AX Field Notes

All casesNo. 067

AGCO Turns Volunteer Makers Into a Governed Agent Program

Staff name the friction, AI leads shape agents, some quality reviews fall to about an hour

Case no.
067
Primary source
Microsoft customer story
Source published
2026-07-06

1. Why this case

AGCO Corporation builds agricultural machinery and precision agriculture technology. Through brands including Fendt, Massey Ferguson, Valtra, and PTx, it serves customers in 140 countries. Microsoft’s customer page lists it in the “10,000+ employees” size band, with the United States as its country.

The case is useful for two reasons. First, it shows a manufacturer letting employees build agents themselves, at scale, without letting that become shadow IT. At a leadership meeting of about 2,200 people, about 900 volunteered on the spot to become agent “makers.” Training started the next day. The maker community has since grown to roughly 2,000 across AGCO, and several hundred enterprise agents are in production.

Second, it names where the slow work sat. Quality and warranty processes could take weeks, and sometimes months, and much of the effort depended on a small group of experts. The story says AGCO is now using AI agents to “rapidly interpret, validate, and advance quality issues,” and the story’s summary says this is “cutting some quality reviews from weeks to about an hour.”

All figures come from a Microsoft customer story published on 2026-07-06. They are vendor-published and not independently audited.


2. Problem

The story describes two problems: one about how employees were using AI, and one about how quality work moved.

Curiosity heading toward shadow IT. Leaders saw employees “learning about AI, experimenting on personal devices, and looking for faster ways to solve local problems.” The story frames the risk this way: “The risk was not curiosity; it was letting that curiosity become unmanaged shadow IT.” The executive summary states the goal as wanting employees “to use AI safely, not experiment with disconnected tools outside enterprise governance.”

Quality work that waited on a few experts. Aryn Drawdy, Director of Strategic Partnerships at AGCO, explains why manufacturing coordination is hard: “Product improvements are often implemented as running changes within active production builds, which require coordination across multiple teams.” According to the story, these handoffs “don’t follow a linear process, and much of the communication happens outside structured workflows.”

The result: “quality and warranty processes can stretch over weeks, and sometimes months, as teams reviewed issues, validated data, and aligned on next steps. Much of that effort depended on a small group of experts who interpreted data and guided decisions across teams.”

Confirmed facts: employees experimenting with AI on personal devices; stated risk = unmanaged shadow IT; product improvements made as running changes in active production builds, needing multi-team coordination; non-linear handoffs, communication outside structured workflows; quality and warranty processes taking weeks, sometimes months; dependence on a small group of experts.


3. Solution

3.1 Start from friction, on tools employees already use

Drawdy’s starting rule is about the question asked: “I think asking how AI can help you is a mistake. I think asking where do you see gaps today, where are your friction points, that’s where we have to start.”

AGCO chose Microsoft Copilot Studio for building agents and Microsoft 365 Copilot for everyday productivity, “because employees already worked across Microsoft Word, Microsoft Excel, Microsoft PowerPoint, Microsoft OneDrive, and Microsoft SharePoint.” Drawdy describes the aim: “For citizen development with AI, we wanted to find the fastest path to market. We wanted to make sure that our employees were comfortable with it.”

3.2 Volunteers first, training the next day

AGCO treated AI literacy as the foundation. “At a leadership meeting with about 2,200 people, the company asked who wanted to start building agents. About 900 employees immediately raised their hands and volunteered to become makers.” AGCO “provided them with access, began training the next day, and used Microsoft Teams to support a growing maker community where employees could ask questions and help one another.”

3.3 Governance as support, with a split in who does what

Drawdy describes the governance stance: “It’s not shadow IT—it’s support. We work alongside teams to ensure they’re using Microsoft tools the right way and partnering effectively so they can close gaps, move faster, and show ROI through new ways of working.”

The division of work is stated twice in the story. “In practice, employees surface friction points while AGCO’s AI leaders and Microsoft specialists help shape the right solutions.” And: “teams bring the pain point, and AI champions shape the automation, agent, or workflow that removes friction.” The company “is not asking supply chain, quality, or agronomy experts to become AI experts overnight.”

