You’ve probably felt it: the quiet dread when an AI moves from answering questions to actually doing things. Deleting files. Sending emails. Spending money. That’s the moment the stakes change. And that’s exactly the moment Tencent’s WorkBuddy decided to accelerate.
In just three months, this AI agent hit 12 million daily active users, a 65-75% DAU/MAU ratio that beats Slack at its peak, and a rumored $300 million annual revenue run rate. But here’s what nobody’s talking about: WorkBuddy’s secret isn’t its AI model. It’s something far more boring—and far more dangerous for competitors.
When you strip away the hype, the real moat isn’t the brain. It’s the harness.
WorkBuddy is built on a system called Harness—a set of engineering layers that manage context, tools, execution loops, and recovery. The model is interchangeable. DeepSeek, Kimi, GLM, MiniMax, Hunyuan—WorkBuddy routes tasks to whichever fits best. The true competitive advantage isn’t picking the smartest model; it’s building the infrastructure that makes any model reliable enough to trust with real work.
Think of it this way: a powerful but unpredictable horse is useless without a harness. The harness doesn’t replace the horse—it makes the horse usable. Apply that to AI. The model is the horse. The harness is everything else. Most companies are fighting over horses. WorkBuddy built the harness.
But the deeper insight—and the one that will reshape enterprise software—is that even the best harness solves only half the problem.
Individual productivity is soaring. Teams are still stuck in the mud.
Tencent’s own product lead, Liu Yi, put it bluntly: ‘Individuals feel great. Organizations feel nothing.’ A worker who used to need two hours for a PowerPoint now finishes in twenty minutes. But the project still waits for approvals, cross-department dependencies, and review cycles. The bottleneck shifts from content creation to organizational friction.
This is the twist that most AI companies ignore. They sell you a faster horse. They don’t sell you a new road system. WorkBuddy’s enterprise play is exactly that: not just a faster tool for individuals, but a way to rewire how tasks flow through an organization. It’s moving from a personal agent to an organizational interface.
That’s why pricing models are shifting too. Traditional SaaS charges per seat. WorkBuddy is experimenting with task-success-based pricing—charging for outcomes, not access. If you think that’s a small change, ask yourself: what happens when every software vendor adopts a pay-per-result model? The entire industry flips from selling tools to selling guarantees.
Neutrality is death. I’ll say it plainly: the future of enterprise software is not a collection of tools. It’s an agent orchestration layer that sits between you and every system you use.
WorkBuddy’s origins reinforce this. It started as CodeBuddy, an AI coding assistant for 12,000 internal engineers. Non-technical employees started using it to organize research, write reports, manage files. The team noticed and built WorkBuddy in 48 hours over a weekend. That speed wasn’t magic—it came from years of building the underlying Harness system. The 48-hour prototype was just the tip of a very deep iceberg.
If you’re a business leader, here’s the practical takeaway: stop obsessing over which model is smarter. Start asking how your organization will actually execute tasks that involve multiple systems, permissions, and handoffs. The companies that win the AI era won’t be the ones with the best chatbot. They’ll be the ones that build the operating system for work itself.
WorkBuddy is still far from perfect. Third-party tests show it scores only 2.31 out of 4 on complex tasks. It’s great at structured text, weak at design and cross-system operations. But the direction is clear: the battle for the enterprise is no longer about AI that answers questions. It’s about AI that takes actions—and takes responsibility for the results.
That’s a shift that changes everything. And it’s happening faster than most people realize.
FAQ
Q: If the model doesn't matter, why does everyone obsess over which AI is best?
A: Because models are the flashy, easy-to-compare part. The hard work—building reliable execution loops, context management, and error recovery—is invisible but far more valuable. In the long run, the harness wins over the horse.
Q: What's the practical implication for my company?
A: Stop evaluating AI tools solely on demo quality. Ask about failure recovery, task completion rates, and integration with your existing systems. The real value comes from how well the AI handles complexity, not how smart it sounds in a single conversation.
Q: Isn't this just a rebranded version of old enterprise software?
A: No. Old software sold tools for humans to operate. This sells an operating system where agents do the work and humans set the goals and handle exceptions. The pricing model may shift from per-seat to per-task, which fundamentally changes the vendor-buyer relationship.