---
title: "The five stages of an agentic workforce"
description: "From random acts of AI to humans and agents on one org chart: what each stage looks like, who owns the AI, and the one move that gets you to the next."
date: "2026-10-23"
author: "Michael Meinhardt, CEO"
category: "Resources"
source: "https://www.superorgs.com/blog/agentic-workforce-stages"
---

# The five stages of an agentic workforce

From random acts of AI to humans and agents on one org chart: what each stage looks like, who owns the AI, and the one move that gets you to the next.

The Agentic Workforce Maturity Model has five stages, from random acts of AI to a company-wide workforce of humans and agents on one org chart. Every stage asks the same question: does this piece of AI have a name, a manager, a cost, and a job, the way every employee does?

We built the model by tracking a real company through all five stages as it happened. Next to each stage we put where third-party research places most companies today, so you can see where your company sits against the market, not against an ideal. [What separates the companies getting a return on AI](/blog/return-on-ai) has the research behind the question.

![The five stages as rising columns: Random Acts of AI, Workflow Automation, Named Agents, Agentic Pods, and Company-Wide Workforce, each with a market figure and its next move. Below, who owns the AI at each stage: no one, whoever built it, a manager per agent inside one department, a pod of owner, builder and expert, and the company on one chart.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-1-five-stages.png)

Figure 1. The five stages, and who owns the AI at each one.

## Stage 1: Random acts of AI

**What it looks like.** Employees have their own AI licenses, but usage is personal and almost invisible above the individual. Nobody has a shared picture of what is being used, for what, or with what result.

**Where the market sits.** This is the most crowded stage. [Airtable's 2026 research](https://www.airtable.com/articles/ai-agent-security) puts 56% of organizations here, using AI for individual productivity but not across systems. The habit exists, but the company's structure has not caught up with it.

![An AI spend inventory at stage 1: a chat assistant, a writing assistant, a coding copilot, and personal accounts, each with an unknown seat count, no clear owner, and a cost buried in other software or expensed. Total AI spend: no one owns this number.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-2-ai-spend-inventory.png)

Figure 2. A stage 1 inventory. Illustrative.

**The move to stage 2.** Stop trying to measure usage and start trying to see it. That means one shared, honest inventory: every tool, every seat, and every dollar, in one place, visible to more than one person. [Our guide to AI agent sprawl](/blog/ai-agent-sprawl-how-to-find-every-agent-your-company-runs-and-who-owns-it) covers where to look.

## Stage 2: Workflow automation

**What it looks like.** The personal habit becomes a company habit, in the shape of point automations: lead routing, data enrichment, SDR sequences, renewal alerts. Dozens or hundreds of them, often stitched together in a tool like n8n. It works, and it is a real step past stage 1, but the gains fade and nobody can say why.

**Where the market sits.** [Gartner predicts](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027) that over 40% of agentic AI projects will be canceled by the end of 2027, and we think this is the stage it describes. The automation logic is rarely the problem. Nothing at this stage carries a name, an owner, or a register of what exists, so when something breaks, there is nobody whose job it was to know.

![Live automations found by asking around: lead routing built by a former employee and broken for three weeks, data enrichment with an unknown owner, SDR sequences degrading, renewal alerts running, health scores that two people think they own, and competitive intel failing silently. Plus 140 more across sales, marketing and customer success, with zero documented owners.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-3-live-automations.png)

Figure 3. A stage 2 automation list, found by asking around. Illustrative.

**The move to stage 3.** Give every workflow an owner before you build the next one. Not a team: one person for each workflow. The instinct at this stage is to keep building, because building feels like progress. Stop long enough to attach a name and a cost to what already exists, so the next automation adds to something you can see.

## Stage 3: Named agents

**What it looks like.** Automation starts to look like a workforce. Agents get roles, managers, defined tasks, even performance reviews. The catch is that it happens one department at a time. Marketing's agents do not know RevOps's agents exist, and the same job gets built twice.

![Named but siloed: marketing, RevOps, and customer success each have a manager and their own agents. Paid media is built twice, in two departments that cannot see each other. Each department sees its own agents; no one sees all of them.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-4-named-but-siloed.png)

Figure 4. Named agents, one department at a time. Illustrative.

**Where the market sits.** [Deloitte's 2026 survey](https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/services/consulting/2026/state-of-ai-2026.pdf) found that only 21% of companies report a mature model for governing autonomous agents. Named ownership inside one department is real progress, but it solves the problem locally and leaves it unsolved for the company.

**What helps.** Sendoso learned that asking an employee to imagine what an agent should do rarely produced anything useful, while asking the same person to write a job description for a new hire did. Your best people know how to describe a job. Very few have practice describing an agent. A job description also keeps each agent's scope small, which matters: when one large all-purpose agent gets something wrong, nobody can tell who is accountable.

