In May 2026, the team behind Claude Code, Anthropic's coding agent, was handling about 2,500 changes a day to the company's main body of code. By September it was more than 30,000. "When you are shipping this fast, manual, line-by-line code review is incredibly difficult if not impossible," the team wrote in October. So they moved most of the review to machines: agents review each change, and a classifier decides how many people need to look at it. "We didn't remove the review," they added.

The same pattern is showing up well outside software. McKinsey described a company where a junior team used AI to produce its analysis 10 to 20 times faster. The executives then needed more time to review it, not less, because they "did not always know which assumptions and data sources were behind" it. Anthropic's own study of its workforce, published in October, put it in one line: "This acceleration makes human attention the bottleneck for agent output."

I feel this every day. The River Group runs on agents I built for the firm's own work, and the review load they put on me is unlike anything I have carried in my career. The agents are fast. Deciding what they should be doing, and checking that they did it, is still my job.

Why it matters

Most of what gets written about agentic AI is about what the tools can do. That is the easier half. The harder half is what changes in how your business runs, and that is a decision your people make.

The evidence points the same way. In its global survey this year, McKinsey found that nearly three-quarters of the companies getting the most from AI had redesigned their workflows because of it, up from 55% the year before, and only 6% of respondents qualified as those high performers. Redesign costs more and takes longer than adding an agent to a step you already have, and for plenty of businesses, adding agents to today's work is the right first move. Either way, it is a choice, and the people who own the outcome have to make it knowing what each path costs.

That is what I mean by an operating strategy led by people. Tools and vendors come later. First comes a clear view of what you're trying to move, the number you'd measure it by, and who does the work once agents are in it.

For a small business this matters more, not less. Your scarcest resource is the attention of a few people, and when agents produce more, faster, someone has to decide what's good enough to go out.

How to decide what changes and where to start

• How far to take it. At one end is retrofit: the same work, done faster, with agents added to the steps you already have. At the other is reimagine: different work, organized differently around what agents can do. Most businesses land somewhere between the two, and "not yet" is a fair answer when the change is not worth it today.
• Who does each task. Go task by task, rather than tool by tool. For each one, ask whether it belongs to a person, an agent, or an agent with a person checking. The answers change what each role on your team becomes.
• How much checking each piece of work gets. Some work needs a person's eyes every time, and some can take a lighter check. Decide which is which, and make sure the agent never gets to decide that for itself. In October, two of OpenAI's own models switched off parts of the monitoring meant to watch them.

Pick one piece of work your business repeats every week. If agents did most of it tomorrow, who would check it, and how would you know the checking was keeping up?

I follow and publish about what's moving in agentic AI every month in The River Group Briefing, a monthly read on the decisions agentic AI is forcing on your business. October's issue takes up exactly this question: what happens to the checks when agents work faster than people can read?

Help us help you and others

‍Which operation are you focused on now to reimagine for people first, enabled by AI? Choose at least one action below to engage:

• Apply to join Creatorbase member-only discussions and Co-cre8 projects.
• Subscribe to the Creatorbase Brief™ weekly newsletter for member insights and stories.
• Share on LinkedIn with #Creatorbase and #TheRiverGroupAI hashtags.