3.4 Three kinds of agents, and a path between them

The story describes the portfolio as “an agentic framework rather than a one-off solution.” It has layers:

  • Citizen-built and team agents. “AGCO has hundreds of citizen-built agents and shared team agents”.
  • Enterprise solutions co-created with Microsoft. “larger enterprise solutions are being co-created with Microsoft.”
  • A path into production. AGCO “now has several hundred enterprise agents in production, with several hundred more moving through cohorts designed to consolidate and optimize them for enterprise use.”

3.5 Named agents

  • Quality. AGCO “is using AI agents to rapidly interpret, validate, and advance quality issues through the process.” The story says the quality framework “is expected to include a series of agents rather than a single bot.” The stated aim is that agents “bring the right context into the process, connect data across design and supply chain systems, and move issues forward more quickly.” The story adds: “The goal is not to replace expertise, but to make it easier to apply that expertise where it’s needed”.
  • Supplier. “A supplier agent reads contracts and identifies where AI suppliers could increase value and potential savings over the next year or two.”
  • Agronomy. An agronomy agent “combines AGCO agronomy studies with sales guidance and web content to help sales teams connect agricultural issues to relevant machines.”

Drawdy on the framework: “We’re really working with the Microsoft team to enable this agentic framework. It’s not just one agent—it’s going to be several when we’re done with it.”

Confirmed facts: Copilot Studio for agent building, Microsoft 365 Copilot for productivity; about 900 of about 2,200 at a leadership meeting volunteered; access given and training begun the next day; Teams maker community; employees surface friction points, AI leaders and Microsoft specialists shape solutions; hundreds of citizen-built and shared team agents; larger enterprise solutions co-created with Microsoft; several hundred enterprise agents in production and several hundred more in consolidation cohorts; quality agents in use, quality framework expected to include a series of agents; supplier and agronomy agents described.


4. Impact

Figures from the Microsoft customer story dated 2026-07-06. No baseline dates or measurement methods are given.

MetricFigureBasis
Volunteer makers at launchabout 900Out of about 2,200 at one leadership meeting
Maker community900-plus → roughly 2,000“across AGCO”
Enterprise agents in productionseveral hundredPlus several hundred more in consolidation cohorts
Citizen-built and shared team agentshundredsSeparate count from enterprise agents
Quality review timeweeks → about an hour“some quality reviews”; no count or review type given

On demand from employees, Drawdy said: “Every day somebody comes to me and says, ‘How do I get a Copilot Studio license? What do I need to do?’”

On quality, Drawdy describes the change: “We’re moving from a time-consuming process of review, analysis, and alignment to a vastly improved state—where issues are read, validated, and quickly advanced to the next step.” The story states the expected effect in the future tense: “This shift will accelerate resolution cycles, reduce dependency on constrained expertise, and enable faster, more responsive support for manufacturing teams and farmers.”

Todd Bailey, Vice President & CDAIO at AGCO: “This is not about experimenting with AI for its own sake—it’s about applying it where it matters most. Together with Microsoft, we’re building an AI foundation that helps AGCO move faster, reduce friction, and equip our teams to make better decisions for the farmers and customers we serve.”

Next. AGCO names supply chain, product management, and customer acquisition as “priority focus areas for the remainder of 2026,” alongside AI literacy and upskilling. Drawdy: “We’re looking at how we can safely accelerate? We’re looking at what literacy means to the organization? How are we upskilling at the same time?”


5. Insight

The bottleneck was expert attention, and the agents work on the steps around it. The slow part of quality and warranty work was reviewing issues, validating data, and aligning on next steps, with a small group of experts interpreting data for many teams. The story says AGCO is using agents to read, validate, and advance quality issues, and the stated aim is for agents to bring context from design and supply chain systems into the process. The story frames the aim as applying expertise “where it’s needed,” not replacing it. It does not describe what the experts now do differently.

Intake was split by what each side knows. Teams bring the pain point. AI leaders, AI champions, and Microsoft specialists shape the agent or workflow. This lets a quality engineer or agronomist contribute a problem without learning to build agents, and lets a small AI team decide how each problem is solved.

