When “Make a PPT” Becomes a Single Sentence: Wenxin Assistant Task Engine 2.0 and Its Zero-Barrier Promise
On August 20, at Baidu AI Day, He Jingzhou, head of the Baidu App R&D Center, made a straightforward statement:
“Let everyone use AI for free, and make AI a basic capability that every user can access with ease.”
On the same day, Wenxin Assistant Task Engine 2.0 officially launched.
What Baidu is trying to redefine is not simply an AI assistant. It is the basic interaction model between people and AI: instead of teaching users how to operate a tool, let users simply describe what they want—and let AI take care of the rest.
From “Helping You Find” to “Helping You Finish”
For decades, the traditional search experience has followed a familiar pattern:
Enter keywords → receive links → read → filter → understand → organize → judge → execute.
Search engines solve the problem of finding information, but the final result still depends heavily on the user.
Wenxin Assistant Task Engine 2.0 is attempting to move the endpoint one step further—from information retrieval to actual delivery.
According to Wenxin Assistant product manager Wang Yichun, the new engine covers more than 400 professional skills, increases the usability rate of personalized generation by 50%, and improves overall efficiency by approximately three times through end-to-end delivery.
Those capabilities span more than 30 industries.
But the most intuitive example for ordinary users may be surprisingly simple:
“Make me a PPT.”
Making a PPT Without Starting From Scratch
The traditional PPT workflow is familiar to almost everyone:
Think about the structure → search for information → organize the information → write the copy → find a template → format the slides → revise → revise again → revise one more time.
A simple presentation can take half a day. A complicated business deck can take several days—or even a week.
Task Engine 2.0 changes the workflow completely.
Imagine entering a request such as:
“Create an M&A analysis report for the technology sector.”
Instead of handing the user a pile of search results, the engine can coordinate multiple steps in sequence:
Brainstorming → data collection → information synthesis → document generation → PPT creation.
The entire process is designed as an end-to-end loop:
Information Collection → Processing → Creation → Final Delivery
The key difference is that the user is no longer responsible for manually connecting every step.
The expected output is not simply a collection of references or a rough draft. It is a usable deliverable that can continue to be refined through natural-language instructions.
And PPT creation is only one example.
Task Engine 2.0 also targets professional analysis reports, quantitative trading interfaces, scheduled tasks, and other complex workflows.
For example, a user could request a simulated quantitative trading terminal based on historical CSI 300 data, including daily backtesting, a signal dashboard, and a strategy engine.
That represents a meaningful shift in AI productivity tools:
The AI is no longer merely generating content. It is assembling a working result.
Free Is the Real Signal
The most interesting aspect of Task Engine 2.0 may not be what it can do, but how users are allowed to access it.
In 2026, paid AI subscriptions have become increasingly common. Claude Pro and ChatGPT Plus, for example, are both positioned around a $20 monthly subscription, while enterprise AI APIs typically charge according to usage.
Baidu is taking a different route with Task Engine 2.0:
Free access.
This is not simply about giving something away.
It reflects a broader infrastructure strategy.
Search engines became valuable not because users paid for every search, but because free access encouraged massive usage and kept users inside the ecosystem.
Baidu appears to be applying a similar logic to AI:
If users can freely ask AI to create presentations, write reports, analyze information, or build functional tools, then AI gradually becomes a default layer of the Baidu ecosystem rather than another standalone subscription product.
The usage data points in the same direction. In June, Wenxin Assistant's daily active users reportedly increased 83% year over year, while average daily conversation rounds more than doubled.
The strategic question therefore changes from:
“How much can we charge each user?”
to:
“How much AI can each user consume?”
That is a fundamentally different growth model.
English Version Part 2
AI Agent App Stores Are Redefining the Office Productivity Category
The launch of Wenxin Assistant Task Engine 2.0 also provides an interesting case study for the emerging AI Agent application marketplace.
Over the past several months, numerous products have entered the AI productivity race.
