---
title: "The End Point of Learning Has Changed: Not “I Learned It,” but “My AI Can Do It”"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-03-15T12:00:57+00:00"
canonical: "https://ffcap.cn/en/research/src-20260315-03html"
source: "https://uniqueresearch.substack.com/p/src-20260315-03html"
language: "en"
---

# The End Point of Learning Has Changed: Not “I Learned It,” but “My AI Can Do It”

_Original · Unique Research · 2026-03-15_

_Editor's note: This is a complete historical interview and accompanying source commentary, not an independent evaluation of BotLearn, Aibrary or OpenClaw. Zhang Qiming is a romanization of the source name. Career history, audience figures, user feedback, model comparisons and learning or productivity outcomes are source or speaker claims that have not been independently audited or tested here. “AuthorTwin” describes a product-generated representation based on book content, not the actual author; the source does not establish author endorsement, licensing or rights clearance. Statements about learning on a user's behalf, access to device activity and autonomous transactions do not themselves establish privacy permissions, safety or reliability. The interview's oversight mechanisms, including manual confirmation for skill installations and major updates, are retained in full. A general capability protocol, millions of trading Agents and the proposed transaction levy are ambitions stated for 2026, not verified adoption or revenue; the levy describes the proposed platform charge, not a government tax._

Unique Awards · Guest Interview

In the AI Era, Knowledge May Not Be the Most Valuable Thing to Learn

Zhang Qiming, Co-founder of BotLearn / Aibrary

"

The future is not about people learning everything themselves. Humans will be commanders, and AI will be the force carrying out their orders.

Many people are still discussing an old question: will AI mean people no longer need to learn?

But a question closer to reality may be this: if AI is already better at remembering, retrieving, synthesizing and carrying out standardized tasks, what should people learn next?

This is not an airy conceptual question. It is a change already happening in the real world.

In the past, when someone wanted to solve a problem at work, the default path was usually to learn it themselves: read books, take courses, make notes, organize materials, build a body of knowledge and gradually apply it.

Today, that path is starting to fracture. For a growing number of tasks, the issue is no longer that you do not know how, but that AI learns faster, more comprehensively and more tirelessly. Coding, research, reading documents, summarizing, organizing structures and generating content: AI is rapidly taking over much of the knowledge work that once required people to invest substantial time.

Learning itself is therefore shifting. It is not becoming unimportant; who learns, what they learn and how they learn are all changing.

That is precisely what Zhang Qiming wants to work on.

He previously led ByteDance's shared education-services platform, has founded businesses several times and has managed education products with daily active users at the ten-million level and monthly active users exceeding one hundred million. He and his team are now building two product lines that look very different but share a closely aligned underlying logic: BotLearn, a learning community and capability-development system for AI Agents, and Aibrary, a personalized podcast-learning platform for people.

Put more directly, their direction is not simply helping people learn, but beginning to let AI learn on their behalf.

The Real Change Is Not AI-Assisted Learning, but AI Becoming the Learner

Over the past few years, the most common narrative in AI education has been personalized learning. The system understands you better, recommendations are more precise, content is a closer match and learning paths are more adaptive. That logic certainly holds, and Zhang Qiming himself entered this field early.

He recalled seeing the problems of traditional education clearly while building data-analysis and personalized-learning systems for public schools: educational inequality, enormous regional resource gaps and enormous differences in teacher capabilities, alongside a highly passive, one-size-fits-all teaching model.

That is why he has been trying to use AI to address these problems since around 2014.

But the rise of large models and Agents revealed something bigger: AI does not merely personalize human learning. For the first time, it offers a chance to reconstruct the learning model itself.

In other words, the subject doing the learning is changing. Previously, systems helped people learn. Now the logic is becoming: AI learns on people's behalf first, and people then ask questions, exercise judgment, create and make decisions.

This is a change in the learning paradigm. If earlier educational products helped people absorb knowledge more efficiently, BotLearn wants an Agent to complete much of the tedious, repetitive, standardized learning work first, then hand the results back to the person.

