Costa Rica’s largest dairy cooperative, Dos Pinos, has deployed roughly 80 autonomous AI agents using Microsoft Copilot Studio to automate a range of critical business functions—from packaging review to legal drafting and IT service management. The details emerged in a Microsoft customer story published on May 14, 2026, and they offer one of the most vivid examples yet of enterprise-scale AI agent adoption in a heavily regulated food-production environment.

Dos Pinos, which supplies over 900 products to markets across Central America and the Caribbean, built the agents on Microsoft’s low-code platform to tackle repetitive, error-prone work that traditionally required manual oversight. While the cooperative has not disclosed the full inventory of its agent workforce, the published account specifically calls out four domains: packaging review, legal drafting, risk analysis, and IT support.

Packaging review: stopping label mistakes before they reach consumers

Food labelling is a minefield of regional regulations, ingredient declarations, and nutritional formatting. A single misprinted allergen warning or an incorrect net-weight figure can trigger costly recalls and regulatory fines. Dos Pinos’s packaging-review agents continuously scan artwork files, proofread text against approved regulatory databases, and flag discrepancies—misspelled ingredients, incorrect country-of-origin statements, or outdated barcode formats—before the label goes to print.

The agents are built on Microsoft’s multi-modal capabilities, meaning they can inspect both the text layer of a PDF and the rendered image of a label, catching layout misalignments that a simple text parser might miss. According to Dos Pinos IT leaders cited in the story, the agents have already prevented several high-risk labelling errors that would have otherwise gone unnoticed until post-production inspection.

Dairy cooperatives juggle an enormous volume of contracts—supplier agreements, distribution deals, co-manufacturing addendums—and each one requires precise language that aligns with Costa Rican and international trade law. Dos Pinos assigned a subset of its Copilot agents to legal drafting. The agents ingest existing templates, learn the cooperative’s preferred clause structures, and then generate first-pass drafts for new agreements.

Legal teams review and approve the output, but the upfront time saved is substantial. The agents also monitor regulatory changes published in government gazettes and automatically suggest clause updates when, for example, a Central American tariff schedule is revised. This proactive posture, Dos Pinos says, has cut the average contract turnaround by more than half.

Risk and compliance: turning raw data into actionable alerts

Dairy production involves complex supply chains, cold-chain logistics, and strict sanitary standards. The “risk work” agents described in the Microsoft story pull telemetry from IoT sensors on storage tanks and delivery trucks, cross-reference that data with weather forecasts and shipping schedules, and highlight potential spoilage windows. If a refrigeration unit on a delivery route trends toward failure, the agent can pre-emptively reroute the shipment and notify the receiving warehouse.

On the compliance side, agents simulate audit trails against HAACP and ISO 22000 standards, ensuring that documentation is complete before a real auditor ever books a visit. The cooperative’s risk-and-compliance head told Microsoft that the agents have transformed a reactive, paper-heavy process into a real-time monitoring operation that runs 24/7 without adding headcount.

IT service: a self-healing helpdesk

Dos Pinos uses Copilot Studio to power an internal IT service agent that handles password resets, software license assignments, and basic troubleshooting for its 5,000-plus employees. Because the agent is integrated with Microsoft 365, it can understand the user’s role, device, and recent activity, offering personalised remediation steps. Escalations to human technicians now come with a full diagnostic summary, slicing mean-time-to-resolution.

One notable detail from the customer story is that the IT service agent also monitors the health of the other 79 agents. It tracks their token consumption, run frequency, and error logs, alerting the AI ops team if an agent starts producing low-confidence outputs or exceeds its allotted compute budget. This meta-governance layer is what makes a deployment of 80 agents operationally sustainable rather than chaotic.

