Why Most Companies Aren’t Agentic Yet: Tencent’s 4 Missing Pieces for Trusted Agents (4 of 5) (Tech Strategy)

In Part 3, I went through the architecture for an agentic enterprises. Which brings me to the big question:

Why aren’t businesses going agentic-first?

Because going agentic rightfully makes businesses nervous.

Are the agents good enough? What is the performance? What do they cost?
Are the agents reliable?
What about security? What about protecting data?

Huixing argued that enterprises need 4 essentials for sustained deployment of agents:

Production ready capabilities. It needs to be easy.
Reliable execution
Dependability
Measurable value

So, you need to offer more than WorkBuddy and the graphics above. You need at least four other capabilities for manageable and trusted agentic operations.

Tencent’s Answer: Easy Deployment, an Open Platform, Security and Governance

According to Tencent, there are four required capabilities that line up against these four essentials:

Easy to deploy, train and manage digital employees. For Tencent, this is the WorkBuddy managed agents (WMA) platform.
An open platform. You need to be able to work with everyone. And you can’t be locked in to one model, messenger, or ecosystem.
Enterprise -grade security. This includes real-time monitoring, traceability and the ability to do investigations. And it needs to be everywhere (cloud, operations, edge, etc.)
Governance and control. Management needs to have total understanding and control of everything the agents are doing.
- Tencent Cloud offers an Observability Center. This gives management end-to-end visibility in agent operations. This includes:
  - Real-time monitoring
  - Session tracing
  - Execution details
- Controllable behavior. No moves that aren’t approved. No accidental deletions.
- Cost management / token consumption.

Put Together, Here Is the Full Agentic Enterprise Architecture

Put that together and this is what you really need for an agentic enterprise.

You can see:

Seamless integration across the Tencent ecosystem
- WeChat / WeCom
- Tencent Meeting
- Tencent Docs
- Ima
- Lots of other stuff being added
Open integration with third party ecosystems
- IM, 3^rd party collaboration office software
- ERP, CRM, OA
- MCP
- Marketplace
Open and diverse model ecosystem
- Co-design (with software, hardware, and apps. Not standalone models)
- Day 0. Simultaneous day one readiness and deployment. New model works immediately with hardware and infrastructure.
- Enterprise self-developed large models.
- TokenHub (open source, closed source)

This last point is important.

For Tencent, WorkBuddy is the desktop app a person uses. That’s for the individual.
But WorkBuddy Managed Agents (WMA) is the cloud platform underneath where a company builds its own agents and runs them.

WMA Moves Agent Work to the Cloud as a Managed, Pay-as-You-Go Service

Tencent describes WMA as a cloud-hosted agent runtime service that puts every enterprise agent on a production-grade foundation.

It basically moves agent task execution to the cloud, so companies can create digital assistants based on their own business processes

WMA was officially announced in October for:

Running digital employees
Building custom vertical agents
Embedding agent capabilities into a company’s own applications

It is sold as managed, maintenance-free hosting that is production-ready, with elastic scaling and pay-as-you-go pricing.

ADP Lets Companies Build Their Own Agents. That Protects Their Alpha Over Time.

At the bottom of that last graphic, I put Agent Development Platform (ADP). This is the full Tencent capability where you create your own agents. Which is critical for protecting your company alpha. Much of a company’s expertise and unique capabilities will be in their own proprietary and customized agents.

Agent Runtime Puts Each Agent in a Sandbox. That Solves Much of the Trust Problem.

There is also Agent Runtime at the bottom. This is Tencent’s full hosting and execution layer that manages deployment, operations, and lifecycle management of the Agent. It uses the Agent Sandbox, which is the infrastructure layer providing an isolated environment (the “sandbox”).

This is actually the solution to a lot of the trust issues. You keep the agents in a secure environment that limits what they can do. They are limited by both policy and technical architecture.

Here’s the Full Tencent Agent Stack, From the Desktop App Down to GPUs

So it’s:

WorkBuddy. The desktop app a person works in. This ties into a greater ecosystem. Both Tencent’s and external to Tencent.
WorkBuddy Managed Agents (WMA). This is the service businesses can use to manage their agents. It mostly sits on top of and uses ADP and Agent Runtime.
ADP. This is where companies build, deploy, govern and improve their own agents.
Agent Runtime. This gives each agent task an isolated sandbox to run its code, use tools and touch data. And then tears it down.
Tencent Cloud compute and GPUs. The other hardware underneath.

Agentic Organizations Need New KPIs

Huixing had some good comments about how you measure an agentic organization. Here is my interpretation.

The right side is interesting. Going forward we want to start seeing the system self-learn. And we want to see the agents increasingly take their own initiatives. We’ll see.

Here’s the final slide from Huixing.

***

Ok. That’s it.

I think we’re almost to a real clear design for agentic organizations. I’ll present another version of this thinking from McKinsey & Co later this week.

Cheers, Jeff