— How a change in a unit of measurement is rewriting the basic assumptions of cost, pricing, and organization
When the output of intelligence can for the first time be measured, compared, and traded, what will economics encounter? On the afternoon of October 26, CITIC Press Group and the Tencent Research Institute held the exchange forum 'Token Economy — When Intelligence Becomes a Measurable Service.' Guests from the economics community, research institutions, and the front lines of industry gave answers that were not consistent.
According to data from the National Data Administration, China's daily average Token call volume grew from about 100 billion in early 2024 to 500 trillion in June 2026 — growth of thousands of times over in two years. In actual applications, two phenomena — 'unit prices getting cheaper' and 'bills getting more expensive' — appear simultaneously, forming the core tension running through the entire discussion.
Token: The Emergence of a Unit of Measurement
Bai Huitian, one of the authors of Token Economy and a senior researcher at the Tencent Research Institute, views the Token as a lens for understanding this era: it represents a change in the unit of measurement — it is both the smallest unit for large models to process information and the smallest unit for billing products and services. Just as bushels, barrels, and kilowatt-hours are used to measure grain, oil, and electricity respectively, the Token is the first unit used to measure the output of machine intelligence.
Xiang Shuai, a well-known finance scholar and financial writer, also believes that people now say 'a drop of water' and 'a kilowatt-hour of electricity,' and will soon say Token as well. She added that intelligence is being stripped from humans and made into an industrial product, and the hallmark of industrialization is that products become more sophisticated while prices get cheaper.
But both also cautioned: one cannot look only at the quantity of Tokens. Bai Huitian gave an example: consuming the same number of Tokens to revise an email versus to discover loopholes in a contract and avoid losses of hundreds of thousands of yuan have completely different values — this is precisely the book's core judgment: cost is measured linearly by Tokens, but value unfolds non-linearly according to the consequences of the task.
When Supply Is No Longer Scarce, Where Does Value Come From?
Bai Huitian noted that intelligence can be supplied almost without limit, yet human attention remains finite: when the time to write an article is compressed to half an hour, the time saved may be filled with more work — people may not be liberated, but instead become busier.
Yuan Xiaohui, an author of Token Economy and founder of the New Possibility Lab, pushed the judgment one step further: with AI supplying intelligence, the bottleneck has shifted from 'whether it can be made' to 'whether it can become value.' She gave an example: a Pomodoro timer mini-program hand-built by two first-year middle school students in Shenzhen had beautiful visuals and received tens of thousands of likes, but no one has paid for it yet; AlphaFold made discovering protein structures easier, but whether it can be turned into targeted drugs still depends on the second half driven by real-world feedback. 'Innovation has not been solved by AI; that later segment is the bottleneck.'
Xiang Shuai believes that humans' 'specialized intelligence' has not disappeared — creativity is using unique life experience to call upon general intelligence. Therefore, demand lies not in traditional 'grand consensus,' but in micro scenarios: abstract needs must be discovered within scenarios, co-created with users, and consolidated into data, then compounded — her summarized 4C framework (Scenario, Co-creation, Consensus, Compounding).
Is the Token Economy Economic? One Challenge and Three Responses
At the roundtable forum, Li Gang, deputy director of the Tencent Research Institute, opened with his article 'The Token Is Not Economic,' raising four observations: the API prices of vendors' lightweight models have not fallen as expected, but have instead risen; Agents cause repeated context calls, wasting large numbers of Tokens; profits in the industry chain are increasingly concentrated upstream; and downstream players have invested ahead of finding stable cash-flow scenarios, so risk is accumulating in the industry chain.
Yuan Xiaohui pointed out that many cost problems need to be solved through engineering means: the price difference between a cache hit and a miss can be about 50 times, and the cost structure on the input side has been rewritten. Bai Huitian proposed that in their research, they observed enterprises beginning to make 'intelligent budget management' decisions: for the same task, is it more cost-effective to buy Tokens or to hire people.
Zhang Wenzhao, head of customer success for the northern region at Tencent WorkBuddy, provided empirical evidence in his keynote speech using first-hand AI bills. WorkBuddy is positioned as an 'all-scenario, AI-native enterprise workbench,' and in his view, office work is moving from tool-based work and collaborative work into the stage of digital-employee work. He shared three cases: Tencent's internal HR team, with more than 400 people, launched over 3,600 applications based on WorkBuddy within three months, whereas in the past requesting a system from the IT team often meant monthly delivery — now HR can build applications themselves; an employee at a manufacturing client used WorkBuddy to build production-line monitoring software in 4 days, saving the company over a million in software costs; a teacher in education used WorkBuddy to cut lesson preparation time from a full day to a few dozen minutes.
Zhang Wenzhao also offered a cost-pathway approach — 'first use it broadly, then use it deeply, and only then maximize the value of the bill.' For example, copywriting tasks need not call expensive models, while high-value tasks such as programming are worth using good models for.
Will Companies Still Exist in the Future?
Xiang Shuai believes that one of the most important reasons companies exist is to save transaction costs, and as digital tools continue to lower the cost of collaboration and intelligence can be called upon at any time, organizations will move toward flattening and dimensional reduction, all the way to the 'one-person company.' Humans, as combinations of skills, are more likely in the future to move toward 'matching of tasks and skills.' She emphasized that this is a 'factor revolution' — when intelligence becomes a basic input, organizational structures and modes of production must be reorganized around it, and the hardest part is organizational change.
Li Gang argued that because some functions of organizations are structural, they will not disappear because of AI — for example, the legal-person system caps liability at the amount of invested capital, giving failures the chance to stand up again; organizations can also share risk and undertake larger goals, and not everything is suited to a one-person company; organizations also carry social functions such as identity.
The differing views in the roundtable discussion together sketch a boundary: organizational forms will evolve but will not disappear; and what determines the effectiveness of the transformation is precisely the organization's own capacity for change — echoing a judgment in the book Token Economy: enterprises that only connect to APIs without making accompanying organizational changes see productivity gains close to zero.
The Place of People: From 'Able to Take Responsibility' to 'Not Falling Off the Tail'
The discussion ultimately returned to people themselves. Yuan Xiaohui argued that humanity's next comparative advantage is precisely being 'able to take responsibility' — Agents have infinite patience, but humans can be accountable for results. Li Gang also believes that genuine connections between people will become increasingly precious.
Xiang Shuai cautioned that technology inevitably presents a J-curve that first falls and then rises, and 'society needs a bit of damping': the engine has already sped ahead — how do we keep everyone from 'falling off the tail'?
This forum did not offer a final answer, but all the discussion landed in the same place: when intelligence becomes a measurable factor of production, what truly needs to be answered is not 'how much is a Token worth,' but where people are to be placed.
In January of this year, the president of the Tencent Research Institute proposed the claim that AI should 'put people at ease and amplify people.' In the context of the Token economy: putting people at ease means intelligent services need to be explainable and accountable, with someone responsible for the results; amplifying people means technology should expand, not replace, human judgment, creativity, and willingness to take responsibility.
From bushels, barrels, and kilowatt-hours to Tokens, the succession of units of measurement has never been merely a change in bookkeeping; what it must ultimately answer is still whether, after efficiency gains, people can obtain more disposable time, more reliable security, and fuller room for development. This is also precisely the landing point of the book Token Economy — the endpoint of the Token economy is still the human being.