Kimi's Office Pilot, for example, focuses on interoperability between PPT, Word, Excel, and PDF workflows.
Other AI products are targeting automated presentation generation, document creation, and conversational editing.
Microsoft's ecosystem is also increasingly incorporating AI agents directly into productivity software, allowing users to create and modify presentations through natural-language instructions.
But many of these products share a common model:
Subscription fees or usage-based pricing.
Wenxin Assistant Task Engine 2.0 takes a different position:
Free + End-to-End
The difference is subtle but important.
It is not:
“I'll help you find the information, and you finish the work.”
It is closer to:
“Tell me what you need, and I'll deliver the result.”
Nor is it simply:
“Here is an AI-generated first draft. You finish editing it.”
The ambition is:
“Let the system run the workflow and deliver something you can actually use.”
This could become an important benchmark for AI office agents.
Users may gradually stop asking whether an AI tool can generate text, slides, or charts.
Instead, they may ask a much more practical question:
Can it complete the entire task?
The Hidden Infrastructure Question: What Happens to Compute Costs?
There is another side to this story.
Every time a user says:
“Create a PPT for me.”
the system consumes computing resources.
Every research step, document generation, reasoning process, tool call, revision, and final rendering requires model inference—and therefore tokens and compute.
That creates an interesting paradox.
The easier AI becomes to use, the more compute users are likely to consume.
And when the service is free, the incentive to experiment becomes even stronger.
A user who would hesitate to spend $10 on an AI workflow may have no reason to hesitate when the same workflow is free.
This is where infrastructure becomes critical.
This is the problem Unexhub is trying to address.
Unexhub positions itself as a Token aggregation and distribution compute trading center, connecting different AI model providers and making compute resources available through a more flexible market mechanism.
Its proposed model includes several layers:
Instant settlement — suppliers can receive their share as token usage is generated, reducing the cash-flow pressure associated with long settlement cycles.
API bidding — prices can move dynamically with supply and demand rather than remaining locked to a single static price list.
Early-stage zero commission — qualifying early suppliers can enter the ecosystem without platform commission during the initial period.
Multi-provider aggregation — developers and businesses can access models from different providers through a unified infrastructure.
Millisecond-level routing — requests can be routed according to model availability, pricing, and performance requirements.
For developers, the potential benefit is straightforward: instead of being locked into one provider's pricing structure, they can access a broader supply pool and seek more competitive token costs.
Unexhub states that developers can pay based on actual token consumption, with eligible usage potentially reaching as low as 30% of prevailing market prices, while suppliers can connect their APIs and receive usage-based revenue through real-time settlement.
The value proposition is therefore not simply “cheap tokens.”
It is about creating a marketplace in which developers optimize costs while suppliers improve monetization efficiency.
Two Sides of the Same Transformation
Wenxin Assistant is lowering the human-side barrier.
Unexhub is attempting to lower the infrastructure-side barrier.
At the application layer:
“Make a PPT” → “I know how to use a design tool” → “I know how to write prompts” → “I can simply describe what I want.”
At the infrastructure layer:
“Buy compute from one provider” → “Compare multiple providers” → “Access aggregated supply” → “Let the market discover the price.”
Put these two changes together, and a broader transformation becomes visible:
The cost structure of AI office automation is being redesigned from both ends.
AI agents are reducing the amount of human skill required to complete a task.
Compute marketplaces are attempting to reduce the amount of infrastructure cost required to execute that task.
Let's Talk About the Practical Value
The launch of Task Engine 2.0 ultimately raises a very simple question:
If AI can complete an entire office task from beginning to end, how much should that capability cost?
Baidu's answer is striking:
For users, it can be free.
That does not necessarily mean every AI agent will eventually be free.
But it establishes a new baseline.
Users may begin to expect that basic AI productivity capabilities—creating presentations, drafting reports, processing information, generating structured documents—should be available with almost no technical barrier.