How it works under the hood: Copilot Studio and the Azure backbone

All 80 agents are built with Copilot Studio, Microsoft’s low-code tool for creating and orchestrating autonomous AI assistants. The platform offers a drag-and-drop canvas for defining workflows, natural-language triggers, and knowledge sources—SharePoint libraries, Dataverse tables, Dynamics 365 modules, and third-party APIs. Dos Pinos connected its agents to the cooperative’s existing SAP ERP system and its legacy on-premises label-design software via custom connectors built in Azure Logic Apps.

Each agent runs in the Microsoft 365 security envelope, inheriting the same identity-based access controls that govern human employees. That architecture matters enormously for an enterprise in a regulated sector: it means governance, compliance, and audit capabilities are applied uniformly. Administrators can view a dashboard of every agent’s permissions, the data sources it touches, and the decisions it has made.

Microsoft provides a set of responsible-AI tools that Dos Pinos activated for this deployment. The system detects and mitigates groundedness issues—hallucinations where an agent invents an ingredient name or a legal clause—by comparing outputs against verified knowledge bases and flagging low-confidence results for human review. Additionally, content safety filters block harmful or off-topic outputs, which is critical when agents interact with customer-facing systems.

The business case: faster, safer, and more predictable

Dos Pinos has not released exact return-on-investment figures, but the cooperative’s chief digital officer indicated in the Microsoft story that the agents collectively eliminate roughly 30,000 hours of manual, high-focus work annually. Those hours are being reallocated to strategic initiatives—product innovation, sustainability programmes, and market expansion—rather than simply cut.

More importantly, the risk-reduction value is substantial. A single labelling recall can cost a dairy processor millions of dollars in lost inventory, freight, and brand damage. By catching errors before labels are printed, the packaging agents directly protect revenue. Similarly, the compliance agents reduce the probability of a failed food-safety audit, which could jeopardise export licenses.

Lessons for enterprises: governance is the real breakthrough

While the headline figure of 80 agents grabs attention, the deeper story is how Dos Pinos manages them. The cooperative established an AI Centre of Excellence before scaling beyond a handful of pilot agents. This team defines the taxonomy of agents, sets confidence-score thresholds for automated decisions, and maintains a “human-in-the-loop” policy for all legally binding outputs.

Microsoft highlights three governance practices from the Dos Pinos deployment that it now recommends to other large-scale adopters:

  • Identity-centric access: Every agent authenticates with its own Entra ID service principal, ensuring that data access can be audited per agent, not just per user.
  • Lifecycle management: Agents are version-controlled in Azure DevOps, promoted from development to production through gated pipelines, and automatically deprecated when their business case expires.
  • Continuous monitoring: The IT service agent’s meta-monitoring function is itself monitored by a human operations team, creating a layered oversight model.

These practices address the number-one concern that CIOs voice about AI agents: that autonomous, low-code-built assistants will proliferate without visibility, creating a “shadow AI” problem that is even harder to control than shadow IT.

A turning point for regulated industries

Food and beverage companies have historically been slow to adopt bleeding-edge automation because of the perceived fragility of AI models in high-stakes scenarios. The Dos Pinos case suggests that era is ending. With the right platform and governance framework, a mid-sized cooperative in Latin America can orchestrate dozens of agents across functions that touch product safety, legal risk, and corporate IT—and do so in a way that satisfies both internal auditors and external regulators.

Other dairy processors and consumer-packaged-goods manufacturers will undoubtedly study this deployment as a template. Microsoft’s decision to publish the story signals its confidence that Copilot Studio is ready for production workloads far beyond office productivity. The company has also announced plans to release an industry-specific agent template pack for food manufacturing in the second half of 2026, likely inspired by early adopters like Dos Pinos.

For IT leaders weighing their own agent strategies, the message is clear: enterprise governance is not an afterthought—it is the prerequisite that turns a handful of clever bots into a trusted digital workforce. Dos Pinos didn’t just build 80 agents; it built the scaffolding that keeps those agents safe, compliant, and aligned with business goals. That scaffolding is what will determine whether AI agents become a permanent fixture of the modern enterprise or just another overhyped experiment.