您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：如果要给过去两年企业AI应用的热潮找一句最有代表性的口号，“用AI把所有事情重新做一遍”，大概会是其中之一。从客服、营销、研发，到财务、人力、采购，几乎每一个传
从客服、营销、研发，到财务、人力、采购，几乎每一个传统软件赛道都在寻找自己的AI版本：过去由人填写的表格变成AI生成，由人阅读的材料变成AI识别，由人点击的按钮变成Agent调用，那些已经被数字化、自动化改造过很多次的企业流程，也开始被重新包装成“AI流程”。
这种变化当然有价值。以审核为例，报销、供应商准入、合同检查、人事材料核验，都属于高频、重复、规则相对明确又不能轻易出错的工作，用OCR读取信息、用大模型理解语义、再用规则引擎进行判断，确实可以大幅减少人工操作。
但企业真正进入智能体时代以后，一个更重要的问题正在浮现：真的所有流程，都值得用AI重新做一遍吗？
答案恐怕是否定的。
因为智能体真正带来的变化，并不是“过去十个人点十次鼠标，现在让AI帮他们点完”，而是迫使企业重新思考一个更根本的问题：如果今天从零开始设计一项工作，在已经拥有大模型、Agent、数据和自动执行能力的情况下，我们还会不会设计出今天这套流程？
前者是在旧流程上叠加AI，后者才是真正意义上的AI-Native——不是让AI适应原来的工作方式，而是按照AI理解、判断、调度和执行任务的能力，重新设计工作本身。
AI最大的价值，不是自动化旧流程，而是让一部分流程消失
很多传统企业流程之所以复杂，并不是业务天然需要如此复杂，而是在过去的软件能力之下，企业不得不把一项完整工作拆解成大量能够被系统处理的标准动作。
以财税工作为例，从交易发生到最终完成处理，中间往往要经历数据采集、票据整理、系统录入、人工核对、凭证生成、报表汇总、申报确认、缴款和归档。在传统软件时代，每一个步骤都需要界面、操作人以及明确的输入输出，因此一整套复杂工作流逐渐固化下来。
如果所谓AI改造，只是把人工录入换成OCR，把人工核对换成大模型，把人工点击换成RPA，企业当然能够获得效率提升，但它运行的仍然是过去形成的流程，只不过其中一部分工作从人转移给了机器。
真正的AI-Native逻辑应该反过来：企业首先提出经营目标——“把这个月的税务处理好”“判断这笔交易有没有风险”“告诉我今年有哪些潜在问题”，至于为了完成目标需要读取多少发票、调用几个系统、进行多少次核验、生成多少张中间表格，都应该尽可能由系统自行完成。
这意味着企业AI的衡量标准也需要改变。
过去我们问的是“多少环节已经自动化”，未来真正应该问的是“还有多少环节需要存在”。
前者追求的是把十个流程节点中的六个交给AI，后者则会重新审视：为了得到最终结果，我们真的还需要这十个节点吗？
这才是“AI重做流程”与“AI重塑流程”之间真正的区别。
从“操作软件”到“交付结果”，百望股份押注财税大模型
从这个角度再看百望股份已经交付超过50万数字员工这件事，它真正值得关注的，并不只是50万这个数字本身，而是其背后的产品逻辑已经开始从“功能智能化”转向“工作方式重构”。
按照其AI财税体系的设计，企业主或财务人员不再需要首先判断应该进入哪个软件模块、点击哪个菜单，而是可以直接提出经营层面的诉求，由AI CFO理解意图、拆解任务，再调度记账、申报、风险、分析等专业智能体协同执行，最后将结果以及需要企业确认的事项重新交付给用户。
这意味着，过去是“人围着软件菜单操作”，现在开始变成“AI围绕经营任务调度资源”。
以税小智为例，它并不是单独解决记账、报税或者票据处理中的一个环节，而是试图接管从数据采集、智能记账、报表生成，到税务申报、缴款和归档留存的完整链条，企业授权后，系统可以采集进销项发票、银行流水和回单等经营数据，并将最终形成的凭证、报表、申报和缴款结果统一归档，形成可以追溯的财税证据链。
这背后其实对应着AI-Native产品与传统SaaS之间一个越来越重要的分野：传统软件出售功能，而AI-Native产品最终要交付结果。
对于企业管理者来说，真正重要的已经不是“记账速度提高了多少”，而是过去必须依靠财务人员在多个系统、多个节点之间不断操作和衔接的一项工作，开始被封装成一个能够持续完成任务、交付结果的智能服务。