The combination of 400+ professional skills, end-to-end task execution, and free access makes this more than a conventional AI assistant upgrade.
It is a statement about product design:
The interface to AI does not have to be a tool. It can be a task.
The Real Competition Is Moving From Models to Workflows
Of course, free does not mean perfect.
The quality of complex presentations, the rigor of professional reports, the reliability of quantitative trading tools, and the consistency of long-running agent workflows all need to be validated through real-world usage.
But the importance of Task Engine 2.0 lies elsewhere.
It lowers the cost of experimentation almost to zero.
Users who previously thought:
“I don't know how to make a PPT.”
may now simply say:
“Make one for me.”
That small change in language represents a much larger change in computing.
The next generation of AI competition may no longer be primarily about who has the strongest model.
It may be about:
Who can understand the task?
Who can execute the entire workflow?
Who can deliver the best result?
And who can do it at the lowest sustainable cost?
Baidu is lowering the application barrier.
Unexhub is attempting to reshape the underlying token market.
And somewhere between the two, a new AI economy is beginning to take shape:
Humans describe the outcome.
Agents execute the work.
Compute becomes a tradable resource.
When “make a PPT” becomes a single sentence, the real revolution is not that AI can make slides.
It is that users are gradually being freed from the need to know how the work gets done.
中文版 Part 1
当“做 PPT”变成一句话:文心助手任务引擎 2.0 的零门槛宣言
8 月 20 日,在百度 AI Day 开放日上,百度 APP 研发中心负责人何晶舟提出了一句非常直接的判断:
“让所有人都能免费使用 AI,让 AI 成为每个用户都能轻松获取的基础能力。”
就在同一天,文心助手任务引擎 2.0 正式上线。
这一次,百度想改变的已经不只是一个 AI 助手,而是人与 AI 之间的交互方式:
过去是学习如何使用工具;现在是直接告诉 AI 想要什么。
至于“怎么做”,交给 AI。
从“帮你找到”到“帮你做到”