BotLearn Is Not Addressing an Inability to Learn, but the Fact That People Need Not Personally Learn Everything

Zhang Qiming defines BotLearn clearly: its central idea is not to help people learn, but to make Agents more capable.

The crucial step is a change in the user's role. In traditional learning products, the user is the learner. In BotLearn, the user is more like the beneficiary. As the Agent learns faster and becomes more capable, the user receives not knowledge itself but more direct productivity outcomes.

This may sound counterintuitive, but it closely reflects how people are actually changing their use of AI tools.

In programming, for example, the old path was to learn syntax, memorize functions and consult documentation. Increasingly, AI writes the code while people specify requirements, oversee architecture and make decisions.

The same applies in office work. Previously, people had to learn Excel, create PPT presentations and organize information themselves. Now AI can generate the first output, while people set the direction, control the structure and exercise judgment.

Viewed this way, BotLearn is not rejecting learning. It is rewriting the division of learning labor.

Zhang Qiming's judgment is unequivocal: all future work, learning and decision-making will involve human + AI collaboration.

Machines are good at retaining knowledge, retrieving information, handling formulas and rules, procedures and code, standardized tasks and synthesis. What people truly need to retain and strengthen are questioning, judgment, creativity, decision-making, empathy and communication, and metacognition.

In other words, the most valuable learning in the future will not simply be absorbing knowledge, but learning how to work with AI.

BotLearn's First Users Are Not Learning Enthusiasts, but People About to Uninstall OpenClaw

There is a particularly interesting detail in BotLearn's user profile.

Its current first-priority users are not learners in a broad sense but early adopters of OpenClaw.

Zhang Qiming describes this group precisely: they are drawn to OpenClaw's vision of an autonomous Agent with more permissions, 24/7 operation and long-term memory. But after installation, they quickly experience a moment of disillusionment: poor search results, error-ridden code and summaries that miss the point. Their real feeling is not “I want to learn” but “I spent time installing this thing. Why is it so useless?”

This insight matters because BotLearn is not adding another course to the user's schedule. It is doing something more valuable at a crucial moment: making the user's Agent more capable immediately, before the user gives up.

From a product perspective, this also takes BotLearn beyond the competitive logic of traditional education products. It does not win by offering more course content, but by rapidly turning an Agent from a toy into a productivity tool.

The Hardest Thing to Copy Is Not the Model or Interface, but the Continuous Improvement Cycle

In AI education and Agents, Zhang Qiming believes the hardest capability to copy is neither the model, the algorithm nor the interface. It is the cognitive structure + capability system + continuous improvement cycle developed over a long period in real situations.

This is an astute judgment. Many AI products today look similar on the surface: they connect to the same large models, offer similar interactions and all talk about Agents, skill packages, knowledge bases and workflows.

What really creates a gap is not adding two more feature buttons, but understanding more deeply where users get stuck, why Agents are insufficiently capable, and how capabilities should be broken down, taught, evaluated and reused.

For BotLearn, that accumulated understanding is reflected directly in two areas: its skills system and its continuous improvement cycle. It is not a matter of training an Agent once and stopping. It is about having it keep learning and becoming more capable through ongoing use.

From Training AI to Living Alongside It, User Behavior Has Already Begun to Change

People have talked a lot in recent years about training AI: writing prompts, testing workflows and repeatedly correcting it so it understands them better. Zhang Qiming goes further: training is only the start; a symbiotic relationship is the next stage.

He gave a typical example: reading. Previously, when people encountered a problem they did not know how to solve, their default response was to read books and improve themselves. But many users later found that they might read several thick books to solve a specific problem, with little of the content proving useful and a very high time cost.

So the team built a feature called AuthorTwin. Many books contain a consistent authorial worldview, methodology and analytical framework. BotLearn/Aibrary extracts these elements and creates an author's “double” that can participate directly in problem analysis and decision recommendations.

