Recently, Zhang Shaofeng, founder and CEO of Bairong Intelligence (6608. HK), visited multiple AI native enterprises overseas. He noticed a significant change: after new tasks appear, the team will first hand them over to AI to complete, with humans responsible for setting goals, checking results, and taking over when AI encounters difficulties. This means that AI has been able to play a role similar to that of a "colleague" in organizations.

Why can AI take over complete tasks?

Traditional software mainly provides functions, and the user is still manual. However, intelligent agents have the ability to understand intentions, decompose tasks, call tools, and check results, so they can continuously advance a task. Zhang Shaofeng believes that this makes AI a prerequisite for participating in enterprise operations as a labor force.

However, he emphasized that big models are not equivalent to intelligent agents. Zhang Shaofeng compares large models to the "brain" of intelligent agents. Only deploying large models, AI usually can only generate content or provide suggestions. Completing a multi link enterprise task requires planning, memory, tool calling, and execution capabilities, as well as access to the enterprise knowledge base, business systems, and permission system. He referred to the work beyond these models as the '99 kilometers' that enterprise AI applications must complete.

Many companies have already integrated large models and launched intelligent assistants, but their business returns are not significant. Model availability does not equal business availability, let alone the ability to replicate on a large scale. When facing complex tasks, enterprise level AI needs to have high stability, security, and compliance. Any problem in any link may affect delivery.

In terms of scene selection, Zhang Shaofeng advocates prioritizing tasks with high frequency, measurable results, and clear responsibility boundaries. For example, text and voice interaction, unstructured data processing, and long-distance work with clear rules. Repetitive execution and information organization can be handed over to AI for manual verification and receipt of results.

How can an employee manage hundreds of AI?

Bairong Intelligence refers to intelligent agents with clear positions and the ability to independently complete tasks as "silicon-based employees". The company has currently deployed over 200000 silicon-based employees, with service scenarios including customer service, marketing, operations, finance, legal, and recruitment. The internally calculated "silicon to carbon ratio" exceeds 150:1. Bairong has built a "Silicon based Employee Home" for this purpose, which integrates job positions, identities, training, assessment, scheduling, auditing, and exit into unified management.

He emphasized that the silicon to carbon ratio cannot be simply understood as the scale of personnel replacement, as it reflects the management radius of the organization. He shared an internal real-life case where the company had a department dedicated to serving small clients with an annual contribution income of less than 500000 yuan. They provided remote services to over 2000 companies through phone, video, and other means, originally requiring 50 employees. After adopting AI to improve efficiency, the number of employees was reduced to 5 and 18 silicon-based employees. But the other 45 employees were not laid off, instead they learned how to build intelligent agents to serve more enterprises. They have transformed from a back-end functional department to a front-end business department, and their income has actually increased. This case illustrates that AI is not replacing humans, but rather 'repetitive labor' itself, and humans will be transferred to more creative positions. ”

Why are AI companies starting to charge based on results?

After AI has the ability to deliver complete tasks, the enterprise level AI application model changes accordingly. Bairong Intelligence has proposed the RaaS (Results as a Service) strategy, which allows for billing based on work orders, effective calls, or actual workload that meet quality standards and are successfully completed in customer service, operations, and other scenarios; In the scenario of facilitating transactions, profits can also be distributed based on business results. The customer's investment is directly linked to the actual work completed by AI.

As for the scenario with the fastest implementation of enterprise level AI, Zhang Shaofeng believes that programming and customer service will be at the forefront, because the value measurement standards for these two scenarios are relatively unified: whether the program runs smoothly, and whether customer problems have been solved? Among them, in the field of customer service, the AICC (AI Intelligent Contact Center) created by Bairong Intelligence is the breakthrough point for enterprise level intelligent agents to achieve cross industry scale implementation.

For example, in the incoming call scenario of a leading logistics company, AI receives about 15000 calls per day, of which 80% can independently close the loop, and customer satisfaction remains above 95%. During a 23 minute phone call that was disassembled on site, the intelligent agent was able to actively slow down the speech rate, continuously memorize the context, extract key information, and call the business system within the permission boundary to provide executable solutions to elderly users' dialect accents, scattered information, repetitive expressions, and operational difficulties.

Looking ahead, Zhang Shaofeng believes that measuring an enterprise level AI application ultimately depends on three things: whether the task can be completed, whether the cost has improved, and whether customers are willing to continue expanding their use. These determine whether AI can become a long-term productivity tool for enterprises.