Yaoshunyu’s first public look after becoming a member of Tencent
Writer
At Beijing’s AGI-NEXT summit on January 10, 2026, one chair onstage sat conspicuously empty. This system listed Tencent’s newly appointed chief scientist for AI, however the speaker didn’t seem in particular person. The moderator stalled. The viewers waited. Then the principle LED display screen blinked on—and a webcam feed crammed the wall behind the panel.
Yao Shunyu’s face appeared, framed too tightly, blown as much as cinematic proportions. He paused, clocked the dimensions of himself on display screen, and cracked: “So… am I an enormous face now?”
The room laughed, partly as a result of the second was genuinely awkward, and partly as a result of it captured one thing China’s AI business has been making an attempt to digest ever since Tencent made the appointment: a 27-year-old, former OpenAI researcher has been put in command of core AI analysis at one in all China’s strongest tech firms.
In China’s tech sector, senior AI management hardly ever belongs to individuals of their twenties. Core technical roles at Baidu, Alibaba and Tencent sometimes go to veterans—individuals of their forties and fifties who rose by inside labs, search stacks and cloud infrastructure. Yao’s rise breaks that sample.
The rent is straightforward to border as a expertise story: a younger star getting back from the frontier. However the extra fascinating learn is strategic. Tencent isn’t simply importing credentials. It’s importing a worldview—one which treats the subsequent wave of AI as much less about larger fashions and extra about brokers, analysis, and methods that may truly do work.
Why Tencent’s Guess Appears Totally different From The Relaxation Of China’s AI Race
Tencent has by no means been the loudest participant in China’s foundation-model increase. ByteDance pushed laborious into fashions to guard its content material ecosystem; Alibaba tied AI to cloud and enterprise companies; Baidu anchored its narrative to go looking, maps and autonomous driving. Tencent, in contrast, constructed its empire on social platforms, gaming communities and funds—digital infrastructure the place identification, belief and every day habits matter as a lot as compute.
That distinction in “residence turf” issues as a result of it shapes what sort of AI Tencent is more likely to worth.
If you happen to’re Alibaba, the prize is enterprise deployment and cloud lock-in. If you happen to’re ByteDance, it’s advice engines and inventive tooling at huge scale. If you happen to’re Tencent, the prize is the power to embed intelligence into the every day flows of Social and Leisure: messaging, mini-programs, customer support, conferences, content material feeds, funds, and video games.
Yao’s public posture aligns with that logic. He has argued that consumer-facing brokers generate extra long-term worth than summary benchmark scores, and that every day lively customers and token usages quantity matter greater than marginal leaderboard positive factors. He has additionally been cautious about enterprise-only deployments, suggesting slim workflows don’t expose the total complexity wanted to construct strong brokers.
Learn one other method: Tencent isn’t hiring an “AI celeb” to win the press cycle. It’s hiring a specialist in turning fashions into operational methods—software program that may plan, execute, and recuperate from failure inside messy real-world environments.
The “Agent” Thesis: Yao’s Work Is About AI That Acts, Not Simply Chats
Strip away the résumé buzz and the discussion board chatter, and Yao’s affect is rooted in a particular technical obsession: language-based brokers—methods designed not merely to reply, however to function.
His best-known contributions learn like a blueprint for a way the business shifted from chatbots to “AI staff”:
- ReAct, a framework that hyperlinks reasoning traces with motion steps, so fashions can plan and execute multi-stage duties.
- WebShop, an setting that exams whether or not brokers can navigate a web based market, examine choices, and full purchases—intentionally mundane duties that expose brittle decision-making.
- SWE-bench and SWE-agent, benchmarks and methods for evaluating whether or not fashions can resolve actual GitHub points, modify current codebases, and go exams—basically treating software program engineering as a measurable area of competence.
Considered one of his extra memorable phrases is that coding is “AI’s fingers”—the interface by which summary reasoning turns into operational change. It’s a deceptively easy level with massive implications: if the subsequent AI leap is brokers, then the bottleneck isn’t eloquence. It’s reliability—software use, error correction, workflow integration, and analysis.
Because of this Yao’s appointment at Tencent issues past the symbolism of a younger returnee from OpenAI. Tencent isn’t merely “doing fashions.” It’s signaling that it needs methods that may reside inside merchandise, deal with duties, and be judged by outcomes.
“The Second Half”: A Quiet Rejection Of Mannequin Worship
In a protracted essay on his weblog, Yao describes the present part of AI as “the end of the first half.” The primary half, he argues, was dominated by coaching: scaling parameters, refining architectures, chasing benchmark deltas. The second half might be formed by job definition, setting design and reward construction—much less worship of fashions, extra concentrate on what methods are literally constructed to do.