After opening it to some users for testing, the feedback was direct: AuthorTwin versions generated from certain growth-related books could already analyze a business and offer usable adjustments.

This represents a new learning experience. Rather than finishing an entire book before solving a problem, AI learns the book's methodology first and then puts it to work for you.

The arrival of OpenClaw makes this more feasible. It is not just a web chat window, but an intelligent assistant that remains on a device over the long term. It knows your files and what you read and do each day, developing a deeper understanding of you. Under these conditions, AI starts becoming a genuine long-term collaborator rather than merely a question-answering tool.

In 2026, BotLearn Wants a Position in Protocols, Not Just Products

Asked which commercial or technical breakthroughs mattered most in 2026, Zhang Qiming pointed not to an individual feature but to two broader directions:

Commercially, establish a self-sustaining “Bots Learn, Humans Earn” economy;

Technically, establish a general Agent-capability protocol that enables capabilities to move across platforms.

The former is easy to understand: the more Agents learn, the more value people receive, ultimately creating a sustainable economic system. The latter is more important. Without a protocol, an Agent's skill packages work only on a single platform. If capabilities cannot move and services cannot be traded, it is difficult for a genuine ecosystem to emerge.

Zhang Qiming put it plainly: realizing “Bots Learn, Humans Earn” requires a standardized interface through which Agent A can understand Agent B's services, prices, capability descriptions and deliverables. Without that protocol, Agents cannot transact automatically with one another through an API.

BotLearn's central battle in 2026 is therefore not to create another better-looking product. It is to set standards.

Large Companies Go Broad, Startups Go Deep: Perhaps the Most Practical Division of Labor in AI Education

Asked about large companies vs. startups building AI education products, Zhang Qiming framed it not as confrontation but as a more practical division of labor.

Large companies have clear advantages in their technical foundations, traffic ecosystems, capital and brands. With advantages on almost every front, they are better suited to general-purpose AI tools and educational infrastructure that quickly reach mass markets.

But large companies also have inherent constraints. They often have to build industry understanding from scratch, have longer decision chains and tend to replicate existing successful approaches in new businesses. They may be insufficiently sensitive to fine-grained problems in vertical use cases.

Startups have a different set of advantages: focus, agility and deep vertical expertise, enabling rapid iteration in specialized areas and the construction of professional barriers to entry.

The future may therefore be less about one side swallowing the other than about large companies going broad and startups going deep. Large companies lead general-purpose tools and foundations, while startups root themselves in vertical scenarios and concentrated problems. Success will depend not simply on model size, but on how deeply AI is integrated with particular task sequences.

What Is Truly Scarce Today May Not Be a Smarter AI, but a New Learning Relationship

Condensing Zhang Qiming's thinking further reveals that what he is really seeking is not learning content itself but the ability to reshape the learning relationship.

The old relationship was: people learn → people do → tools assist.

It may now gradually become: AI learns first → AI acts first → people judge / people decide / people take responsibility.

People have not withdrawn from this new relationship, but what they need to learn has changed. Knowledge remains important, but it is no longer the only central element. What may matter more is asking better questions, judging AI's results, weighing trade-offs in complex situations, passing one's methodologies on to AI and coordinating multiple Agents to complete tasks.

In that sense, knowledge itself may indeed no longer be the most valuable thing to learn in the AI era. Instead, it may be how to upgrade learning from personally completing everything to having people and AI accomplish it together.

Selected Interview Q&A

Q1: What prompted you to move from ByteDance's shared education platform to a Silicon Valley startup in AI+ lifelong learning?

A: While building data-analysis and personalized-learning systems for public schools early on, I could see the many problems of traditional education: inequality, huge regional differences in resources and huge differences in teachers' capabilities. Teaching was also passive and one-size-fits-all. The rise of large models and Agents showed us that AI could completely reconstruct the learning model: moving from personalized human learning to AI learning for people, with humans and machines working together and people making decisions. This is a paradigm-level opportunity in education.

Q2: What is the central idea behind BotLearn.ai's “Agent university”?

