Everyone Wants AI Agents. Nobody Wants to Be Responsible When They Go Wrong
Play Audio Version
AI agents are arriving. It’s no longer science fiction to have a team of AI-powered workers all performing tasks on your behalf.
The New York Times recently published a list of ideas for how people can use this tech in their personal life. Finding flight deals, booking hard-to-get restaurant tables, even applying for jobs. All these tasks can be handed to your personal agent.
The advice for most personal users has focused on Muse from Meta. This is because it’s freely available to consumers. Dots from OpenAI costs $100 a month to use and Anthropic charges are more closely related to what you are doing with the agents.
The ongoing rollout of agents is controversial for some businesses though. Amazon has banned Muse from visiting their site and shopping on behalf of customers, but companies like Walmart are welcoming the AI agents with open arms.
It is the personal use of agents that is getting most coverage in the business media at present, but there is a quiet change also taking place in the enterprise. Many people in information-focused jobs are exploring how this changes their own workflow. Can knowledge workers all be responsible for a team of agents that can make everyone more productive?
It’s been common in software development for more than a year now.
This is an interesting development as we look beyond coding, because most of the discussion about the enterprise use of IT has focused on generative AI – how can employees be more productive through better and easier access to information?
Now there is the possibility that employees can have better information systems, but also a team of agents working on their behalf. Each morning you should be able to set off your team of AI agents on specific tasks, that can be reviewed as each agent delivers what they needed to do.
AI agents can now perform many functions inside businesses. They can initiate actions that used to require a person, interact with systems, and even make operation decisions – where they have been given permission.
This all creates new challenges for enterprises. We are no longer just talking about co-pilot summarizing a long document or searching through FAQ details. Now we need to consider how to create an information framework that allows agents to change financial records, access customer data, and even make payments on behalf of employees.
This challenge is here today. The consumer adoption of AI agents is just creating more visibility around how AI agents can perform tasks on behalf of people. Although, it also creates an imperative for CIOs to answer questions from their own employees about why they have automation and AI agents available on their personal phone, but they are more restricted in how they use AI in the enterprise.
The real issue is not the technology – it is how to create a safe and stable environment within the enterprise.
AI orchestration, defining the right level of permissions, monitoring agents, and the governance you apply to the use of agents all need to be in place.
Imagine an energy utility accounting team using agents to find and contact people who are struggling to pay their bills. The automated contact is relatively simple and could easily just be triggered by who has paid, who has not paid, and what their payment history looks like. However, with an agent involved in this process, there is the option for more decision-making.
Should an agent mark a customer account as ready for disconnection because of unpaid bills or should a human always supervise and check decisions like this?
Getting this right inside the enterprise is not just about how good the AI technology is or how intelligent and trained your agents are. It is essential that all these questions around governance are asked before you even allow agentic automation to take place.
Shadow AI use is becoming a major problem for many CIOs and enterprise leaders. People find AI so useful in their personal life that they are using and applying AI tools to their work systems and information – without permission.
This can create problems where confidential data is being shared with external systems, but more importantly with agentic systems, this can create a situation where agents are making decisions without any governance around this decision-making authority.
IBA Group has an enterprise AI agent platform called STAIR. This allows enterprises to choose their own preferred LLM and cloud configuration, but then to quickly apply permissions, policies, and central security to how agentic AI is used across the entire enterprise.
The real agentic opportunity for enterprise IT leaders is not simply to deploy more AI agents, but to quickly make it possible for employees to use them confidently, safely, and at scale across the entire business.
An established and reliable platform like STAIR can accelerate adoption by making governance part of the infrastructure, rather than a problem that every department must solve independently. You can quickly move beyond pilots and experiments to enterprise-wide agentic AI use.
The companies that succeed with agentic AI will not necessarily be the early adopters. You don’t need to be first. You just need quickly create an environment where thousands of agentic agents can all operate safely.