为争议的门槛极低,尤其在北美,银行App里一键即可发起争议。
Levels的突破在于,他用AI重新平衡了这个不平等。过去,手动处理每个争议需要大量时间和专业知识,因此大多数小卖家要么忽略争议(自动输掉),要么雇佣昂贵的法务团队(不现实)。现在,一个用自然语言描述的、AI驱动的自动响应系统,可以针对不同用户类型和活动模式生成极其详细且有力的证据。这不仅仅是效率提升,而是让以前实际上无法行使的“抗辩权”变得可行使。
这个案例的隐喻意义远远超越争议处理本身。它暗示了一个更广泛的趋势:AI正在让商业规则的实际执行者从“大玩家”扩展到“每个人”。过去,只有大型企业才能负担得起专门的合规团队、专业的争议响应、精细的客户风险管理。现在,AI让这些能力以近乎零成本复制到个人创业者手中。这是一种权力的去中心化,但它并非没有代价。
代价在于,规则本身可能变得更加复杂和不透明。当更多的参与者开始用AI驱动系统进行争议响应时,争议处理就不再是简单的“买家说没用,卖家说有用”,而变成了一场AI与AI的军备竞赛。买家可能也开始用AI生成更合理的争议理由,平台则必须用更复杂的AI去甄别真假。这种螺旋上升的结果,是商业系统中的“规则层”被彻底技术化,普通人越来越难以理解和解构这些规则。
这种趋势在更大的范围内已经显现。Karpathy在演讲中提到的“代理原生经济”(agent-native economy),正是对这种现象的理论化。他将产品和服务分解为传感器、执行器和逻辑,并指出这些组件现在可以分布在传统计算、AI计算和未来神经计算的范式之间。这意味着商业系统的设计逻辑正在根本改变:不是先有业务流程,再有技术实现,而是先有AI能理解的信息结构,再以此为基础设计业务流程。
一个具体的例子是Karpathy提出的“用.md技能代替.sh脚本”:与其写一个复杂的Bash脚本去安装软件,不如写一份详细的文字说明,让AI在运行时刻自适应地执行安装、调试和配置。这听起来像是一个技术优化,但它隐含的前提是:AI现在可以理解并执行高度语境化的指令。这意味着商业操作中许多曾经需要人类判断的环节——比如“在客户A的场景下应该先做什么”——可以交给AI处理。但这种委托并非中立,它意味着AI的训练数据和推理逻辑会内化为商业操作的“潜规则”。
反对者会指出,这种观点夸大了AI的能力。目前的LLM仍然存在严重的幻觉问题,它们在关键决策中的可靠性远未达到可以完全无人监管的程度。Levels的案例中,AI生成的证据仍然需要人类最后审阅(尽管他分享的截图显示系统已经做到了端到端)。而且,更强大的AI争议系统也可能被滥用,制造虚假证据,反而加剧不公平。
这些质疑点是合理的,但它们恰好说明了问题的核心:AI不是中性的工具,它正在成为商业规则的实际制定者和执行者,而且这一转变已经发生。问题不在于是否接受这个现实,而在于如何确保这个新规则体系是透明、公平、可解释的。当越来越多的商业决策——从贷款审批到争议处理到合规审查——由AI驱动时,我们需要新的“商业宪法”来界定AI的权限边界、人类参与的程度和问责机制。
从Levels的案例中,我们可以看到一个积极的信号:一个独立开发者用AI工具重新获得了对自己商业命运的掌控。但从更宏观的视角看,这也是一个警示:当规则越来越由技术决定,那些不掌握技术或不愿理解技术的人,将越来越被动地接受由别人定义的商业秩序。AI不是乌托邦,也不是反乌托邦,它是一面镜子,照出我们商业系统中既有的权力不平衡,并赋予我们工具去改变它——前提是我们愿意主动设计这种改变。
参考来源
- 🏆 For the first time in a decade on @Stripe I've started winning disputes with my vibe coded dispute responder
- I used to ignore disputes so I almost always lost them, now I've started winning, this one is the first big dispute for $1,199 USD!
- Whenever a dispute comes in, my site gets a webhook notice from Stripe, it then starts collecting evidence and generates a PDF with entire user's details, when they signed up, and most importantly what they did in the app
- In this case the user used the app for months, generated thousands of photos then tried to get the money back from their bank
- The evidence has to be REALLY detailed, and REALLY good, which is why it's perfect to vibe code it, you can get quite detailed with different types of users and activity on your app, and put that all in the PDF
- I'm shocked because I again I never would win disputes before
- People in US especially abuse the [ chargeback ] or [ dispute ] en masse, unlike the rest of the world, it's easily built into their banking app next to every transaction, so it's one tap to get free stuff. And why not? You get free stuff!
- It's destructive for business owners like me on many levels, if I get over 1% disputes on my account, I risk getting shutdown permanently by Stripe, Visa and MasterCard, like permanently for life, not just my business but on my personal name too, it's ruthless
- Disputes are also super expensive for business owners: you don't just pay back the amount they disputed, for every dispute you pay $30, which you only get back if you win!
- But with AI we can now create our own tools to fight back against dispute abuse and finally win! 🎉 - https://nitter.net/levelsio/status/2049847252680614105#m
- Fireside chat at Sequoia Ascent 2026 from a ~week ago. Some highlights:
- The first theme I tried to push on is that LLMs are about a lot more than just speeding up what existed before (e.g. coding). Three examples of new horizons:
- 1. menugen: an app that can be fully engulfed by LLMs, with no classical code needed: input an image, output an image and an LLM can natively do the thing.
- 2. install .md skills instead of install .sh scripts. Why create a complex Software 1.0 bash script for e.g. installing a piece of software if you can write the installation out in words and say "just show this to your LLM". The LLM is an advanced interpreter of English and can intelligently target installation to your setup, debug everything inline, etc.
- 3. LLM knowledge bases as an example of something that was *impossible* with classical code because it's computation over unstructured data (knowledge) from arbitrary sources and in arbitrary formats, including simply text articles etc.
- I pushed on these because in every new paradigm change, the obvious things are always in the realm of speeding up or somehow improving what existed, but here we have examples of functionality that either suddenly perhaps shouldn't even exist (1,2), or was fundamentally not possible before (3).
- The second (ongoing) theme is trying to explain the pattern of jaggedness in LLMs. How it can be true that a single artifact will simultaneously 1) coherently refactor a 100,000-line code base *and* 2) tell you to walk to the car wash to wash your car. I previously wrote about the source of this as having to do with verifiability of a domain, here I expand on this as having to also do with economics because revenue/TAM dictates what the frontier labs choose to package into training data distributions during RL. You're either in the data distribution (on the rails of the RL circuits) and flying or you're off-roading in the jungle with a machete, in relative terms. Still not 100% satisfied with this, but it's an ongoing struggle to build an accurate model of LLM capabilities if you wish to practically take advantage of their power while avoiding their pitfalls, which brings me to...
- Last theme is the agent-native economy. The decomposition of products and services into sensors, actuators and logic (split up across all of 1.0/2.0/3.0 computing paradigms), how we can make information maximally legible to LLMs, some words on the quickly emerging agentic engineering and its skill set, related hiring practices, etc., possibly even hints/dreams of fully neural computing handling the vast majority of computation with some help from (classical) CPU coprocessors. - https://nitter.net/karpathy/status/2049903821095354523#m
- 测试招募开启!《遗忘之海》剧情概念PV 「余烬」 - https://www.bilibili.com/video/BV17Y9UBYEPd