The Proactive Agentic Agent
AI can now do the work, but it still needs to be told what to do.
Like many of you, I have been blown away by the pace of AI development in 2026. While AI chatbots were gathering steam in 2023 and 2024, the promise of agentic AI was going to propel AI from a better way to search the web to a fundamentally different way to get work done.
And now, with Claude Cowork and MCP, agentic AI is here, and it is indeed transformational. The other day, Cowork redlined a document for me based on information it found in my email and Office files, saving me at least an hour.
And yet, I still want more.
To pull on the thread of the truly great assistant, Cowork is now one of the best assistants you can hire…provided you tell it exactly what you want it to do.
But what makes for a truly one-of-a-kind assistant? It’s the one who doesn’t need to be told what to do. The one who sees an email from an important client asking for an updated report, and, on their own, prepares the report, drafts the reply, and queues it up in your outbox, ready for your approval.
This is the go-getter assistant, otherwise known as a proactive assistant.
What I’m looking for now is the proactive agentic agent. Completing tasks for me when prompted is game-changing. But I’m ready for the agent who knows what I need to get done without being told what to do.
For instance, the proactive agent should see a task in my Todoist or an email in my Superhuman and put the pieces together for me.
With MCP, Claude can now pull data from Notion, Slack, and Google Drive. And, with Cowork, it can assemble Word, Excel, and PowerPoint files.
The ingredients are all there; I just want it to go from being a great assistant to one that has my back and makes me better at what I do.
Today’s agentic AI still requires a fair amount of prompting. And yes, some of what I am looking for could be handled by scheduling a recurring task in Cowork to check my email. But that’s not the proactive agent I’m looking for.
The proactive agent I’m looking for operates 24/7 in the background and acts on new data: new emails, new Slack messages, new tasks. And when I fire up my email client, emails are already drafted, with reports waiting for me.
I’m not writing this because I doubt we’ll get there; I’m writing this because I 100% believe we will, and quickly. At the pace Anthropic is shipping, we’re talking weeks, maybe even days.
So I’m writing this down as a marker: ‘Here’s something we can’t quite do yet, but will soon.’ And I don’t think it will be long before we see truly proactive agentic agents.