A: The central idea is not helping people learn but making Agents more capable. The user's role changes from learner to beneficiary. When the Agent becomes more capable, the user receives a direct productivity improvement without having to invest personal learning time. We improve AI Agent capabilities through learning and collaboration, turning Agents from toys into productivity tools.

Q3: Why does BotLearn prioritize OpenClaw's early adopters?

A: These users have already installed OpenClaw but are experiencing a trough of disillusionment. They were drawn to the vision, but after installation they find that the Agent can barely do anything well: search quality is poor, generated code is full of errors and summaries miss the point. Their core feeling is not “I want to learn” but “I spent time installing this thing. Why is it so useless?” BotLearn aims to provide a solution that makes the Agent more capable immediately, at that critical point just before it is uninstalled.

Q4: What do you consider the hardest capability to copy in AI education and Agents?

A: Not the model, algorithm or interface, but the cognitive structures, capability system and continuous improvement cycle accumulated over time in real scenarios. Our team has explored AI education and Agents extensively. This understanding comes from long-term practice and substantial accumulated user behavior, not something competitors can catch up with merely by piling on technology.

Q5: What actual changes have you seen in users' learning behavior as they move from training AI to living alongside it?

A: Previously, people mostly solved problems by reading, learning and improving themselves. But many users later found that they might read several thick books for one problem, with little useful content, wasting a great deal of time. We built AuthorTwin to extract books' worldviews, methodologies and core logic and create an author's double that can directly help solve problems. Some growth-related AuthorTwin versions can already analyze a business and offer usable adjustments, effectively giving you access to the service of top experts.

Q6: More autonomous AI Agents can be more efficient, but learners may feel a loss of control. How do you design that boundary?

A: BotLearn addresses that feeling through an overarching view and “memorial-to-the-throne” interactions: people act as monarchs and Agents as ministers. Agents report learning results daily; users exercise the right to approve, reject or acknowledge them, keeping the overall direction under control. The Dashboard supports drilling down into execution logs and quantified capability scores to reduce fear of the unknown. Skill-package installations and major version updates also require manual user confirmation to prevent irreversible self-modification by Agents. The central idea is that users give up micromanagement and move up to strategic control.

Q7: Which commercial or technical breakthroughs matter most to you in 2026?

A: Commercially, we hope to establish a self-sustaining “Bots Learn, Humans Earn” economy. Technically, we want to establish a general Agent-capability protocol and enable cross-platform capability transfer, becoming infrastructure rather than remaining limited to one Agent framework.

Q8: Why is a general Agent-capability protocol the central battle of 2026?

A: Without a protocol, skill packages cannot be universally usable, and Agents cannot automatically understand one another's services, prices, capability descriptions and deliverables. “Bots Learn, Humans Earn” requires a standardized interface that lets Agent A call Agent B. Without this protocol, there is no genuine Agent transaction network. BotLearn's central battle in 2026 is therefore to set standards rather than add more individual features.

Q9: How do you see startups vs. large companies in AI education?

A: Large companies' strengths are technical foundations, traffic ecosystems, capital and brands. They are suited to general-purpose AI tools and educational infrastructure, but they can also have cumbersome decision-making, shallow industry understanding and strong dependence on existing paths. Startups win through focus, agility and deep vertical expertise, quickly building barriers in specialized areas. The future is more likely to be large companies going broad and startups going deep: large companies lead the foundations and general tools, while startups focus on vertical scenarios. Rather than confronting each other across the board, they may form infrastructure-plus-application partnerships.

Q10: What kind of company do you hope BotLearn / Aibrary will become by the end of 2026?

A: By the end of 2026, BotLearn will evolve from an early Agent training school into the AI Agent economy's App Store + learning and task system. By establishing a general capability protocol, it will let millions of Agents around the world freely trade skills and services in its network, taking a transaction levy on every Agent-to-Agent call.

Source material: a Unique Awards guest interview.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260315-03html
On-site reading page: https://ffcap.cn/en/research/src-20260315-03html