# AI 企业的新模式：Steve Yegge 提出 Counting House，让 AI 在严密护栏下接管千万级业务

DevFeed: [AI 企业的新模式：Steve Yegge 提出 Counting House，让 AI 在严密护栏下接管千万级业务](<https://devfeed.tech/articles/ai-steve-yegge-counting-house-ai-109788.md>)

Original publisher: [Read original article](<https://tonybai.com/2026/10/31/steve-yegge-counting-house-ai-enterprise/>)

Author: Tony Bai

Published: 2026-10-30T22:00:00Z

Content type: article

Language: zh

Sources: [Tony Bai](<https://devfeed.tech/sources/tony-bai.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [agent orchestration](<https://devfeed.tech/topics/agent-orchestration.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-a428c850142a](<https://devfeed.tech/tags/ai-a428c850142a.md>), [api](<https://devfeed.tech/tags/api.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [confusionmatrix](<https://devfeed.tech/tags/confusionmatrix.md>), [costoferror](<https://devfeed.tech/tags/costoferror.md>), [countinghouse](<https://devfeed.tech/tags/countinghouse.md>), [deterministicorchestrator](<https://devfeed.tech/tags/deterministicorchestrator.md>), [digitalclerk](<https://devfeed.tech/tags/digitalclerk.md>), [enterpriseai](<https://devfeed.tech/tags/enterpriseai.md>), [go](<https://devfeed.tech/tags/go.md>), [goldset](<https://devfeed.tech/tags/goldset.md>), [harness](<https://devfeed.tech/tags/harness.md>), [harness-engineering](<https://devfeed.tech/tags/harness-engineering.md>), [highthroughput](<https://devfeed.tech/tags/highthroughput.md>), [idempotency](<https://devfeed.tech/tags/idempotency.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [middleware](<https://devfeed.tech/tags/middleware.md>), [precision](<https://devfeed.tech/tags/precision.md>), [recall](<https://devfeed.tech/tags/recall.md>), [safety](<https://devfeed.tech/tags/safety.md>), [saga](<https://devfeed.tech/tags/saga.md>), [saga-06ebb2f7c546](<https://devfeed.tech/tags/saga-06ebb2f7c546.md>), [sagapattern](<https://devfeed.tech/tags/sagapattern.md>), [statelessagent](<https://devfeed.tech/tags/statelessagent.md>), [steveyegge](<https://devfeed.tech/tags/steveyegge.md>), [tag-2b3122b544e9](<https://devfeed.tech/tags/tag-2b3122b544e9.md>), [tag-43bf51bf8ffb](<https://devfeed.tech/tags/tag-43bf51bf8ffb.md>), [tag-5acad3d0279b](<https://devfeed.tech/tags/tag-5acad3d0279b.md>), [tag-5f4fb9043aec](<https://devfeed.tech/tags/tag-5f4fb9043aec.md>), [tag-6073b9aeb854](<https://devfeed.tech/tags/tag-6073b9aeb854.md>), [tag-8691f6f2d889](<https://devfeed.tech/tags/tag-8691f6f2d889.md>), [tag-9ae76ab74b6f](<https://devfeed.tech/tags/tag-9ae76ab74b6f.md>), [tag-9eb0df9036c0](<https://devfeed.tech/tags/tag-9eb0df9036c0.md>), [tag-c4fc0fe165d0](<https://devfeed.tech/tags/tag-c4fc0fe165d0.md>), [tag-e868d312559a](<https://devfeed.tech/tags/tag-e868d312559a.md>), [tag-ec474cd01110](<https://devfeed.tech/tags/tag-ec474cd01110.md>), [tag-f70bb7f905e5](<https://devfeed.tech/tags/tag-f70bb7f905e5.md>), [temporal](<https://devfeed.tech/tags/temporal.md>), [tieredautonomy](<https://devfeed.tech/tags/tieredautonomy.md>), [token](<https://devfeed.tech/tags/token.md>), [wheelhouse](<https://devfeed.tech/tags/wheelhouse.md>)

## AI overview

Counting House proposes scaling enterprise AI by assigning stateless agents one business case at a time, with narrow permissions and task specific evaluation. Organizations first build human reviewed gold sets and measure error types and business costs, then grant autonomy in stages for each task, with the ability to reduce or disable it when results worsen. Models may propose structured actions, but a deterministic orchestrator checks eligibility, parameters, approvals and idempotency before execution, while managing failures and audit trails. The approach suits high volume work with limited, definable intents and usable historical cases; it requires ongoing evaluation, data preparation and cross team effort, and may not fit open ended work or settings lacking reliable human fallback. Its proposed durable asset is the organization's own business rules and evaluation system, which can help assess replacement models and their costs.

## Source excerpt

AI 时代，企业需要的未必都是能够自主规划、长期协作的智能 Agent。对于客服、银行、保险等每年处理数百万乃至数千万个业务案例的场景，Steve Yegge 提出的 Counting House 提供了另一种思路：让无状态的 AI 数字职员在严格的权限、评估与编排机制下工作。通过黄金样本、分级自主权、错误成本分析和确定性事务执行，企业可以逐步将 AI 从辅助工具升级为规模化业务处理系统。真正的竞争壁垒，也许不在于使用了哪个模型，而在于企业能否把自身业务规则转化为可验证、可持续演进的工程体系。