I've long held a thesis that AI agents are the future operating system for employees. My v1 of this was an employee performing the majority of their work through an agent. We saw early glimpses of this with coding agents, but it wasn’t until Claude Cowork came out that the vision solidified into how the general knowledge worker would work. You give the agent the right context, access to the right tools, and an objective. Press enter, and it goes out and executes.
I shared my first experience testing this with some basic automation of weekly reports, which I still use today. I increasingly find myself setting up scheduled tasks to automate my busy work. From transcribing notes into a CRM after a demo to prepping for meetings, everything is just smoother.
Anthropic is seeing this across a sample of 600,000+ companies that use Cowork. Their analysis of 1.2 million anonymized Cowork sessions from May 11–31, 2026, found that half of all usage fell into just two categories:
33.4% was business process and operations. All the busy work between the objective you receive and the outcome you want. This is exactly how I've been using it in my day-to-day work.
16.4% was content creation and copywriting. I use this to proofread documents I'm writing. Others use it to fill in a blank page, often the greatest point of friction to progress.
While this all feels individual, it has become the new way forward-looking employees operate. It truly is an interface into a new way of working.
Why executives are pushing AI
Executives and managers see the instant value because agents work the way management always has. You collect the right context, set a clear goal, and provide the team with clarity on the desired outcome. Okay, so not all managers/executives do that, but the good ones do. The team goes to work and gets a v1 in place. They bring it back to management, who reviews it, provides feedback, and iteration happens until the objective is met.
Sound familiar? That's exactly how power users are using agents today.
So it can feel like a shock for individual contributors who have been doing the deep-dive work, striving for perfection, but always being the one to do it. As much as I hate the phrase “people will be managing a team of agents”…well, people will be managing a team of agents.
Your new team of agents
The team dynamic gets really interesting with agents. An individual contributor can now access a team of agents that can provide different perspectives on a problem. It becomes a multi-agent debate in which specialized agents look at the problem through a particular lens, offer their opinions, and a separate agent decides the best course of action based on those perspectives.
Zoom out, and this emulates exactly how humans work today. As a leader, you want a diverse range of perspectives on a problem. You want the super smart neckbeard who thinks everything is impossible sitting next to the eternally optimistic business person who thinks anything is possible. The answer lies somewhere in between, but it's up to someone to facilitate that discussion and drive things forward.
Imagine a world where every employee has a team of agents at their fingertips, in an interface (aka the harness) that collaborates and drives everything toward the target outcomes.
This is why executives are so bullish on the future of work.
From scheduled tasks to digital employees
I've been experimenting a lot with Claude Tag recently, and I'm seeing early indications of these digital employees filling the gaps. I'm evolving from a series of scheduled Claude tasks to having Claude sit in Slack (which the company runs on). Claude keeps track of what’s going on and automatically picks up tasks for me.
My random questions don’t have to eat up a human knowledge worker's time (as much) anymore. I can ask Claude to pull information and get me ready for a meeting. It’s becoming the connective tissue between processes.
Daniel Miessler believes we just reached the first level of Artificial General Intelligence (AGI) with Claude Tag. I don't know if I'm quite ready to agree, but his argument hinges on the definition of AGI. He's pushing for it to be "An AI system that can replace an average knowledge worker." His strongest point: everyone, including himself, was waiting for a technology upgrade to reach AGI. He makes the most powerful observation: instead of a model update, AGI may first appear as an integration into a workflow.
When you meet people where they work and solve busywork there, that’s the start of a powerful transformation. This is the key for every business moving forward.
How do you secure a workforce that's increasingly not human?
This is what I spend so much of my time thinking about, because we're heading toward a future where agents far outnumber employees in an organization. The unfortunate truth is that adoption is outpacing security architecture, especially in things like identity.
But that doesn’t mean you do nothing. I continuously come back to the same security fundamentals that have carried us through every emerging technology:
Get visibility into what's in the environment. We talk about agents as an operating system, but today, companies are completely in the dark about what agents are doing. I had a conversation this week where the head of IT said the business is using Claude on personal accounts and connecting it to resources in the environment. He had no idea how bad it was because he had no visibility into the environment.
It's an uncomfortable spot to be in. Start by gaining visibility into the agents in your environment, the tools they're connected to, and the permissions they’ve been granted.
Find the misconfigured agents. Spoiler alert: the number-one finding we see in customer environments is that agents are given way too many permissions for the tasks they perform. Full wildcard access to local systems and production environments, meaning they can access anything and run any command.
It’s not the warm, fuzzy feeling you want to have about your agentic workforce. The visibility you gained gives you the first step in clawing back that access and getting it down to a reasonable level. This is all about least agency.
Monitor for malicious or rogue behavior. This is Evoke’s core focus because agents require an upgrade in how security tools should work. Look at how we got to EDR. We started with signature analysis in antivirus: malware gets identified, a signature gets created, and you hunt for it across your systems. That evolved into heuristic scanning (aka "next-gen AV"), which monitored system behavior for signals that stand out as malicious. Then EDR took behavior to the next level by building in investigation and containment.
Agents require a deeper introspection. You need to understand how the user is using the agent, just as they would an application. You're watching the actions the agent takes, whether it's deviating from its goal or being misdirected by something malicious or just accidental. It's a purely behavioral analysis game, overlaid with known malicious actions and indicators.
That's not something companies can tackle themselves today. With the agentic workforce ramping up faster than anyone expected, the best time to get started securing them is now. If you’re ready to reclaim visibility and control over what agents are doing in your environment, let’s chat.
