I recently read Ethan Mollick’s latest post, where he called this moment the twilight of the chatbots.1 His point is that the way we work with AI is shifting: from working alongside it and checking every step, to handing it whole tasks and letting it run. It is what made me want to focus this issue on AI agents. Read any AI news, sit in on any conference, and you hear the same thing: teams are going agentic, and the industry has moved from chatbots to agents.
And it is happening now, well beyond engineering. When OpenAI studied its own teams, non-technical functions like legal, finance, and recruiting shifted most of their AI work to agents once a no-code option arrived, several of them adopting faster than the engineers.2 Microsoft’s 2026 workplace study points the same way, with agents now in use across every industry.3
In this issue I want to make the case for why domain expertise matters more in the agent era, not less, and then walk through what an agent actually is and the parts that make one up.
Before you read further, I am running another free Maven Lightning Lesson, Think in Workflows: Agentic AI for Business Leaders, where we dig into why thinking in end-to-end workflows is where the real leverage with AI lives. If that is the shift you are trying to make with your team, come join us.
The shift: A chatbot fills in what you do not know. An agent does the work and hands it back. Your job changes from doing the task to judging whether it was done well.
Why your expertise matters more, not less
When AI does more of the work, it is easy to assume your expertise counts for less. It counts for more.
A study of people using Claude Code on real tasks found something the headlines missed. Software engineers did not have an edge over people from other fields. What predicted success was not job title. It was domain expertise: the more someone understood the problem, the better the results they got.1
An agent can produce a mountain of work in an hour. So the bottleneck is no longer whether it can do the task. It is whether you can tell that it did the task well. Catching the answer that is confident and wrong, the assumption that does not hold, the case it missed: that is judgment, and judgment is the part that does not commoditize. Before long the same models will be everywhere. Your judgment will not be. Mollick’s advice for this moment is to think of yourself less as a user and more as a manager.1
And it is not only deep expertise that gets more valuable. It is the judgment around it. When Microsoft asked people which human skills matter more as AI takes on more work, the top two were quality control of AI’s output (50%) and critical thinking (46%). Most went further: 86% said they treat what AI gives them as a starting point, not a final answer, and that they stay responsible for the thinking.3 The job is shifting from producing the work to evaluating it, refining it, and owning it.
So before you can manage an agent well, you need to know exactly what one is.
The five parts of an agent
An agent is not a smarter chatbot. It is a different shape of working, and it takes five parts:
A goal. An outcome, not a single question. “Turn these five articles into my Monday brief,” not “summarize this.”
Context. What it knows about your world: your company, your voice, your standards, what good actually looks like. This is where your expertise goes in, and the part most people skip.
Tools. It can act, not just talk. Search the web, read a file, draft an email, pull your calendar.
Memory. It holds the context and the thread across steps, so you are not re-explaining every time.
The loop. It takes several steps in a row and checks its own work, instead of stopping after one answer.
When all five are present, you have an agent. Missing even one, and you do not.
The rule: The agent does the work. You decide if it is right. Guard the second thing. It is the only part that stays scarce.
What leaders should know
In June, Anthropic released Claude Tag, an AI agent that works as a teammate inside Slack. Anyone on the team can tag it, hand off a task, and it plans the steps, uses your connected tools and data, remembers the context of the channel, and delivers the result back in the thread. That is all five parts of an agent, and for teams that run on Slack, it sits right where the work already happens.4
So what: Agents like this are becoming teammates, not tools you open and close. The skill that matters most now is knowing what an agent is, how to put it to work, and how to judge what it hands back.
Found this useful? Forward it to one person who needs to hear it. Getting this for the first time? Subscribe free here.
Think Better. Execute Faster.
See you next week - Purti
Clarity Prompts - The weekly AI operating system for sharper leaders.
Sources
Ethan Mollick, “The twilight of the chatbots,” One Useful Thing, June 30, 2026. oneusefulthing.org
Drew Johnston, David Holtz, et al., “The Shift to Agentic AI: Evidence from Codex,” OpenAI, June 2026. arxiv.org
Microsoft, “2026 Work Trend Index Annual Report: Agents, Human Agency, and the Opportunity for Every Organization,” WorkLab, May 2026. microsoft.com
Anthropic, “Introducing Claude Tag,” June 23, 2026. anthropic.com
