Artificial intelligence agents are often pitched as virtual workers capable of running entire business processes with minimal supervision. But a new Forbes Innovation article by contributor Bernard Marr argues that for complex tasks, a single agent is rarely the right tool. Instead, Marr makes the case for multi-agent workflows, where specialized AIs each handle a distinct part of a process, and he lays out a practical design guide for putting them to work.
The article, published on Aug 24, 2026, at 01:08am EDT under the Enterprise Tech topic, contends that the more you ask one agent to do, the harder it becomes to understand what went wrong when it fails. That single-agent approach, Marr writes, makes failures difficult to diagnose and complicates any attempt to improve the system. The multi-agent alternative, by contrast, simplifies design, improves error detection, and allows for easier management and iteration.
Why One Agent Is Not Enough
Marr's central claim is that complex jobs, such as running marketing campaigns or managing end-to-end QA processes, quickly overwhelm a single AI agent. When one system is responsible for everything, a breakdown anywhere in the chain becomes a mystery. You cannot easily tell whether the problem lies in research, drafting, publishing, or something else entirely.
The multi-agent approach solves that by breaking the work into pieces. Each agent has one relatively simple and clearly defined task. That primary advantage, Marr argues, is what makes the whole system more manageable. If one component fails, you can see exactly where. If a task changes, you can adjust a single agent without rebuilding everything.
Individual agents can be swapped out or retrained without breaking other elements. That modularity also makes it easier to update or adapt the process as business requirements evolve. Instead of wrestling with a monolithic system, you are tuning a small team of AI specialists.
Real-World Examples in Marketing and Customer Service
Marr provides two concrete examples to illustrate how multi-agent workflows operate in practice. In marketing, one agent handles research, another drafts the content, and a third publishes it. Each step is separate, measurable, and replaceable.
In customer service, the division of labor is even more granular. One agent classifies incoming requests. Another consults the knowledge base for relevant information. A third drafts responses. A fourth decides whether the issue needs to be escalated to a human. That separation of duties means each agent can be evaluated on its own performance, and a failure at any step is immediately visible.
The multi-agent approach does introduce a new challenge, however. Marr acknowledges that getting agents to work together effectively requires careful design. The benefit is that you can diagnose and adjust one component at a time, rather than guessing at the source of a problem in a tangled single-agent system.
Six Steps for Designing Multi-Agent Workflows
Marr's article offers a step-by-step guide for organizations that want to build these systems. The first step is to pick a job where results can be easily measured and success objectively defined. Without clear metrics, you cannot tell whether the workflow is working.
The second step is to design the workflow, not the team. Marr advises mapping out every task first, then deciding how many agents you need and what each one will do. That order matters. If you start with the agents, you risk forcing the process to fit the tools.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Third, give each agent a narrow, clearly defined job with consistent inputs and outputs. There should be no duplicated work or overlapping responsibilities. Fourth, ensure clear hand-offs. Each agent must know what information to expect from the prior agent and what to pass to the next.
Fifth, track the performance of every agent, not just the final outcome. Evaluate quality at each step, so you can see where improvements are needed. Sixth, identify human intervention points in advance. Marr stresses that you must know where human-in-the-loop is non-negotiable, such as for payments or declining customer requests, and decide ahead of time what needs human involvement or approval.
Resist the Urge to Give Agents More Power
Marr's article includes a pointed warning for organizations tempted by increasingly capable AI. He believes that as agents become more powerful, the urge to give them more responsibility should be resisted. The focus, he argues, should be on building teams of simpler, specialist agents rather than on creating ever more complex autonomous systems.
For most organizations, Marr writes, the real value of agents will not come from replacing humans with ever more complex autonomous systems. Instead, that value comes from breaking complex processes into manageable, measurable steps that can be improved iteratively. The goal is collaboration, not complexity.
That philosophy extends to how organizations should approach their first attempts. Marr advises starting with a simple workflow to understand where automation and collaboration add value. First attempts will not always work out as planned, and he treats that as a learning experience, not a failure. Designing agentic workflows is new to everyone, not just you, he notes.
The multi-agent approach makes that learning process easier. Because you can diagnose and adjust one component at a time, you can iterate quickly and refine the system without starting over. That iterative improvement is where the real payoff lies.
A Pragmatic Path Forward
The Forbes article, which includes editorial standards and reprints & permissions, positions multi-agent workflows as a more effective strategy than single agents for complex business tasks. Marr's argument rests on a simple observation: complex tasks are easier to manage when they are broken into simple pieces.
The examples in marketing and customer service show how that principle applies in practice. The six-step guide gives organizations a concrete path to implementation. And the warning against overloading agents with responsibility keeps the focus on measurable, manageable progress.
For businesses exploring AI agents, the takeaway is clear. Do not ask one agent to do everything. Build a team of specialists, define their roles carefully, track their individual performance, and keep humans in the loop where it matters. Start small, learn from what does not work, and improve step by step.
The article was published on Aug 24, 2026, and is part of the Enterprise Tech topic on Forbes Innovation. Bernard Marr is a contributor to that section, and the article uses an image from Adobe Stock.