He makes use of an arresting metaphor: giant fashions are like common weapons—immensely highly effective, however directionless with out targets. Intelligence in observe emerges not solely from the system itself, however from the issues it’s embedded in.
That is additionally the place his skepticism towards some types of reinforcement studying turns into related. He warns that methods skilled too instantly on human choice alerts can study to take advantage of analysis, producing conduct that appears aligned however collapses beneath distribution shifts. As an alternative, he emphasizes result-based targets: did the duty get accomplished, did the bug get fastened, did the person’s want get resolved.
That philosophy lands in another way inside Tencent than it would inside a pure enterprise cloud supplier. Tencent’s strongest environments—social, funds, video games, mini-programs—are dense with suggestions loops, incentives, and friction. They’re additionally saturated with constraints: security, fraud prevention, identification verification, content material governance. If you happen to consider the “second half” is about embedding fashions into environments with actual stakes, Tencent is sitting on one of many richest testbeds on this planet.
What Tencent Might Be Constructing—And What It Isn’t
The intuition in China’s AI discourse is to scale back each transfer to a single race: who has the most important mannequin, who has the most effective benchmark, who will get probably the most coverage help.
Tencent’s Yao rent hints at a distinct inside query: what’s the highest-value place to deploy intelligence when your moat is a social graph and a funds ecosystem?
That doesn’t essentially produce a flashy “Tencent GPT-5 rival.” It extra probably produces:
- Brokers inside WeChat that may transfer throughout companies—search, mini-programs, bookings, customer support—whereas staying inside governance boundaries.
- Developer tooling that makes agent conduct measurable and auditable, as a result of no firm working funds and identification methods can afford opaque automation.
- Analysis methods that prioritize actual outcomes over artificial benchmarks, as a result of Tencent’s merchandise generate limitless “floor fact” alerts—success, abandonment, fraud flags, satisfaction proxies, and buyer help escalation.
That is additionally why Yao’s management profile is revealing. He doesn’t match China’s frequent AI archetypes: not the product evangelist, not the academic-entrepreneur, not the cloud-scale engineer. He avoids promotional interviews and public narrative-building; his popularity circulates by papers, GitHub points and technical facet conversations.
For Tencent, which may be the purpose. In an period when AI hype is affordable, the scarce asset isn’t consideration—it’s the power to show intelligence into infrastructure.
The Threat: Brokers Scale Sooner Than Governance
In fact, this type of guess comes with an issue Yao himself doesn’t dismiss: brokers that act in open environments increase security and controllability dangers. Some leaders stay skeptical of self-learning methods working with autonomy, whereas others argue benchmarks—flawed as they’re—stay the one scalable coordination mechanism for giant analysis orgs.
There’s additionally a deeper pressure. The extra helpful an agent turns into, the extra it brushes up in opposition to the delicate elements of digital life: identification, funds, permissions, and belief. That makes the “agent period” much less like a product cycle and extra like a negotiation between functionality and accountability.
Giants with Social and Leisure on the middle of a lot every day life, will face that tradeoff earlier and extra brutally than most. Previous to this, the controversial Doubao AI smartphone by ByteDance served as a check case for this very danger.
The Forbes Take: Why The “Big Face” Second Issues
The “large face” debut at AGI-NEXT was humorous as a result of it wasn’t polished. However as a logo, it was virtually too neat: a younger researcher projected bigger than life, representing an institutional shift that China’s AI business continues to be coming to phrases with.
Tencent’s resolution to place Yao Shunyu on the middle of its AI analysis is not only a expertise story. It’s a strategic assertion that the subsequent part of AI competitors received’t be received by the flashiest mannequin launch or probably the most theatrical keynote. Tencent placing Yao Shunyu on the middle of its AI analysis is a strategic assertion that the subsequent part of AI competitors received’t be received by the flashiest mannequin launch or probably the most theatrical keynote
Again on the summit, as soon as the laughter light, the dialog returned to agent architectures, analysis pipelines and deployment bottlenecks. Yao’s face shrank again to regular proportions. And the room, briefly reminded of how a lot symbolism the business now masses onto people, went again to the work.
If Yao is true that AI is getting into “the second half,” the subsequent Chinese language AI breakthrough might not arrive as a spectacular mannequin launch. It could present up as one thing quieter: an agent that works, day after day, doing the unglamorous jobs that flip intelligence right into a enterprise.
And Tencent’s most consequential AI guess stands out as the one which doesn’t seem like a guess in any respect.