![A job posting for an agent role, Web Specialist: reports to the Director of Marketing; scope, site performance and SEO; not in scope, paid channels and budget; success, one metric, reviewed like any hire.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-5-job-posting.png)

Figure 5. An agent's job posting: who it reports to, its scope, what is out of scope, and how success is measured.

**The move to stage 4.** Make what works in one department visible to the rest of the company, before you build the next agent inside another silo. Ask team leads to write the job posting they would use if the role had to be filled by a person. The scope, the reporting line, and the success metric usually fall out of that exercise.

## Stage 4: Agentic pods

**What it looks like.** A named owner, a technical builder, and a business expert share accountability for one defined part of the business, with one scoreboard everyone can see. This is where AI stops being a department's side project and starts working as an operating unit that proves its value.

![The anatomy of a pod: an owner, named and accountable; a builder, the AI engineer; and an expert from the business, around a set of agents. A shared scoreboard shows operational numbers, task completion, override rate and health score, next to the business number the pod exists to move.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-6-pod.png)

Figure 6. A pod and its shared scoreboard. Illustrative values.

A pod tracks its agents two ways: operationally, through task completion, override rates, and health, and against the business number the pod exists to move.

**Where the market sits.** Early, and without a shared name yet. Leaders have started to describe the same structure:

- **At Uber,** CTO [Praveen Neppalli Naga](https://www.linkedin.com/posts/pneppalli_agentic-ai-adoption-is-on-fire-at-uber-and-activity-7480367291851833344-6Lhm) wrote that the company handpicked about 30 of its most AI-proficient engineers and paired each with a domain expert from a business function.
- **At HubSpot,** [Kieran Flanagan](https://www.linkedin.com/posts/kieranjflanagan_heres-how-to-use-agentic-pods-across-your-activity-7483145550033309697-MQFR), SVP of Agentic GTM and Systems, laid out a version for go-to-market teams: "To me, this model is the only way to transform GTM functions around AI rather than AI sprawl, where lots of tasks are automated by AI, but it's unclear whether they make your overall GTM teams better or worse."
- **[Carolyn Healey](https://www.linkedin.com/posts/carolynhealey_your-newest-ai-agents-dont-need-better-models-activity-7486404418997952512-zni-),** a fractional CMO, put it in two sentences: "Your newest AI agents don't need better models. They need an org chart."

**The move to stage 5.** Resist declaring victory because of one strong pod. A pod that works in isolation is still an anecdote, not a strategy. Give the next function the same structure and the same visibility, and use the same scoreboard for every pod, so leadership can eventually see all of them at once.

## Stage 5: Company-wide workforce

**What it looks like.** Humans and agents sit on the same org chart, across every function, not only the ones that got early attention. Ownership and real cost, not estimated cost, are visible at every level, from one agent up to the whole company.

**Where the market sits.** [McKinsey's AI high performers](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2025), about 6% of the respondents in its 2025 survey, are the closest real-world proxy for this stage. Being here is rare.

**Where it is headed.** The few companies at this stage are aiming for hundreds of agents, beyond go-to-market and into finance, engineering, and back-office work. Just as important as the agent count is training people to manage agents.

Their stated goal is to decouple revenue growth from headcount growth: adding fewer people and more agents as the business scales, without treating that as a cover for layoffs. Some roles shift. People, especially those early in their careers who already work fluently with AI, get a different job: managing the agents instead of doing the repetitive work the agents now do.

![A conceptual chart, not to scale: from today to scale, revenue rises fastest, agents rise with it, and headcount grows slowly.](https://www.superorgs.com/blog/agentic-workforce-stages/figure-7-decoupling-growth.png)

Figure 7. Growing revenue faster than headcount. Conceptual, not to scale.

**What comes after.** Stage 5 is not a finish line. It is an operating state you maintain. The work shifts from building the system to running it, and the companies here are best placed to keep adding agents and functions without losing track of who owns what.

## Your stage is not a verdict

Nobody skips from stage 1 to stage 4, though some get there faster. Most leadership teams will also find different parts of the company at different stages at once: marketing might run solid pods while finance has not left stage 1. That is useful, because it shows where the next investment of attention should go.

What should feel different after reading this is not urgency. Urgency is what put most companies in stages 1 and 2, buying tools because everyone else was. What should feel different is clarity: one small, specific next step at every stage.

## Find your stage

The full guide has a self-assessment for every stage: four statements, and if most are true, that is your stage. It also has a worksheet to map each team, and the sources behind every number. It is free.

- [Get the maturity model](/maturity-model)
- [Read the research](/blog/return-on-ai)