The CEO of Maxio Branden Jenkins stated that the most difficult thing to deal with after the company advanced AI agents was not necessarily a sudden rise in bills, but rather an anxiety among employees about the technology gap. One weekend using the AI code, he found that the account had been over $1,000 for automatic renewal in a short period of time, which had subsequently become an internal warning case.

One weekend to trigger the overrun

Jenkins said that he would normally call Claude directly on his cell phone for encoded and debugged tasks. Because the account has an automatic filling, the deduction will continue at $1,000 as long as the token is exhausted, and the cost will not be found to have risen until he has seen the dashboard.

In his view, the use of AI was often wasteful, not simply because of the number of calls, but because of inappropriate model selection and the continuous deviation of the dialogue process from its original objective. An error in the agent ' s execution would also result in continuous consumption of token, which would eventually result in unnecessary expenditure.

The report mentions that Gartner estimates that agent AI consumes token per task, five times as many as 30 times as normal chat robots. Other surveys indicate that budget overruns occurred in a number of AI projects in United States enterprises over the past year.

The company started to distribute the model by assignment.

After that experience, Jenkins began to adjust his way of using it. He assigns simple tasks to lighter models, general codes to intermediate models and high-cost models to more complex planning tasks.

He also mentioned that some third-party tools could reduce token consumption by compressing output content. For example, AI is required to respond with shorter, more direct sentences to reduce unnecessary costs. According to him, such methods can significantly reduce costs.

He also noted, however, that these techniques were usually in the hands of HF users or technicians and were difficult for ordinary staff to access. This difference in capabilities further exacerbates the use gap within the team.

More stress comes from staff anxiety.

Jenkins said that in the company ' s promotion of AI, he focused most on three points: efficiency, differences in the distribution of tools, and staff concerns. The last of these is the most difficult to address.

He mentioned that he would build automation tools directly in the operations for which the management team was responsible, which put pressure on some managers. He had been directly informed that he should have initiated those programmes, but now felt behind.

Similar sentiments appear in a broader team, including concerns about whether the post will be replaced and the team will be downsized. At the same time, the company ' s initial opening to the entire workforce of ChatGPT and its subsequent authorization of a higher-priced Claude to approximately 50 sales and market employees also raised questions about the unequal distribution of resources in other sectors.

Inclusion of AI agents in the organizational structure

In response to these problems, Maximo has begun to formally incorporate AI agents into the organizational design. Jenkins states that management, in combing the departmental structure, will include not only staff in statistical reporting relationships, but also AI agents, who are directly managed by each other, in a mixed organizational chart.

The company has also expanded the DevOps team to take over internal tools that have been developed by the staff themselves but have taken on a key role in its operations. Jenkins believes that these tools, if only in the hands of individual staff members, can raise issues of continuity, security and outreach.

In his view, it was even more interesting for an enterprise to use AI, not as an occasional four-digit overrun, but as the relationship between income growth and human expansion was changing, and whether employees were willing to accept the new tool.