AI does it faster, but someone has to take ownership

getty

What’s it like to build a business run entirely by teams of AI agents? Bret Greenstein, chief AI officer at West Monroe, explained how he launched an experimental business using fleets of AI agents and subagents running marketing, finance, sales, and product development. The agents were impressive in their performance and speed, but lacked a key ingredient – human oversight. The agents "just kept giving me to-dos," he related. "It tells me you need to approve things or we can’t process, can’t proceed.”

The bottom line is no matter how fast and automated a process can be, humans are still accountable at the end, Greenstein recounted in a recent CAIO Connect podcast. “The product manager is responsible for the code that the agents are creating. The sales leaders responsible for all the research and input that went into that customer or that client. Autonomy is amazing, but it’s still up to us.”

Running a business requires, for example, a finance person who is supported by agents. “I want someone accountable for finance,” he said. "Someone who’s going to keep me out of regulatory trouble. Someone who’s going to make sure my taxes are done right, my billing is done right. Somebody has to own that and then use AI to do as much work as that can be done.”

AI and AI agents are misused if they’re only used as a vehicle to cut jobs and costs. Instead, they will be instrumental in a concept called “agentic process reengineering.” Don’t look at the activities employees or teams are undertaking – look at the outcomes, Greenstein urged. As part of such a transformation, “understand what AI is good at, strengths and weaknesses, and where you could apply it to the different activities. Decompose work not as a process, but as a series of activities that produce an outcome.”

For example, look at and evaluate the steps a finance person does in the month-end close. “What does the month-end close have to look like?” Greenstein asked. "And how much of that work could I reimagine letting AI do it? Then where’s the human-in-the-loop moments to review QA and provide judgment on it?”

The benefit may be a substantial reductions in time and human input. Agentic process re-engineering requires understanding AI, understanding business processes, and understanding functional business domains.

It differs from business process reengineering in some important ways, Greenstein explained. “It’s the combination of traditional business process re-engineering, but optimizing business processes for the work that AI can do. You’re trying to find activities that AI can be very good at. Researching, summarizing, analyzing, all you know, all those things, reviewing against a policy or a standard, extracting entities from documents, all that stuff AI is really good at. Shift those activities in AI and allow people to review outputs or exceptions or where approvals are needed.”

The rhythm of work is different with agentic process reengineering, he added. “You’re gated by moments where humans have to review something, or data feeds that aren’t working. You got to think differently at how you even manage a process like that.”

Advanced models such as the recent release of GPT-5.3-Codex-Spark offer “a thousand tokens a second – you can go from idea to code in seconds,” said Greenstein. “But you still have to have ideas. You know what problem you are solving. You need a vision for how it would work. You need to think about the user experience and what it would do. AI can help you produce this faster, but you still have to be able to think of them."