AI Agents Are Coming

ai agents concept with businessman is using artificial intelligence agentic technology on laptop computer to help make decision marking on business and planning schedule for organization improvement
AdobeStock
What leaders need to know—and do—about agentic AI.

Agentic AI may redefine how businesses operate, compete and grow—faster than many leaders expect. It offers the clearest glimpse yet of what the future of work and business will look like. This is expected and exciting—new ecosystems are already emerging to support the technology and redesign tech stacks and workflows. The opportunity is enormous, and so is the disruption. So, what should leaders know and do now?

What are agents? As with AI, there is no consensus definition, and trying to force one doesn’t really help. In general, “agents” are AI models that are designed to complete evermore complex multistep tasks without human intervention. Most are models built on top of existing LLMs (with all their strengths and weaknesses) and call on multiple datasets and/or other websites to string tasks together. For instance, “make a reservation at Joe’s for 7 p.m. on Tuesday, and schedule a car to get me there on time in light of traffic conditions.”

Are we there yet? At the time of this writing, not really. Most commercially available agent-style AI still requires substantial human input, alignment, guidance and oversight. Other models already cluster agents into “multi-agent” systems that in effect do a good (and improving) job of simulating autonomous function by stringing agents together to solve a task. Today, these systems are more “agent-ish” than agentic. In the future, the landscape will be even more complex, involving the regular interaction of fully autonomous agents with agent-ish models, human-guided agents, narrow AI and existing human processes. This hybrid environment will be the governance challenge for tomorrow’s leaders.

Who has agents and is using them? Increasingly, anyone can access pre-built agents or readily build their own; in concept, there are no technical limits on the number of agents any one person or organization can have. Existing organizations must integrate agents into current workflows and processes (often with significant redesign), while new companies will build around agentic cores from day one.

Agentic vs. agent-ish, does it matter? Yes. Agent-ish models will be better suited for certain kinds of tasks than agentic ones, so both types will be in operation for the foreseeable future, and they require very different strategic and governance approaches.

What types of work can these agents do? The list keeps growing, but they are most suited for repetitive tasks for which a business already has a clear or proven sense of what good looks like (e.g. assembling regular financial reports, creating—and maybe executing—marketing campaigns, summarizing materials and distributing key findings).

OPERATIONAL ADVICE (FOR NOW)

1. Dive in. These tools are here and available and will only get better. Existing operations will need to learn to succeed in an environment with more automated competition and competitors.

2. Know your data. These tools run on data, with all the attendant dependencies on data quality, accessibility and governance.

3. Refresh policies, guardrails and training. Determine how and when agentic tools are authorized for use in the business and with what authorities and approvals. Define, encode and train on core values, standards, principles and ethics—for your people and your agents. This training is essential for their success.

4. Develop a method for agentic use case selection. Not all use cases (e.g., medical diagnoses) or specific problems (e.g., high value vs. routine disputes) will be suitable for truly agentic models. Therefore, organizations need a process for determining which sorts of tools and levels of autonomy are suitable for which problems.

5. Build an agent-tracking and management system. Agents are easily proliferated. Know which ones are in operation and what they are doing. Agents will interact with each other. And agent behaviors can’t be fixed after-the-fact.

6. Establish an agent governance protocol. Agents need to be aligned to human-set objectives (for now). Those objectives need to be reviewed, aligned and deconflicted, with quality standards and boundary conditions that are identified, set and enforced. The workforce needs to be trained in this process as well.

7. Start redefining workflows. Adding agents to existing processes likely won’t unlock the most optimal benefits or competitive advantages. But reimagining and redesigning workflows to incorporate capable agents and well-skilled managers will start to unlock these advantages, and that redesign work can start now.


  • Get the Corporate Board Member Newsletter

    Sign up today to get weekly access to exclusive analysis, insights and expert commentary from leading board practitioners.
  • UPCOMING EVENTS

    JULY

    29

    AI Leadership Forum

    New York, NY

    AUGUST

    7

    Board Committee Peer Exchange

    Chicago, IL

    SEPTEMBER

    18-19

    21st Annual Boardroom Summit

    Washington, D.C.

    MORE INSIGHTS