# Tencent brings WorkBuddy into the software that runs companies

> Tencent connects WorkBuddy to ERP and CRM, launches Buddy One and adds security and spend controls so the agent can move from chat to day-to-day management.
> By José Maldonado · Artificial Intelligence · 2026-10-31

Tencent has decided that its WorkBuddy agent should stop living in a browser tab. On October 30 the company announced the agent's connection to enterprise resource planning systems, customer management and other business tools, plus an office suite called Buddy One that brings together Tencent Docs, Tencent Meeting, its knowledge app and its design tools under a single account. The stated goal is to move from personal productivity to running the business.

The move has one logic that is easy to follow and another that is read between the lines. The first is commercial: an agent that only drafts emails sells as an add-on and competes with any chatbot. One that reads orders, updates customer records and closes tasks in the management system sells as infrastructure, and that changes the budget it is aiming for. The second is about control: without a single identity and a spend dashboard, no large company lets an agent touch its systems.

That second point is the one worth watching closely. Buddy One shares a pool of artificial intelligence credits across applications, and the enterprise offering adds an observability center that shows how much is spent and what is obtained in return. In practice, Tencent is replicating the model that Microsoft and Anthropic have placed at the center of their strategy: the agent matters little if there is no way to govern it.

## WorkBuddy is much more than an office plugin

The difference from an office plugin is one of degree, not concept: it changes who holds the account, who pays for consumption and who can see the trail of what the agent did inside each tool.

WorkBuddy was born as an office agent with two halves: one focused on programming and one on administrative tasks, both hosted on Tencent's cloud. The enterprise version [integrates natively with WeCom, Tencent Meeting and Tencent Docs](https://www.tencentcloud.com/), so the agent can read and write in everyday tools instead of requesting permission via API from each one.

Buddy One changes the unit of sale. You no longer buy an agent, but a shared desktop with a single identity, cross-application search and memory that travels from one tool to another. The company describes it as a way for knowledge and context to flow in a single query, and in practice that means the employee asks once and the system searches documents, meetings and knowledge bases at the same time.

That unification is attractive and it is also a concession. The employee's entire work context ends up in a single space, with one account and one invoice, and the company gains visibility into what is done. For the worker, the convenience of not repeating context comes with the fact that their activity is no longer spread across five tools and is now concentrated in one.

## The agent that reads and writes in your systems

The connection to ERP and CRM is the change with the most consequences. An agent that only reads can limit itself to answering; one that writes can create an order, modify a record or close an incident, and that is where permissions, traceability and the reversibility of each action come into play. Tencent mentions enterprise security controls and a unified identity, without yet detailing how each write is audited.

Comparison with the rest of the sector helps to calibrate. Microsoft opened its own agent harness in Foundry with a managed model, and [Anthropic handed its skills standard to a neutral foundation](https://observatorioblockchain.com/en/ai/anthropic-registro-salida-bolsa/) so that it would not depend on a single vendor. Tencent arrives at the same point by another route: it does not publish a standard, but packages its own with the office suite that millions of Chinese companies already use.

For a company that already pays for Tencent Meeting and Tencent Docs, the proposition is hard to reject on price and easy to accept out of inertia. For one that is not in that ecosystem, the calculation is different, because native integration only works well in-house and open interfaces always arrive with less depth than internal ones.

None of the three routes is neutral. Each vendor packages its agent with the suite it already sells, and the promise of changing harness without changing vendor almost always stays in the sales brochure.

## Shared credits and a dashboard for the boss

The detail that says the most about where the market is heading is the one about credits. Buddy One uses a shared pool across applications, with additional packages that are added to the company's balance and expire after six months. The company does not buy a number of licenses and forget about it, but manages a consumption pot that runs down with use.

That design turns artificial intelligence spending into something visible and budgetable. It also introduces a new tension: when the balance is shared, the team that consumes the most conditions the rest, and decisions about who can use which model move from IT to management. The observability center that Tencent adds answers precisely that question, with spend and outcome data per team.

The logic is not exclusive to China. For a year now companies have been discovering that their agents consume more than expected and that nobody knows what is being paid for, a phenomenon the sector calls Shadow AI and which is beginning to have tools of its own. Charging by shared credits and offering a dashboard is the most direct way to turn that problem into a revenue line.

## The same move as Microsoft and Cognition

That three different companies have reached the same conclusion in the same week is no coincidence. An agent without a single identity, without spending limits and without a record of actions cannot be sold to a bank, and the market has stopped rewarding demonstrations. What sells now is the layer that makes it possible to say how much it costs, who used it and what it did.

That is why this week's news stories look so much alike. Microsoft opened its harness in Foundry, Cognition set per-engineer quotas in Devin and Tencent packaged its agent with the office suite and a spend dashboard. None of the three announced a better model; all three announced the way to control the one they already have.

The consequence for the buyer is that the selection criterion has shifted. The focus is no longer which agent answers best, but which fits the identity you already have, how much it lets you limit spending and what it lets you see when something goes wrong. It is a less flashy criterion and a far more useful one for deciding.

The result is a market where the differentiator stops being the model and becomes the accounting of its use, with the ability to say no to a specific employee as a sales argument.

## Chat was the demo, ERP is the business

Tencent has not presented a smarter agent, it has presented one that can sign off inside the system. That difference is what separates an office product from a piece of infrastructure, and it is also what multiplies the risk: when the agent writes to the ERP, an error does not stay in a badly written paragraph, it becomes a duplicate order or a corrupted customer record.

What comes in the next few months will be decided in unglamorous details. How long it takes to reverse an action, what is recorded in the log, whether the employee can undo what the agent did and who is accountable when something fails. Buddy One has the opportunity to answer those questions well in a huge and undemanding market. If it does, the rest of the vendors will have to copy it.
