The Org Chart With AI Employees On It: Where Agents Sit in the Team

Open your company’s org chart. Every box is a person, every line a reporting relationship, and the whole thing answers one question: who is responsible for what. Now your business is about to add a new kind of worker that drafts quotes, answers customers, reconciles invoices, and researches suppliers, around the clock, without a desk. The question the org chart exists to answer does not go away. It gets harder.

Most companies bolt AI agents on quietly, one team at a time, with no owner, no scope, and no place on the chart. That works for a pilot and falls apart at scale, when nobody can say how many agents exist, what they touch, or who answers when one gets it wrong. Gartner expects the average large enterprise to be running over 150,000 agents by 2028⁶. You cannot manage that with a shrug.

This article is about the unglamorous but decisive work of putting agents on the org chart: where they sit, who manages them, how reporting lines and accountability change, and what the German works-council and EU AI Act rules demand when a digital worker joins the team. It is written for the Mittelstand, where a handful of people carry most of the knowledge and the org design has to stay simple. The companies that get this right will not be the ones with the most agents. They will be the ones who always know who is responsible for what.

The New Box on the Chart Nobody Drew

An org chart is not bureaucracy for its own sake. It is the answer to three questions every functioning team needs settled: who does this work, who is accountable for it, and who do you go to when it breaks. For decades those questions only ever pointed at people. An AI agent that does recurring work breaks that assumption, because it does the work but has no obvious owner, no box, and no manager.

• The agent does the work - it handles a queue, produces output, and affects a number on someone’s dashboard, which is the definition of a workload.
• But nobody owns it - it was spun up by whoever built the pilot, who has long since moved on to the next project.
• And nobody supervises it - there is no manager reviewing its output the way a team lead reviews a junior’s.
• So when it errs, there is a gap - the question “who is responsible for this” has no name attached, which is exactly the situation an org chart exists to prevent.
• And it multiplies - one unowned agent is a curiosity; fifty unowned agents is a governance failure waiting to surface.

The fix is not complicated, but it is deliberate: decide which agents are real team members, give each one an owner and a scope, and draw them into the structure. The rest of this article is how to do that without turning your simple Mittelstand org into a tangle.

When Agents Become Headcount: It Is Already Happening

This is not a thought experiment. The largest professional firms in the world are already counting agents as part of their workforce, naming them, and reporting their numbers the way they report headcount. The framing has shifted from “tool we use” to “worker we have”.

The companies already doing it

• McKinsey - its CEO describes a total workforce of 60,000: roughly 40,000 humans and 25,000 personalised AI agents, with a goal of every employee paired with at least one agent¹.
• Goldman Sachs - it piloted an autonomous coding agent it openly calls a “new employee”, starting with hundreds and planning for thousands alongside its human developers².
• Salesforce - its CEO talks about “digital labor” and a “limitless workforce”, with internal support agents resolving the majority of cases without human escalation³.
• The broader market - more than 68 percent of organisations expect to have integrated AI agents into core operations by 2026, so this is becoming the norm, not the exception¹².

You do not need 25,000 agents for this to matter. The moment your second or third agent starts doing daily work, you face the same question these firms faced: is this a tool, or is it headcount? And if it is headcount, where does it sit?

How the Shape of the Org Actually Changes

The familiar org chart is a pyramid: a few leaders at the top, layers of management in the middle, a broad base of people doing the work. When agents take on the routine base-level work, the pyramid stops being the right shape. Analysts who study this expect two new silhouettes.

From pyramid to diamond and hourglass

• The diamond, for frontline work - agents absorb entry-level tasks, the narrow base shrinks, and the middle bulges with people who supervise agents and handle exceptions⁴.
• The hourglass, for knowledge work - strong junior and senior tiers remain, but the routine middle thins as agents take over coordination⁴.
• The new scarce skill - PwC calls it the AI-literate generalist: someone with enough cross-domain knowledge to supervise agents, interpret their output, and keep them aligned to the business⁴.
• The honest caveat - this is a direction of travel, not a finished state; only a minority of organisations have agents in production today, so most charts are mid-transition⁵.

Who Manages the Agents: The Rise of the Agent Manager

Every team member needs a manager, and agents are no exception. The mistake is to assume that manager is central IT. IT builds and runs the platform, but the person accountable for whether an agent does good work should be the one whose results it affects. A distinct role is emerging to hold that accountability.

What an agent manager actually does

• Owns the outcomes - they are accountable for the agents’ results the way a product manager is accountable for a product, not just for keeping them running⁹.
• Lives in the dashboards - the day is spent in scorecards and observability, watching volume, resolution rate, and exceptions⁹.
• Tunes and escalates - they adjust the agents, decide when to hand a case to a human, and when to retire an underperforming agent.
• It is a recognised category now - Mercer lists “agent supervisor” among the new job titles companies are creating, and most leaders expect HR to manage humans and agents side by side¹⁰.
• Managers are hiring for it - in Microsoft’s research a meaningful share of managers plan to bring in AI workforce managers and agent specialists within the next year and a half¹¹.

In a large firm the agent manager is a full-time job. In a 200-person Mittelstand company it is usually a hat an existing team lead wears, owning the two or three agents in their function. Either way, the principle is the same: every agent has one named human accountable for it.

Four Ways to Place an Agent in the Team

Not every agent sits in the same place. There are four recurring patterns for where an agent fits, and choosing the right one for each agent is what keeps the chart coherent instead of chaotic.