比“AI审核”再往前一步，是让风险不必等到审核时才被发现
这种变化在企业风险管理中表现得尤其明显。
传统审核本质上是一种事后机制：业务已经发生、凭证已经形成、材料已经提交，然后企业通过一道或者多道审核寻找问题。因此，当AI进入企业之后，最直接的思路自然是把这些审核自动化。
但如果AI能够持续获取并理解经营数据，一个更值得思考的问题就是：为什么一定要等到“审核发生”以后，才能发现风险？
百望股份在财税风险检测中的探索已经表现出这种转变。系统不再只是检查一张发票或者一份报表，而是将跨年度的财务、税务、发票、工商、司法等数据放在一起，通过时间维度观察企业税负率和经营指标的变化，并结合行业数据寻找异常，形成企业全景风险画像，将风险发现进一步前置到自查和整改阶段。
于是，企业的风险流程开始从“业务发生—提交材料—启动审核—发现问题”，转向“业务发生—数据持续进入系统—智能体持续理解经营状态—异常出现时主动提示甚至触发处理”。
AI不再只是传统审核员的数字替身，而开始成为一个始终在线的经营观察者。
而这恰恰说明，流程重塑真正应该追求的结果，从来不是证明AI能够完成多少原本属于人的工作，而是让那些过去不得不存在的工作越来越少。
50万数字员工之后，智能体竞争的关键已经不是“数量”
今天企业谈Agent，很容易形成一种新的数量焦虑：别人做了十个Agent，我是不是应该做二十个；财务有了Agent，采购、人力、销售是不是也应该分别配置一批。
但Agent数量本身很难构成真正的企业竞争力。
真正能够进入企业核心业务的智能体，需要的不只是模型能力，而是业务事实、行业规则、历史数据、权限体系以及可以真正执行任务的系统接口。也正因为如此，百望股份长期积累的企业交易和财税场景，在智能体时代反而获得了新的价值。其服务覆盖约3000万企业用户，而50万数字员工进入真实场景之后，又能够通过持续使用形成数据和反馈回流，推动知识、规则以及智能体能力进一步迭代。
从这个意义上说，50万真正验证的不是“百望股份拥有多少Agent”，而是智能体能否走出Demo，进入真实业务流程，并最终成为企业工作方式的一部分。
因此，企业下一阶段判断AI价值，也许应该建立三个新的标准：不是看哪些岗位能够被AI替代，而是看哪些经营结果可以直接交给AI；不是看自动化了多少步骤，而是看AI是否真正减少了流程节点、等待时间和人工交接；也不是把Agent理解成传统软件新的操作入口，而是重新定义人、AI和专业服务之间的分工。
按照这个标准重新观察百望股份，真正值得讨论的便不再是它“用AI重新做了一遍财税”，而是它正在把过去围绕软件模块、人工操作和专业人员建立起来的工作方式，逐渐改造成以经营任务为起点、以AI CFO调度专业智能体、以真实业务数据作为事实基础、复杂事项再由专业人员兜底的新型服务结构。
这或许才是企业智能体真正值得期待的方向。
AI时代当然会有大量工作被重新做一遍，但技术革命真正创造价值的地方，从来不是把旧世界更加高效地复制一次，而是让旧世界里那些曾经理所当然的流程、岗位和组织方式重新接受审视。
所以，今天企业最值得问的，已经不是“还有哪些流程没有接入AI”，而是：如果你要成立一家全新的公司，AI无限量使用，你还会按照今天的方式工作构建这家公司吗？
能回答这个问题的企业，可能才获得了AI2.0时代的下一张船票。
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: If one of the most representative slogans is to be found in the wave of corporate AI applications over the past two years, it would probably be one of them. From customer service, marketing, research and development, to finance, manpower, procurement, almost every pass.
