Skip to content
Back to posts

Welcome to the Team. These Are Your Agents.

Notes by SDG ·

Welcome to the Team. These Are Your Agents.

I’ve been thinking about how work changes when agents become an integral part of the organization. This rabbit hole started when I was reading the PACT paper, and it was an interesting rabbit hole. First, I started wondering who actually owns the agent

https://notesbysdg.com/post/managed-8-ai-agents-coming-soon-to-a-cv-near-you

Then, what happens when agents start managing other agents?

https://notesbysdg.com/post/your-ai-agent-knows-the-rules-can-it-handle-its-manager

Now I have a third question: what happens when you join a company and someone says, “Here’s your team: six people and four agents”?

That sounds slightly funny today. I'm not sure it will sound funny for very long.

Imagine onboarding at a new job where your roster looks like this:

 Procurement agent: Manages supplier renewals.

 Forecast agent: Prepares the weekly forecast.

 Customer risk agent: Monitors churn risks.

 Pricing agent: Watches margins and price updates.

Then someone casually adds, “By the way, don’t touch the forecasting agent without checking with Finance.” Suddenly, your first few weeks get very interesting.

Today, when you join a company, you spend your initial weeks trying to understand the people, the processes, the systems, and the politics. In the future, you’ll have to understand the agents, too. This isn't just about prompt engineering—it’s about learning how to manage a new kind of team member.

The first thing I’d want to know isn’t technical—it's functional: What is this agent’s actual job? Does the forecasting agent just recommend a number, or can it submit it? Can it trigger downstream actions based on that number? Those are vastly different operational boundaries.

Next is authority. There is a world of difference between an agent saying, “I recommend supplier B,” and an agent automatically issuing a purchase order for six months of inventory to supplier B. (Very proactive, but potentially catastrophic.)

Then comes the most critical question: How does the agent think?

Not in a human sense, but through its blend of instructions, policies, context, tools, and feedback loops. When you inherit human reports, you quickly pick up on their working styles—who is cautious, who moves fast, who trusts data, and who relies on gut feel. Agents will develop recognizable patterns and blind spots, too.

 The customer-risk agent might overreact to a temporary drop in usage.

 The forecasting agent might break down during an unusual promotional campaign.

 The procurement agent might execute flawlessly on standard contracts but falter when a deal strays off-script.

You’ll need to know where your agents excel, where they're weak, and who is checking their work. Giving an autonomous agent broad authority without a clear escalation path is like handing a new intern a company credit card and saying, "Use your best judgment." It might go fine, or it might make Monday morning very awkward.

This brings us to culture.

Organizations run on unwritten rules: who really needs to approve an exception, which customers get VIP treatment, which processes everyone follows, and which ones people just pretend to follow. Over time, as agents process historical decisions and feedback, they won't just learn official company policy—they'll learn how the company actually behaves.

When you inherit an agent, one of your primary checks will be figuring out whether it's following official compliance or if it has absorbed "how we do things around here."

Which makes me think about the classic 1:1. When you take over a team, you sit down with your people to understand how they operate. Eventually, we’ll do something similar with agents.

Instead of "How was your weekend?", the agent 1:1 will sound like:

 Show me your key decisions from the last 30 days. Which ones required human approval?

 Where did you encounter high uncertainty? What triggered an escalation?

 Show me cases where a human rejected your recommendation.

 Which other agents or systems do you rely on most?

Today, institutional memory lives in people’s heads. When someone leaves, half the operational context walks out the door. In an agentic organization, that knowledge resides in instructions, system memory, permissions, and workflow graphs. Handover docs will need to cover two org charts: the human team and the agent stack.

This points to a broader shift in how we should prepare managers. The most valuable skill won't be building agents, but leading them—knowing how to delegate, how to audit output, how much authority to grant, and when to override an agent that's confidently telling you everything is fine. The core capabilities remain distinctly human: judgment, governance, and organizational awareness.

And that leaves us with an even bigger question for entry-level talent: If agents are doing the baseline work, how do new professionals gain the functional experience required to manage those agents later?

Traditionally, judgment is built through repetition. A junior analyst learns forecasting by building forecasts; a buyer learns vendor management by working through difficult contracts; a finance leader learns the business by spending years in the numbers.

If agents increasingly automate those entry and mid-level tasks, where does foundational experience come from? How do you develop intuition for a domain if you’ve never done the manual work? How do you know when an agent is wrong if you’ve never spent years learning what "right" looks like?

The future of work isn't just about figuring out what agents can execute. It’s about figuring out how humans build the expertise required to lead them.