The four placement models

1. The personal copilot - tied to one person, who invokes it to work faster. It is a tool, not a team member, and does not get its own box. Most employees will have one.
2. The team member - owns a recurring workload and a queue within one function, reports to that function’s lead, and gets its own box. This is the classic “AI employee”.
3. The shared service - serves several teams at once, like a translation or document-processing agent, and sits in a shared-services or central function with one owner.
4. The supervisor agent - an agent that oversees other agents, monitoring and containing their actions, which Gartner expects 40 percent of CIOs to demand by 2028⁷.

Reporting Lines, RACI and Who Answers for a Mistake

The hardest question an agent on the org chart raises is accountability. An agent cannot be held responsible; it has no job to lose and no licence to revoke. So the line has to point at a human, and the chart has to make that line unambiguous before anything goes wrong, not after.

The accountability rules that hold up

• One owner per agent - every agent maps to exactly one named human who answers for its outcomes, the same one-manager principle you use for people.
• The agent is never the “R” alone - in RACI terms an agent can be responsible for doing the task, but a human is always accountable for the result.
• Decisions have a trail - you need to prove what an agent did, why, and under whose authority, which means identity for each agent and immutable logs of its actions⁵.
• Escalation is designed, not hoped for - every agent has a defined point where it stops and hands a case to its human owner, especially for anything touching money, customers, or staff.
• High-stakes work keeps a human in the loop - the more consequential and less reversible the decision, the tighter the human checkpoint, which is the core of any honest agent governance.

Giving each agent its own identity is also a security and audit requirement, not just an org-design nicety. We cover that layer in depth in our piece on agent identity and access, and the human-checkpoint design in human-in-the-loop governance.

The German Question: Betriebsrat and the EU AI Act

In Germany you cannot redraw the org chart with agents on it without two parties at the table: the works council and the EU AI Act. Both are often discovered late, as blockers, when they should be involved early, as design partners. Here is what actually applies.

Works-council co-determination

• Monitoring capability triggers it - under Section 87 (1) 6 of the Works Constitution Act, the mere technical ability of a system to monitor behaviour or performance triggers co-determination, and almost every agent produces logs that meet that bar¹⁷.
• Selection decisions are covered too - Section 95 extends co-determination to AI-supported guidelines for hiring, transfers, and dismissals¹⁷.
• Reorganising work can be a Betriebsänderung - a German labour court held that introducing an AI system that eliminated several roles in a department was an operational change under Section 111, triggering negotiation duties¹⁶.
• Skipping it is expensive - without works-council agreement, deployment can be declared void, with injunctions and removal orders, so the cheap path is early involvement¹⁷.

The EU AI Act layer

• HR agents are high-risk - agents used for recruitment, evaluation, promotion, or termination fall under the high-risk category in Annex III of the AI Act¹⁵.
• The high-risk deadline moved - under the 2026 Omnibus agreement those high-risk employment obligations were deferred to December 2027, which buys time but does not remove the duty¹⁵.
• Literacy already applies - the AI literacy duty under Article 4 has been in force since February 2025, so staff working with agents must be trained now¹⁵.
• GDPR already restricts automation - Article 22 limits fully automated decisions with legal or similar effect on a person, which keeps a human in the loop for consequential HR calls¹⁶.

How to Draw Your First AI-Augmented Org Chart

You do not need a reorganisation to start. A focused exercise turns a vague sense of “we have some agents around” into a clear chart that shows who owns what. Here is the sequence.

The step-by-step

1. Inventory every agent - list each agent doing recurring work, what it does, and which systems it touches. Most companies are surprised how many they already have.
2. Sort tool from team member - mark each as a personal copilot (off the chart) or a team member, shared service, or supervisor (on the chart).
3. Assign one owner each - give every on-chart agent exactly one named human accountable for its outcomes, usually the relevant line manager.
4. Define scope and escalation - write a one-line job description and a clear point where the agent stops and hands to its owner.
5. Bring in the Betriebsrat and check the AI Act - involve the works council early and flag any HR or high-risk agents for tighter human oversight.
6. Draw it and review it - put agents on the actual chart, then review the fleet on a regular cadence the way you review the team.

If you cannot point to one named owner for each agent, you do not yet have an org chart, you have agent sprawl with a friendly name. The checklist is what turns the second into the first.

How Superkind Fits

Superkind builds custom AI agents for SMEs and enterprises, and we design them to take a clear place in your team, not to float around as ungoverned tools. The approach is process-first: we start from how your org actually works and where an agent genuinely fits, then build it with an owner, a scope, and an escalation path from day one.

• Process-first placement - we map the function, the workload, and the people before deciding whether an agent is a team member, a shared service, or just a copilot.
• One owner per agent - every agent we deploy maps to one named human accountable for it, so there is never a blame gap.
• Built-in scope and escalation - each agent gets a clear job description and a defined point where it stops and hands to its owner.
• Identity and audit logs - every agent has its own identity and an immutable action trail, so you can prove what it did and why.
• Human-in-the-loop by design - the more consequential the work, the tighter the human checkpoint, tuned to your risk appetite.
• Betriebsrat-ready - scoped access and full logging are exactly what the works council needs to approve a deployment.
• Fits the existing chart - we slot agents into your structure rather than forcing a reorganisation around a platform.
• Outcomes, not licences - pricing is tied to a measurable first use case, not per-seat fees.

Decision Framework: Where Does This Agent Belong?

When a new agent is proposed, a few questions place it correctly on the chart and decide how much governance it needs. Run each agent through them before it goes live.

The pattern that works is the same one that works for hiring: get the first one right, learn how to manage it, then scale. For the underlying build-versus-hire decision behind each agent, our piece on new hire vs AI agent sets out the economics in detail.

Frequently Asked Questions

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI agents that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach.