The paradox of the acceleration era:
When "your AI" starts talking to "my AI"
In 2026, artificial intelligence is undergoing a profound role transition: from the "conversational interaction" of the past few years to a full-scale move toward "autonomous delivery."
In the past, AI lived in a chat box in a screen's side panel — you asked, it answered, offering reference suggestions; today, agents (Agents) are beginning to step out of the chat box and become truly embedded in daily workflows. As the AI office productivity agent with the largest user base in China, Tencent WorkBuddy has, after more than 100 version iterations, cumulatively handled over 9000 million task requests and driven Token consumption growth of more than 50 times; according to third-party AI product rankings' desktop data for the 8th month, the total monthly active users of domestic AI office agents on desktop has reached 33.16 million, with WorkBuddy ranking first on the overall list, its monthly active user scale accounting for more than one-third of this category.On the front lines of enterprise business, Mengniu has produced 45 "digital employees" through an internal AI innovation competition, cutting the time for single-farm diagnosis and treatment from 1.5 days to 2 hours; Shanghai Jahwa has deployed WorkBuddy at scale across 9 major departments, compressing the monthly reconciliation of 60 million lines of logistics bills, which used to take 4.5–6 person-days, to under 20 minutes.
Efficiency gains in single-point workflows are already an indisputable fact. But as we sprint along in AI's acceleration, a subtle workplace alienation is quietly taking place:
In many organizations that have adopted AI tools early, the "AI flavor" in reports and documents grows ever stronger. Often, one person uses AI to generate a lengthy, ten-thousand-word article in a few minutes and sends it to the other end; the colleague on the other end has no time to read it carefully, and uses their own AI to condense the long article into a three-point summary.
When communication evolves into "your AI talking to my AI," the volume of information expands severalfold, while genuine collisions of thought between people are diluted.
This is precisely the incisive question posed by the 2026 Tencent Global Digital Ecosystem Summit:Is a team full of "super individuals" automatically a "super team"?
The answer is usually not optimistic: not yet. If departments still rely on lengthy meetings to align on messaging, if the same solutions are still being reinvented across teams, and if key experience remains locked in the minds of a few—then the faster individuals move on their own, the more easily the torrents of information they create become a burden for the organization to digest and sift internally. Single-point efficiency goes up, while group collaboration is canceled out by friction.
When the entire industry is shouting "AI Native," perhaps we need a bit more clarity about being "People Native." As an AI technology company, Tencent has always upheld a clear, warm, and essential value proposition:"Put people at ease, amplify people."
At this year's Global Digital Ecosystem Summit, this proposition was embodied as a set centered on"people-oriented"as its axis:the "AI-native productivity flywheel"—driving efficiency gains with AI-native workflows, ensuring manageability and controllability with an AI-native technical architecture, and promoting human-machine collaboration and the emergence of innovation with an AI-native organization. Behind this flywheel lies a deep dialectic of the agent era about "fast and slow," "machines and people," and "the individual and the community."
Putting people at ease:
Only when the foundation is "manageable and controllable" can
technology move toward "scaled application" in organizations
Why, in the flywheel of enterprise evolution, must the technical architecture immediately take the baton to achieve "efficient operation, manageable and controllable" once workflows are running? Because when AI shifts from "chatting with people" to "doing things for people," the logic of trust between humans and machines fundamentally changes.
When generative AI first appeared, people's attitude toward it was like watching a toddler learning to walk—it could write poems and paint, occasionally answered off-topic, and users found it novel, with great tolerance. But when task requests in workflows explode into the billions, and agents truly enter the core business flows of enterprises, starting to revise code, check accounts, send notifications, and run approvals on people's behalf, tolerance gives way to prudence: Will it overstep and see privacy it shouldn't? If its logic goes off track in the background, who will be accountable for the results?
Speed changes the way we work, but must never lower engineering and institutional standards; stability, security, and availability remain non-negotiable long-term assets in the digital era. Without a "manageable and controllable" technical architecture, AI can forever remain at small-scale personal experimentation and never move toward enterprise-level "scaled application."
In human society, we dare to entrust a matter to a companion not because that person will never make mistakes, but because there are clear boundaries, fallback mechanisms, and transparent communication between us. Tencent's reconstruction of its underlying AI-native technical architecture (Agent Infra), if we strip away the technical shell, is in humanistic and governance terms actually establishing a triple institutional contract about "peace of mind":
First, a well-measured boundary of permissions—"running errands on your behalf, but never overstepping authority."The greatest taboo between a person and an agent is blurred boundaries. In Tencent's security system and WeCom mechanisms, there is an iron rule: an agent may execute tasks on a person's behalf, but never inherits all of the person's permissions by default. It acts only within the scope of authorization of "you" yourself—documents you normally cannot see, groups you cannot enter, the AI likewise cannot touch; and once sensitive actions are involved, such as sending messages externally or modifying core configurations, the system must stop and wait for you personally to complete a second confirmation in the client.
Second, a fault-tolerant space that provides a safety net—"leave room for exploration, install fences around mistakes."Whether human innovation or agent exploration, trial and error is indispensable. Many organizations hesitate to introduce AI at scale because in production systems they "cannot afford mistakes." To this end, the underlying open-source security sandbox opens up an independent "digital workstation" for each agent, drawing four rigid fences around computing, networking, data, and credentials—“Do not interfere with its flexibility; only keep its uncontrollable parts under control.”At the same time, the environment state supports saving and rolling back at any time: even if the agent takes a wrong path while experimenting, it can return to the previous moment and retry via a different route, strictly locking the cost of trial and error inside the fence.
Third, there is a traceable, transparent process—“open the black box so that actions can be traced.”The unknown most easily breeds suspicion. Traditional software monitoring only records a single line of error logs, while an agent's “poor answer” often produces no error at all. Through a full-chain observability platform, the agent's background task execution, knowledge retrieval, and tool invocation trajectories can be fully reconstructed. Only when we can see how it made its judgments step by step and clearly identify which link the deviation occurred in does the machine cease to be a disturbing black box.
At its core, the essence of “putting people at ease” is to establish, in human–machine collaboration, a deep“human-in-the-loop”—no matter how fast the agent runs, at critical decision points, people can always see, understand, and intervene. The steering wheel and the final “braking power” always remain in human hands.
Scaling up from the individual:
From “tool efficiency” to “amplified capability,”
human autonomy and “slowing down in three ways”
Once the underlying guardrails establish a sense of security, individuals gain the freedom to step away from tedious tasks. But this brings another question about self-worth: if machines write faster than I do and compute more accurately than I can, where exactly is the place for humans?
Some time ago, workplace discourse was filled with anxiety-inducing discussions: some worried about the “collapse of the middle layer,” while others debated whether it is the young who, armed with new tools, will phase out the experienced, or the experienced who will suppress the young with their expertise.
Yet in real industrial practice, we have seen a completely different answer: regardless of past experience, background, or age, the real dividing line has never been seniority, but rather who possesses“Agency”—who knows how to use an AI-oriented way of thinking to redefine problems and mobilize resources.
In Tencent's “people-centered” value framework, AI has never been meant to replace or diminish people, but to transform the human role from a “mineral deposit” of data and a “cog” in processes back into the “purpose” of service. This is exactly the core proposition advocated by the“Human Career Evolution Program”at this conference:“Amplify capability, evolve careers.”
In AI-native workflows, this amplification of individual capability manifests as a clear division of labor between humans and machines:WorkBuddy takes on repetitive, deterministic execution work such as cross-system operations, organizing meeting minutes, and verifying reports; meanwhile, humans step back and return to thinking, creativity, challenging, questioning, acceptance, and evaluation.Whether in human resources, administration, legal affairs, wealth management and investment, or content creation, creative design, academic research, and marketing, every professional can summon a domain expert at any time on their dedicated workbench, turning industry know-how into capability modules at hand, and through daily collaboration, letting the digital assistant better and better understand one's work habits.
When the threshold of "how to execute" is greatly flattened, and when generation itself becomes cheap, humanity's core competitiveness is comprehensively shifting from the memorization and accumulation of external knowledge to the tempering of internal qualities. In an era where AI is accelerating the generation of everything, people precisely need to "slow down" on three things:
First, slow down to polish your "Taste".Ten proposals and a hundred lines of code can be generated instantly, but which proposal truly hits the mark? Which architecture truly withstands scrutiny? AI has raised the baseline water level of daily output, but the ceiling of work results always depends on human aesthetic intuition, value judgment, and the wisdom of trade-offs. Whether one can make the right choices and truly get things done ultimately tests a person's taste.
Second, slow down to "use the human brain well and speak human language well".The more readily tools are available, the more one must guard against intellectual laziness. Shedding the burden of mechanical formatting is not meant to make people surrender their right to think, but to keep people from being swept along by dazzling new concepts and to return to first principles — to ask what real problem something solves and what real value it creates. Cut away the clichés and redundancy, and distill deep thinking into the most sincere, most essential viewpoints.
Third, slow down to build "high-density interpersonal trust".Intelligent agents can gather materials and generate meeting minutes for you, but in real social and business interactions, the chemistry that sparks between people face to face, and that power of mutual understanding and empathy in complex situations, cannot be directly generated by any algorithm. The more advanced technology becomes, the more precious genuine human warmth is.
Amplifying the organization:
Integrating into the existing organizational fabric,
thickening assets and spurring the emergence of innovation in human-machine collaboration
When individual capability is amplified and the underlying security architecture supports scaled application, the flywheel engages its most critical link: how to truly evolve from single-point "AI-native workflows" into a collective"AI-native organization", and break out of the vicious cycle of "individual efficiency gains, organizational friction"?
Over the past few years, many enterprises' digital or AI transformations have fizzled out halfway, often because they fell into a radical fantasy of tearing everything down and starting over. But from the perspective of organizational sociology, any normally functioning enterprise is an organism composed of complex division of labor, historical tacit understandings, and business inertia. Today, there is not yet a universally applicable "optimal organizational structure for the AI era",and the evolution toward an AI-native organization is destined to be a long process. No enterprise can afford the cost of "smashing the existing organizational structure, business processes, and IT systems and rebuilding everything from scratch".
Therefore,first seamlessly integrating into the existing organizational system and business processes through an open, connected "AI-native workflow" is not a compromise of expediency, but the most urgent and most objective realistic starting point of organizational evolution.Among the digital-intelligence pioneer enterprises Tencent has long tracked, the practices of the organizations that have truly succeeded are highly consistent:don't touch the organizational structure, don't change business processes, don't swap out IT systems — start from the most repetitive link, and place the intelligent agent in the entry point employees are most familiar with every day.
This is precisely the sociological logic behind Tencent using WorkBuddy and BuddyOne to link its office ecosystem: technology should proactively adapt to people's habits and existing relationship networks, rather than forcing people to accommodate unfamiliar systems.
Through the new experience of "unified accounts, shared memory, unified management" built by BuddyOne, along with the data platform's connections to WeCom, third-party IMs, and systems such as ERP, CRM, and HR, change naturally grows within the daily fabric: in WeCom, WorkBuddy moves directly into the contacts list, becoming a "digital colleague" in the chat window to whom tasks can be assigned at any time; in Tencent Docs, humans and AI engage in real-time "human-machine co-writing" on the same document; in Tencent Meeting, pre-meeting scheduling and post-meeting recording, transcription, and intelligent minutes connect naturally; and product and R&D staff go to the front lines of industries such as education, sitting beside teachers, gradually helping them smooth out lesson-preparation workflows and connect system interfaces.
When discussions in WeCom group chats, decisions made in meetings, and knowledge accumulated in Tencent Docs, Tencent Leking, cloud drive, and ima achieve "shared memory" within authorized scope, enterprises no longer need to re-explain the whole context to the AI every time.
And when workflows truly grow into the organization,the "AI-native organization"'sthree-fold deep evolution follows naturally:
First, promoting human-machine collaboration, giving every position and industry a professional partner.From hosted digital assistants (WMA) for various positions, to industry-specific Agents in healthcare, marketing, risk control, maps, intelligent cockpits, and more, humans in the loop (HITL) inject context and aesthetic taste, while multiple agents in the loop (AITL) work in parallel collaboration to resist organizational entropy, helping managers and frontline staff make more scientific decisions.
Second, thickening organizational assets, consolidating personal experience into collective memory.Through the full-scenario service system of "assessment—training—competition—certification", enterprises can systematically cultivate internal AI talent; in pioneer enterprises such as Mengniu and Shanghai Jahwa that have produced dozens of digital employees through combined competition and training, enterprises further consolidate mature business logic into reusable skills, and with the underlying refinement, de-identification, and review of personal experience, consolidate the effective workflows an individual pioneering employee has figured out into assets the entire team can directly reuse —“Let what one person accomplishes become the starting point for everyone.”
Third, spark the emergence of innovation, forming a positive cycle that feeds back into workflows.As organizational assets accumulate and the barrier to invoking them keeps dropping, ordinary employees who understand the business best are no longer mere executors of processes, but creators of new tools and new workflows. This “emergence of innovation” within the organization continuously feeds back into business workflows, driving the entire flywheel to strengthen in a virtuous loop, ultimately forgingdeliverable, verifiable, sustainableproductivity.
Putting people first: returning to the “human” itself
Over long task chains, tiny errors accumulate with each step. This simple mathematical law reminds us:The better AI becomes at handling deterministic execution, the more humans must take responsibility for the direction and value amid uncertainty.
As agents step out of the dialog box, the AI-native productivity flywheel displayed by Tencent at the 2026 Global Digital Ecosystem Conference—with“putting people first”as its axis—uses an efficient, manageable, and controllable technical architecture to“put people at ease,”and uses open, connected workflows and human-machine collaboration in organizational evolution to“amplify people”—ultimately pointing to an extremely plain return:
not letting communication become idle spinning of machine with machine, but letting exchange return to sincerity between people; not letting technology become anxiety hanging overhead, but making it the foundation that lifts up the abilities and creativity of ordinary people.
What belongs to machines belongs to machines; what belongs to humans belongs to humans. In this era where AI accelerates at full speed, may we possess the composure to invoke intelligence at any moment, while also retaining the resolve to slow down, think, savor, and trust one another. Use technology that puts people at ease to embrace a future that amplifies people.
This article was provided by Tencent and republished by QbitAI with authorization; the views belong to the original author.