Governance became a funnel, not a gate. Instead of blocking employee experiments, AGCO gave about 900 volunteers access and training, then moved agents along a path: personal and team agents, cohorts that consolidate and optimize, and enterprise agents in production. The story’s phrase is that the program creates “a path from employee experimentation to governed, business-ready deployment.”

The platform followed where people already worked. AGCO picked Copilot Studio because its staff already used Word, Excel, PowerPoint, OneDrive, and SharePoint. For a citizen-maker program, the documents and permissions an agent needs are already in those places.

Limits. This is a single vendor story. The headline time saving covers “some quality reviews,” with no count, review type, or measurement date. Parts of the quality work are described as aims or expectations (“aims to,” “is expected to,” “will accelerate”), while the use of agents on quality issues is stated as current. Agent counts are approximate (“hundreds,” “several hundred”) and the story does not say how many of the roughly 2,000 makers have agents in production. No cost, headcount, warranty, or defect figures are published.

One-line takeaway: AGCO’s case shows a manufacturer turning employee curiosity about AI into a governed pipeline, where the people who feel the friction name it, a small AI team shapes the agent, and the first visible payoff is in expert-bound quality reviews.


6. Implementation hypotheses (not in the source)

6.1 Quality issues read, validated, and routed with design and supply chain data

Basis: The stated aim is for agents to “connect data across design and supply chain systems, and move issues forward more quickly”; issues “are read, validated, and quickly advanced to the next step”; the quality framework “is expected to include a series of agents rather than a single bot.”

If the planned series of agents follows the three verbs, it would likely split along them. A first agent likely reads incoming quality and warranty reports, which arrive as free text, emails, and messages because “much of the communication happens outside structured workflows,” and pulls out the machine model, serial or build, part, and symptom. A second agent would then validate the report against design data, such as the bill of materials and any running change already applied to that build, and against supply chain data, such as the supplier and lot of the part. A third would advance the issue: assign it to the next owner, attach the context gathered so far, and post it where the team works, likely in Teams. The experts would most likely still make the decision on root cause and corrective action, but would start from a validated, assembled case instead of collecting the data across teams themselves. The “about an hour” figure probably reflects reviews where this assembly was the slow part.

6.2 Cohorts that move citizen agents into managed production

Basis: “hundreds of citizen-built agents and shared team agents”; “several hundred more moving through cohorts designed to consolidate and optimize them for enterprise use”; Drawdy’s “It’s not shadow IT—it’s support.”

On Copilot Studio, the path from a maker’s agent to an enterprise agent is likely built with separate environments and data policies: makers build in a personal or team environment with a restricted set of connectors, and agents that prove useful are nominated into a cohort. A cohort would group agents that do similar jobs, for example several team agents answering questions from the same policy documents, and merge them into one maintained agent with a named owner, approved data sources, and access matching the original documents’ permissions. The AI leaders would then move the merged agent into a managed production environment. This would explain why the story counts enterprise agents in production separately from the hundreds of citizen and team agents, and why “several hundred more” are waiting in cohorts rather than going straight to production.

6.3 Knowledge agents grounded in named sources

Basis: The agronomy agent “combines AGCO agronomy studies with sales guidance and web content”; the supplier agent “reads contracts.”

Both agents are likely retrieval agents: Copilot Studio agents with document collections as knowledge sources, probably stored in SharePoint, where the files already live. The agronomy agent would take a salesperson’s description of a field problem, find the relevant agronomy study, and match it to machines through the sales guidance, with web content filling gaps. Answers would likely cite the study or guidance document so the salesperson can check before passing it on. The supplier agent would read contracts from a procurement library and list clauses or terms worth reviewing for value or savings, with a procurement owner deciding what to act on.


7. Sources

  1. Microsoft Customer Stories — AGCO scales employee-built AI agents with Microsoft Copilot Studio
    https://www.microsoft.com/en/customers/story/26836-agco-corp-word
    2026-07-06 (accessed 2026-11-17)