过去几十年,搜索引擎的逻辑一直非常明确:
输入关键词 → 获得链接 → 阅读 → 筛选 → 理解 → 整理 → 判断 → 执行。
搜索引擎解决的是“找到信息”的问题,但从信息到最终结果之间,仍然存在一条很长的人工操作链。
文心助手任务引擎 2.0 想做的,是把这个终点再向前推进一步:
从“信息获取”走向“结果交付”。
据文心助手产品经理王亦春介绍,新引擎覆盖 400 余项专业技能,个性化生成可用率提升 50%,通过端到端交付,整体效率提升约 3 倍。
这些能力覆盖 30 多个行业。
但对于普通用户来说,最容易理解的场景可能只有三个字:
“做 PPT。”
做 PPT,不再从零开始拼素材
传统做 PPT 的流程,几乎每个职场人都经历过:
确定主题 → 搭建大纲 → 搜集资料 → 整理信息 → 撰写文案 → 寻找模板 → 排版 → 修改 → 再修改 → 继续修改。
简单的 PPT 可能需要半天。
复杂的商业汇报、行业分析或者投资报告,则可能需要几天甚至一周。
而任务引擎 2.0 试图改变整个流程。
例如,用户只需要输入:
“帮我制作一份科技领域并购分析报告。”
系统就可以围绕这个目标自动组织多个步骤:
头脑风暴 → 数据采集 → 信息整合 → 文档生成 → PPT 制作。
整个流程被串成一个端到端闭环:
信息收集 → 加工处理 → 创作表达 → 成品交付
用户不再需要自己把一个个工具连接起来。
最终拿到的也不只是搜索结果、资料链接或者一份粗糙初稿,而是一份可以直接使用,并能够继续通过自然语言修改的交付物。
而 PPT 只是其中一个场景。
任务引擎 2.0 同时覆盖专业分析报告、量化交易终端、定时任务等更加复杂的工作流。
例如,用户可以要求系统生成一个模拟量化交易终端,接入沪深 300 历史数据,并提供每日回测、信号面板和策略引擎。
这意味着 AI 办公工具正在发生一个重要变化:
AI 不再只是生成内容,而是开始组装并交付一个完整结果。
免费,才是真正的信号
任务引擎 2.0 最值得关注的地方,也许不是“它能做什么”,而是:
它让用户怎么使用。
进入 2026 年,付费 AI 订阅已经逐渐成为行业常态。
Claude Pro、ChatGPT Plus 等产品采用月度订阅模式,企业 API 则普遍按照 token 使用量计费。
但百度选择了另一条路线:
免费开放。
这并不只是“送给用户一个免费工具”。
背后其实是一套更接近基础设施的产品逻辑。
搜索引擎之所以能够形成庞大的用户生态,并不是因为用户为每一次搜索单独付费,而是因为低门槛甚至免费的使用方式,可以不断放大用户规模和使用频率。
百度正在尝试把类似逻辑应用到 AI 上。
如果用户可以免费让 AI 制作 PPT、撰写报告、整理资料、分析信息,甚至生成专业工具,那么 AI 就可能逐渐从一个需要单独购买的产品,变成百度生态中的一种基础能力。
从这个角度看,真正重要的问题就不再是:
“一个用户每个月能贡献多少收入?”
而变成:
“一个用户每天能够使用多少次 AI?”
这是一种完全不同的增长逻辑。
中文版 Part 2
AI Agent 应用商店,正在重新定义办公生产力
文心助手任务引擎 2.0 的出现,也为正在形成的 AI Agent 应用生态提供了一个非常清晰的样本。
过去几个月,AI 办公赛道已经出现越来越多的竞争者。
例如,Kimi 的 Office Pilot 主打 PPT、Word、Excel、PDF 等办公格式之间的协同处理;其他 AI 产品则重点布局自动生成演示文稿、文档制作和对话式编辑;微软生态也在不断把 AI Agent 直接嵌入生产力软件,让用户能够通过自然语言创建和修改演示文稿。
但这些产品大多存在一个共同逻辑:
订阅收费,或者按照使用量收费。
文心助手任务引擎 2.0 则选择了另一种路径:
免费 + 端到端。
区别其实非常明显。
过去是:
“我帮你找到资料,剩下的你自己做。”
现在更接近:
“你告诉我需要什么,我把结果交给你。”
过去是:
“这是 AI 帮你生成的初稿,你自己修改。”
现在则试图变成:
“我直接跑完整个流程,交付一份可以使用的结果。”
这可能成为未来 AI 办公 Agent 的一个重要衡量标准。
用户以后关注的可能不再只是:
“这个 AI 能不能生成 PPT?”
而是:
“它到底能不能把这件事完整做完?”
算力层的暗线:免费 AI 的成本由谁承担?
