Opinion
AI agents need jobs, not just tasks
As agents move from isolated actions to continuing work across business systems, trust will depend on defined responsibility, clear boundaries and human oversight.
A company may ask an AI agent to review historical prices, compare contracts with market data, prepare a request for quotation and draft messages to suppliers. The agent can perform each action. The harder questions begin when the work has to move forward: Which suppliers are eligible? Which considerations matter besides price? When should negotiation stop? Who approves the final allocation, and how does that decision enter the purchasing system?
AI agents are moving beyond answers and recommendations into actions that cross business systems and affect real outcomes. Yet many companies still introduce them one use case at a time: one agent summarizes a contract, another drafts an email and a third creates a purchase request.
Having built autonomous systems in physical retail and now in enterprise operations, I have learned that capability is only the starting point. When agents receive prompts and permissions but no defined role, people still have to connect the outputs, supply missing context, decide what happens next and take responsibility for the result. This helps explain why impressive demonstrations can struggle to become dependable parts of everyday operations.
The distinction between a task and a job matters more than it may appear. A task is an action: compare two documents, contact a supplier or update a record. A job carries continuing responsibility for an outcome. It includes the information required to make decisions, the systems needed to act, the policies that govern those actions, the people or processes that receive the output and the point at which a human manager must intervene.
A job description defines what the agent is expected to achieve, which information it may use, which systems it may access and which decisions it can make. It also establishes which actions require approval, which conditions trigger escalation, how work moves to another person or agent and how the result will be measured. The organization can then see who remains responsible, what good performance looks like and when a person must step in.
Organizations already build trust in human employees through a similar structure: a defined position, appropriate access, company policies, measurable objectives, management oversight and escalation paths for unusual situations. AI agents need the same clarity, while responsibility remains with the people and organizations that deploy them. The analogy has limits because an agent lacks human judgment, motivation and accountability, but it helps companies define the work they are prepared to assign.
This shift also changes the role of human managers. Traditional software waits for people to initiate and complete each step. When an agent carries defined work across several stages, the manager’s job moves from advancing every action to setting priorities, defining boundaries, approving sensitive decisions and handling the exceptions that require context or judgment. The agent can absorb more of the tactical load, while the person remains responsible for the decisions that carry real weight.
The distinction becomes especially important when several agents work together. One may prepare a commercial decision, another execute it in an enterprise system and a third follow the process through delivery or financial closure. The value of the system depends on the intelligence of each agent, the quality of the handoffs, the preservation of context and the clarity of responsibility across the chain. A weak handoff can erase the benefit of a strong model: a policy may disappear when a negotiation becomes an order, or an exception may reach a person without enough context to resolve it.
Companies therefore need to treat agents as part of their operating model. Leaders must decide which digital roles the organization needs, who manages them, how their objectives relate to business outcomes and how authority is divided between people and software.
Performance measurement must start with the value the agent creates. In a sourcing process, that may include the reduction achieved from a supplier’s initial offer through negotiation. Measurement must then cover the full role: whether the agent completes work reliably, follows policy, preserves context, manages handoffs, escalates the right cases and can be improved without adding hidden operational complexity.
The companies that get the most from AI agents will be the ones disciplined about defining, managing and improving the work each agent owns. Adding tasks expands what an agent can do. Giving it a real job is what makes it something the organization can depend on.
Michael Gabay is Co-Founder and CEO of Gain.