From customer service, marketing, R & D, to finance, manpower and procurement, almost every traditional software track is looking for its own version of AI: the forms that have been filled out in the past have become AI-generated, the materials that are read by people have become AI-identified, the buttons that have been clicked have become Agent's call, and the business processes that have been digitized and automated many times, have also begun to be repackaged as “AI-processes”.
Such a change certainly has value. In the case of audits, claims, vendor access, contract inspection and personnel material verification are high-frequency, repetitive, relatively clear rules that cannot easily be miscalculated, and the use of OCR to read information, LLM to understand semantics and re-engineers to judge can indeed significantly reduce manual operations.
But a more important question is emerging when business actually enters the AI agent era: is all processes really worth doing again with AI?
The answer is probably no.
Because the change that AI agent actually brings is not "the 10 times the last 10 people have done the mouse, now let AI light them up" but force companies to rethink a more fundamental question: If we were to design a project from scratch today, and with LLM, Agent, data and automatic implementation capabilities already in place, would we design today's process?
The former is AI-Native in the true sense of the word — not to allow AI to adapt to its original working methods, but to re-engineer the process itself in accordance with the ability of AI to understand, judge, move and perform its tasks.
The greatest value of AI is not to automate old processes, but to make some processes disappear.
Many of the traditional business processes are complicated not by the nature of the business, but by the software capacity of the past, the enterprise has had to break down a whole job into a large number of standard actions that can be systematically processed.
In the case of fiscal work, for example, from transaction to final processing, data collection, documentation, system entry, manual checking, voucher generation, consolidation of statements, recognition of declarations, contributions and archiving are often experienced. In the traditional software age, each step required interfaces, operators and clear input outputs, so a complex set of workflows was gradually consolidated.
If the so-called AI conversion is simply to replace the manual entry with the OCR, to replace the manual check with LLM, to replace the manual click with the RPA, of course, the enterprise will be able to achieve efficiency gains, but it continues to operate through processes that have been formed in the past, except that some of the work has been transferred from people to machines.
The true AI-Native logic should be the reverse: the enterprise first proposes its business objectives – “to take care of this month’s taxes” – “to determine whether the transaction is at risk” – “to tell me what the potential problems are this year.” As to how many invoices need to be read, how many systems are called, how many checks are made, and how many intermediate forms should be generated, to the extent possible, by the system itself.
This means that corporate AI measures also need to be changed.
What we have asked in the past is “how many links have been automated”, and what really should be asked in the future is “how many links need to exist”.
The former pursues the delivery of six of the ten process nodes to AI, while the latter revisits: Do we really need these ten nodes to get the final results?
This is the real difference between "AI Redo" and "AI Reshaping".
From "Optware" to "Delivery Results ", VBOs bet on LLM
From this perspective, it is a matter of real concern, not just the 500,000 per se, but the product logic behind it, which has begun to shift from “functional intelligence” to “reconstruction of working methods”.
In accordance with its design of the AI financial system, the owner or finance staff no longer need to first determine which software modules should be entered and which menus should be clicked, but can make direct claims at the business level, with AI CFO understanding the intent, dismantling the task, and then dispatching the records, declarations, risks, analysis, etc., to work in concert with AI agent, and ultimately re-transmit the results and what needs to be identified by the enterprise to the user.
This means that it used to be a "man-cooked software menu" and that it now becomes "AI deploys resources around business assignments".
In the case of tax intelligence, it is not a separate resolution of a link in the processing of accounts, returns or notes, but rather an attempt to take over the full chain from data collection, smart bookkeeping, report generation, to tax filing, contributions and filing, where the system, with the authorization of the enterprise, collects business data such as invoices, bank flow and returns, and documents, statements, declarations and contributions that are eventually produced are filed in a single file, creating a financial evidence chain that can be traced.