这里还有一个容易被忽略的问题。
用户每说一句:
“帮我做一份 PPT。”
背后其实都意味着一次算力消耗。
资料检索、信息分析、模型推理、工具调用、内容生成、格式调整、最终输出,每一个环节都可能产生 token 消耗和推理成本。
于是,一个非常有意思的悖论出现了:
AI 越容易使用,用户就越容易增加使用量。
而当 AI 免费之后,这种增长可能更加明显。
一个用户可能不会为了一个小任务单独支付 10 美元,但如果这个任务完全免费,他就没有理由不尝试。
因此,真正的挑战并不只是在应用层。
基础设施成本同样需要被重新计算。
这正是 Unexhub 试图解决的问题。
Unexhub 将自身定位为 Token 聚合分发的算力交易中心,希望连接不同规模的 AI 模型与 Token 供应商,通过更加灵活的市场机制,让开发者和企业获得更多算力供给选择。
其核心机制包括:
即时结算:供应商产生 Token 调用后,可以更快获得对应收益,降低传统账期带来的现金流压力。
API 竞价模式:价格可以根据供需关系动态变化,而不是长期锁定在单一厂商的固定价目表上。
早期入驻零佣金:早期加入的供应商可享受零佣金权益,降低进入平台的成本。
多模型聚合:将不同模型和供应商汇聚到统一的交易与分发体系中。
毫秒级路由:根据模型、价格、供给和性能需求进行请求路由。
对于开发者而言,这种模式的核心价值并不只是“更便宜”。
更重要的是:
开发者可以拥有更多供应选择,供应商也拥有更多变现渠道。
Unexhub 提供的机制,是让开发者按照实际 Token 使用量进行支付;在符合条件的场景下,开发者最低可享受约为市场价格 3 折的使用成本,同时供应商可以通过 API 接入渠道,并获得实时调用分成。
换句话说,它试图建立的并不是简单的“低价 Token 平台”,而是一个:
让开发者降低采购成本,同时让供应商提高算力变现效率的 Token 交易市场。
两端同时发生的成本革命
把文心助手和 Unexhub 放在一起看,会发现两者其实在解决同一个问题的不同侧面。
文心助手降低的是人的成本。
Unexhub 尝试降低的是机器的成本。
在应用层:
“做 PPT” → “会使用设计工具” → “会写提示词” → “只需要说清楚想要什么”。
在基础设施层:
“向一家厂商购买算力” → “比较多个供应商” → “进入聚合供给池” → “让市场发现价格”。
两种变化叠加之后,一个更大的趋势开始出现:
AI 办公自动化的成本结构,正在从两端同时被重构。
应用层削平人的技能门槛。
基础设施层则试图削平机器的成本门槛。
说点实在的:免费意味着什么?
文心助手任务引擎 2.0 最终其实在回答一个非常简单的问题:
如果 AI 可以从头到尾完成一项办公任务,那么这项能力到底应该卖多少钱?
百度给出的答案非常直接:
对用户来说,可以免费。
这并不意味着所有 AI Agent 最终都会免费。
但它至少建立了一个新的体验基线。
当用户已经习惯:
一句话生成 PPT;
一句话生成分析报告;
一句话整理资料;
一句话完成复杂办公任务;
那么未来用户对 AI 产品的期待自然也会发生变化。
400+ 项专业技能、端到端任务执行、低门槛使用方式,这些已经不只是营销概念,而是在重新定义 AI 办公助手应该做到什么程度。
真正的变化其实只有一句话:
AI 的入口不一定再是一款工具,而可以直接是一项任务。
真正的竞争,正在从“模型”转向“工作流”
当然,免费并不意味着完美。
复杂 PPT 的设计质量、专业报告的逻辑严谨度、量化交易工具的实际可用性,以及长链路 Agent 的稳定性,都需要经过真实场景验证。
但任务引擎 2.0 真正重要的地方,也许并不在于它是否已经解决了所有问题。
而在于:
它把用户尝试 AI 的成本降到了非常低。
过去,一个不会做 PPT 的人可能会说:
“我不会做 PPT。”
现在,他可能只需要说:
“帮我做一份 PPT。”
看起来只是少了几个操作步骤。
实际上,这是一次交互方式的改变。
未来 AI 的竞争,也许不再只是比较:
谁的模型参数更大?
而是比较:
谁更懂任务?
谁能完成完整工作流?
谁交付的结果更好?
谁能以更低的可持续成本完成它?
百度正在降低应用层的门槛。
Unexhub 则试图重新组织底层 Token 市场。
而两者之间,正在形成一种新的 AI 经济结构:
人负责描述结果。
Agent 负责执行任务。
算力则逐渐成为一种可以被聚合、交易和动态定价的资源。
当“做 PPT”真的变成一句话时,真正发生改变的并不是 AI 学会了做幻灯片。
而是:
人开始不再需要知道“这件事究竟应该怎么做”,只需要告诉 AI“自己想要什么”。
这或许才是零门槛 AI 办公时代真正值得关注的变化。