Treat AI Agents More Like Human Workers

Tuned-in business leaders seeking advantage from applying AI in their organizations no longer ask if the technology can assist or run tasks, but instead ask which tasks it should be deployed on. Agentic AI agents can reason across time horizons, learn from outcomes, and collaborate with other AI agents to optimize performance. They can provide emotionally intelligent responses to enquiries, and escalate instances where human judgment is needed.

For any business, the potential results are difficult to ignore: better decisions, faster cycles, and dramatically lower unit costs. Agentic AI is not just new technology, but a whole new operating model. It’s also true, though, that these systems can reinforce bias, obscure accountability, or trigger compliance failures if they are not properly managed with deliberate and close attention.

Benefiting from agentic AI means a raft of organizational changes, from new roles and hires to new incentives, KPIs, and training. It also means treating AI agents very much like human workers, with defined roles, accountability, and performance metrics embedded into the operating model. Attention COOs and CIOs. Yours are the frontline leadership roles and responsibilities for preparing the organization to deploy and employ AI agents.

Our agent colleagues

Rather than executing fixed instructions, agents act more like collaborators. They can make autonomous judgments and decisions, and take action to achieve defined objectives. Single-agent systems can perform end-to-end tasks independently. Multiagent systems operate as decentralized networks of agents that interact and collaborate.

For example, for a finance business, three separate agents may work together—one assessing creditworthiness, another modeling risk exposure, and a third ensuring regulatory compliance—to optimize the customer journey in real time. Clearly, embracing AI of this type and at this level means reshaping the way organizations operate. This unlocks gains in speed, scale, and precision, but also brings new categories of risk, along with the need to manage these.

There’s certainly a price of entry to this world in terms of investment and time, requiring both to be spent on infrastructure, interoperable data ecosystems, and deep integration across functions. To work at scale—and it only makes sense at scale— leaders need to overhaul accountability, ethics, and governance, so that humans and intelligent machines can work effectively together.

Managing a hybrid workforce

Just like effective human workers, AI agents need to be well managed if they are to be productive and safe, and that means corporate consideration of how they are funded, evaluated, and integrated. For example, leaders understand the full cost of human talent, which likely comprises salary, benefits, bonuses, and training. They now need to focus on the total cost of ownership (TCO) for AI agents, too, including IT systems, model retraining, orchestration layers, governance tools, and compliance. The best agents, like the best human workers, should be able to work across functions, while underperforming agents should be retrained, more closely performance-managed, or retired.

Every agent needs a job description and to have its results monitored, just as with any human team member. Agents must also adhere to guardrails just as humans adhere to policies, especially in regulated sectors. Elevating AI agents from tactical tools to strategic workforce assets means holding them to the same (or similar) standards used for people.

We work in ‘smart ops’

The ways in which organizations make decisions will change, with humans and AI agents working together to make decisions. Who leads whom will depend on the task. And as agents take on high-frequency or transactional work, employees shift into roles that require more oversight, ethics, and judgment.

The shift reaches far beyond an implementation plan, encompassing a whole new mindset. Each digital worker—like each human worker—needs a clearly defined role and objective, along with a measurable impact on business performance, governance, oversight, and opportunities to work elsewhere if its performance is deemed up to the job.

Deciding which decisions to automate

Although agentic AI offers potential across nearly any function, service operations remain the sharpest proving ground. These environments are rich with high-volume, repetitive tasks and data trapped in silos, making them ideal for intelligent automation. But the question is no longer what companies can automate. It’s which decisions they should automate—and where human judgment still matters.

Organizations chasing automation without considering if they are focusing on the right tasks are likely to expend time and resources to little effect. Instead, the decisions about the application of agentic AI entities should be based on careful consideration of the risks involved and the degree of judgment required. Low-risk, low-complexity decisions are prime for full automation. High-risk, high-judgment scenarios will still require human oversight, perhaps supported by AI copilots. Some areas of business will have no direct AI involvement, which is as it should be.

Getting started

Deploying agentic AI at scale means organizational changes that are unlikely to be straightforward and will certainly take time. Pointers to support your discussions and decisions about how to get ready to welcome a new generation of agentic coworkers include focusing on cross-functional alignment. Senior leaders—starting with the COO and the chief information officer—should own outcomes.

Elsewhere, human roles must shift toward exception handling, judgment-based decision making, and customer experience. Companies need coordinated, thought-through training and hiring programs to meet these demands. Further, AI’s effectiveness depend on reliable data, so parallel projects that aim to modernize data infrastructure will be an essential part of the journey for many.

And finally, in transformations of this scale, communication at all levels is critical to reducing resistance and sustaining efforts. In fact, not taking people with you is the most common reason for transformation failures. That’s unlikely to be any different for the shift towards Agentic AI readiness and adoption.