Folks, I've hit the jackpot!

Today, at the Bund Conference, I've stepped into a brand-new role: Factory Director.

However, the workforce in this factory I'm overseeing consists entirely of AI agents.

My mission for the day is straightforward: construct an AI agent, delegate tasks to it, observe its collaboration with fellow agents, and ultimately present the outcomes to me.

Let's dive in, starting with "recruitment".

There's already a pool of over 300 job-level digital experts and more than 600 skills at our disposal.

For these roles, you can simply pick templates and then piece together the necessary skills like a puzzle.

Once the role is chosen and skills are incorporated, link it to the company's proprietary data and business systems, and voilà, a digital employee comes to life.

This process is somewhat akin to crafting a character in a video game, except the tasks it undertakes can have a real impact on a company's operations.

You might employ versatile AI agents like QianWen or WorkBuddy to gather information, create spreadsheets, draft reports, or execute tasks.

Yet, a company's sales, customer service, procurement, finance, and risk control departments each have their unique data, permissions, and workflows. Thus, corresponding role-based AI agents must also carry out their respective responsibilities.

Hence, the term "factory".

Just as the steam engine empowered machines, and the assembly line later figured out how to organize machine groups for production, large models have endowed AI with formidable capabilities, and the Agent Factory tackles the challenge of organizing these capabilities for practical work.

Overseas tech giants such as Google, Microsoft, and Amazon are all constructing agent factories because as agents start to infiltrate businesses en masse, more efficient production, scheduling, and governance become inevitable evolutionary paths.

But when there's a multitude of agents, genuine challenges emerge. What if tasks are duplicated? Or if someone gains access to data they shouldn't?

What if an agent gets stuck midway, and who takes over the subsequent task?

Hence, the crux lies in "management".

For instance, when serving a customer, some agents pinpoint needs, others devise solutions, some assess risks, and yet others manage follow-up service.

Different roles collaborate in a relay fashion to accomplish a task step by step.

On the surface, it appears as if a few agents are merely exchanging messages.

But beneath the veneer, the sophisticated Harness Framework developed by Ant Digital Tech is silently overseeing the entire operation.

Who initiates the process, who follows, which tools are utilized, where bottlenecks occur, and who receives the results next—all are transparently visible here.

This is what I witnessed today that might hold the utmost value.

Constructing a single agent is already swift, but the true litmus test for an Agent Factory is its ability to effectively manage a cohort of intelligent AIs.

It's truly a relief when I don't have to overexert myself.

But having said that, can we truly relinquish control and place our trust in these digital employees?

A miswritten paragraph can be rectified, but in the future, when they assist companies in placing orders, making purchases, granting authorizations, or even conducting transactions directly, every step must be crystal clear.

Who authorized it? What data did it access? What actions did it take? Can issues be traced back with ease?

Ant Digital Tech has incorporated a layer of blockchain technology here to maintain trustworthy records of some key operations, rendering them traceable and facilitating audits.

In the future, when agents from two companies autonomously negotiate business, this capability will prove even more invaluable.

Alright.

The employees are created, tasks are assigned, progress is monitored, and results are delivered.

Today's KPI as Factory Director is deemed accomplished.

After this experience, I've gained a deeper insight into why businesses require an Agent Factory.

When we, as individuals, utilize AI, we're concerned with how much it can accomplish for us personally.

But businesses are more interested in whether dozens or hundreds of agents can collaborate seamlessly, each fulfilling their designated roles, with processes that are controllable and issues that are traceable.

The moniker "Agent Factory" might initially seem somewhat perplexing.

But after traversing the production, management, and delivery processes firsthand, it does seem rather apt.

As for when this factory will truly penetrate the core operations of more businesses, we shall continue to observe.

Time to clock out!