When it comes to agentic AI, the fear of factor avoidance is palpable. Organizations, in part, because it seems like everyone else is doing it. But fomo is not a business strategy. To make agentic AI work, business leaders must ignore the hype and focus on understanding which agents can do what, how and why, at what cost, and focus.
Our own work has proven that AI agents that independently plan and execute complex multi-story tasks can deliver significant value by accelerating timelines and reducing costs. And this is just the beginning. AI agents working to work, communicate and learn with humans, communicating and learning, can become an original paradigm shift for how work is done.
Clear business value
But enthusiasm doesn’t always lead to impact, something many businesses are beginning to recognize. According to a study, 40% of agent AI projects could be canceled by the end of 2027 due to increased work cost and costs.
In recent research, McKinsey studied 50 agent AI initiatives in which we are directly involved. With Hinsight’s wisdom, we’ve identified three critical factors in Agent AI success.
1. Start with workflows, not agents
Agentic changes are more successful when they are focused on agent-integrated workflows, redesigned workflows that touch agents rather than procedures designed for other technological eras. And the Corollary is also true: even the most powerful AI agent is tied to faulty and inefficient workflows.
Already, agents are successfully deployed in multi-step, dynamic workflows, such as software development and customer service and customer service. Even the most daring leaders successfully deploy agents for boundary use cases. For example, an alternative legal services provider made significant efficiency gains where they carefully modernized their contract maintenance process. Each time an agent made a change in the document editor, it was entered, classified, and fed back into the agent’s logic and knowledge base. In designing an agentic workflow, the team determined when and how to integrate human input. Agents highlighted outliers and anomalies for people to review. Over time, agents have been able to codify new practice and provide more sophisticated legal reasoning, but it is up to lawyers to navigate critical decisions.
2. Stop punching
Many enthusiastic early adopters say their output is known as “slop”, which can be done quickly but requires considerable effort to correct. It’s annoying. Worse, it mistrusts agents and the idea of conversion in general. To do better, companies need to systematically invest in agents as they do in the workplace, with job descriptions, training, monitoring and continuous improvement targets.
3. Engage the workforce to support AI agents
There must be people walking, training and evaluating agents in an ongoing way: “Start and release” is not good enough. As agents begin to achieve more, the roles will change. Leaders are working to bring employees together in a new human-agent hybrid operating model, including devices, to effectively prepare, design, monitor, track, and correct their work, and bring them together to perform more complex tasks.
The underlying principle is that Agentic AI should work with non-time-consuming business priorities, such as productivity and teamwork. The question, then, can agents help solve real-world problems and create value as well as in other technologies?
And the answer is: not always. For tasks that involve generative AI applications like chat, generative AI applications like Chatbots are a better choice. For highly structured or automated tasks such as data entry, rule-based approaches—if x, then y—may be more efficient. High-stakes decisions with little room for error are the domain of leaders and managers.
Yes, agentic AI could be a once-in-a-generation opportunity, hence the fomo effect. Success does not come from enthusiasm, but from a thorough analysis of how this tool can be used for the right task, for the right task.