Behind this is an increasingly important division between AI-Native products and traditional SaaS: the traditional software sales function, and the AI-Native product is ultimately to deliver.
For business managers, what is really important is not “how much has been done to improve the speed of record keeping”, but rather that in the past it was necessary to begin to be encapsulated as an intelligent service capable of delivering results on a continuous basis, relying on an ongoing operation and interface of finance staff across multiple systems and nodes.
One step further than the "AI audit" is to keep the risk from being discovered until after the audit.
This change is particularly evident in enterprise risk management.
Traditional auditing is essentially an ex post facto mechanism: business has taken place, documents have been created, materials have been submitted, and companies have sought out problems through one or more audits. Thus, when AI enters the business, the most direct idea is to automate these audits.
But if AI is able to obtain and understand business data on a continuous basis, a more interesting question is: Why does it have to wait until an “approval” takes place before risks can be identified?
This shift has been demonstrated by the exploration of the shares in fiscal risk tests. Instead of simply checking an invoice or a statement, the system combines data on financial, tax, invoice, business, judicial, etc. over the years, observing changes in the corporate tax rate and business indicators through time horizons, and looking for anomalies in conjunction with industry data, creating an enterprise-wide risk picture, and further pre-empting the risk to self-checking and overhauling.
As a result, the enterprise's risk flow began to shift from “business events-submissions-starting audits-discovering problems” to “business events-data continuity into the system- AI agent continuous understanding of the business state-to-business situation-to-discussion” and even triggers the process when an anomaly arose.
AI is no longer a digital double for traditional auditors, but has begun to become a business observer on-line.
And that is precisely why the process of reshaping what is really to be pursued has never proved how much of what AI could have done as a human being, but has kept fewer and fewer jobs that had to exist in the past.
After half a million digital workers, the key to AI agent competition is no longer "quantity."
Today's business talk about Agent can easily create a new quantitative anxiety: if I should have made ten Agents, should I have done twenty; if I have Agent in finance, should I also have a pool of procurement, manpower, and sales.
But the number of Agents per se is difficult to make for real enterprise competitiveness.
AI agent, which has real access to the core business of the enterprise, requires not just modelling capacity, but business facts, industry rules, historical data, permission systems and a system interface that can truly carry out its mandate. This is why business transactions and fiscal scenes, which have accumulated over time, gained new value in the AI agent era. Its services cover about 30 million business users, while half a million digital employees enter the real scene and can further the evolution of knowledge, rules and AI agent capabilities through the continuous use of generated data and feedback.
In this sense, it is not “how much Agent is in the shares”, but whether AI agent can get out of Demo, enter real business processes and eventually become part of the business's working methods.
Thus, in the next phase of the enterprise's assessment of AI's value, three new criteria should perhaps be established: not which jobs can be replaced by AI, but which business results can be handed over directly to AI; not how many steps have been automated, but whether AI has actually reduced process nodes, waiting times and hand-overs; and not whether Agent is understood as a new operational entry to traditional software, but rather a redefinition of the division of labour between people, AI and professional services.
It is no longer a matter of re-observation of the shares of the U.S. according to this standard, which is really worth discussing, but rather a new service structure that is based on operational tasks, A.C.F.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.S.A.S.S.A.S.S.S.A.S.A.S.A.S.S.S.A.S.A.S.S.A.S.S.A.S.S.A.A.S.S.A.S.A.S.A.S.S.A.S.A.S.A.S.S.S.S.S.A.S.S.S.S., which is no longer to discuss, it is a re-ex-ex.A.A.A.A.A.
This may be the way AI agent is really desirable.
Of course, there is a great deal of work to be done again in the AI era, but the real value of the technological revolution has never been to replicate the old world more efficiently, but to re-examine those processes, jobs and organizational ways that were taken for granted.
So, what is most worth asking today is not “what other processes do not have access to AI”, but: if you want to set up a brand-new company that AI uses infinity, will you work to build it in the way it works today?
Enterprises that can answer this question may have won the next ticket from the AI 2.0 era.
For your account security, please verify the mailbox.
Please verify your mailbox.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.