In Five Weeks, 67 Prumo Group Interns Built Their Own AI Agents
Client | Prumo Group (teams from Prumo and Porto do Açu) |
Participants | 67 interns across compliance, safety, communication, procurement, and HR |
Format | 5 weeks: 4 weeks of active workshops + 1 week hackathon |
Outcome | Non-technical interns building custom agents + 4 standout projects presented to the executive board |
IMAGE · OPENING
Photo of the intern cohort during the program, ideally during the project pitch showcase, or an active session with students on their laptops.
Wide horizontal format, high resolution.
1. Context
Prumo Group had been rolling out artificial intelligence at scale. Microsoft Copilot was already the enterprise standard, accessible to all employees for everyday tasks like summarizing emails, drafting reports, and organizing notes.
The next frontier was more demanding. The company aimed to deploy AI agents capable of automating complex, multi-step business logic beyond simple chat prompts. While the software licenses were in place, there was a shortage of people equipped to design automations, leaving substantial optimization potential untapped across department workflows. Interns represented the ideal catalyst: early-career professionals eager to learn and embedded directly within the departments that stood to gain the most. The challenge was their lack of technical background: to most, Copilot was merely a chat interface, and building autonomous agents from scratch was completely unfamiliar territory.
The diversity of represented departments defined the program's architecture. Rather than training a single homogenous team on a single task, the curriculum had to establish a shared foundation across highly varied operational domains. Exercises had to remain broadly accessible while immediately adaptable to unique business contexts.
2. How we did it
The program spanned five weeks: four weeks of interactive sessions followed by an intensive hackathon week dedicated to individual projects. Spreading the curriculum over weeks rather than compressing it into a bootcamp was vital for this cohort. Each interval allowed interns to assimilate concepts, experiment within their daily jobs, and return with questions born of real workplace trials. Three AutoU facilitators guided the cohort through lectures and mentoring clinics, organized through a dedicated Microsoft Teams hub.
Stage | Focus | Format |
|---|---|---|
Weeks 1 to 3 | AI fundamentals and agent construction in Copilot Studio, presenting concepts and applying them instantly. | Active workshops with real-time co-creation |
Week 4 | Each participant designs and builds an agent solving an authentic challenge in their department. | Hackathon with structured toolkits |
Week 5 | Review, interactive knowledge quiz, project pitch presentations, selection of top innovations, and awards. | Dynamic quiz and executive pitches |
Microsoft Copilot Studio was chosen as the operational platform. Because the organization already utilized Copilot at user tier, activating complimentary 30-to-60-day Copilot Studio trials provided ample runway. Building natively inside the Microsoft ecosystem was a strategic choice: Prumo had standardized on it, ensuring every skill learned was immediately operational. Training on an external tool would have taught capabilities disconnected from daily corporate reality.
The curriculum covered everything necessary to build agents from zero: generative AI fundamentals, prompt structuring, grounding agents on enterprise document bases, and configuring conversational architecture—triggers, variable capture, prompt nodes, and generative responses. Above all, one methodology governed instruction: breaking complex workflows into small, manageable stages—the indispensable prerequisite for turning automation ideas into reliable agents.
Teaching centered on active co-creation. Rather than passively viewing slides, students built alongside the instructor, executing each configuration step in their own environments. Two benchmark agents emerged from class exercises: 'Atom' (vacation scheduling and approval routing) and 'Aurora' (vendor evaluation and comparative scoring). These models provided intuitive blueprints easily understood across departments.
IMAGE · CLASSROOM CO-CREATION
Shared screen of agent configuration in Copilot Studio, or photo of an active workshop with instructors guiding students as they replicate steps on their laptops.
3. Results
Within five weeks, a cohort that began with zero agent-building awareness delivered two collaboratively built agents and dozens of autonomous individual automation projects.
Before | After |
|---|---|
• Knew Copilot only as a chatbot, never having built an autonomous agent | • Confidently construct and publish multi-step agents in Copilot Studio |
• Unaware of where automation applied within their department | • Map complex workflows and identify automation opportunities independently |
During the culminating hackathon, each participant identified a friction point in their department and engineered a custom agent. Four standout projects were selected for presentation to executive leadership:
Project / Department | What the agent accomplishes |
|---|---|
Safety Equipment Management | Automates cataloging and communication for workplace safety assets, previously handled manually. |
Brand Compliance Checker | Evaluates corporate PowerPoint slide decks against official brand and typography guidelines. |
Employee Recognition Assistant | Assesses peer nominations and verifies whether actions meet internal recognition criteria. |
Compliance Intake & Review | Analyzes incoming PDF documents, drafts compliance opinions, and routes notices, dramatically reducing operational load. |
None of these were toy prototypes. All were created by the interns themselves, addressing operations they handle daily. As an accolade, the four finalists received private executive mentoring. The program yielded an unplanned milestone: corporate enthusiasm led the interns to pitch their solutions directly to executive directors during company-wide townhalls—an extraordinary exposure opportunity for early-career professionals.
IMAGE · EXECUTIVE PRESENTATION
Photo of interns presenting their selected automation agents to company directors and leadership.
4. Key Learnings
The overarching revelation was that most business units already possess everything required for automation without realizing it. Across all mapped workflows, a familiar pattern appeared: well-defined procedures awaiting automation, repetitive clerical routines, rich data stores lacking analytical interfaces, and corporate manuals ripe for conversion into agent grounding. The raw materials were all present. The academy supplied the missing link: the acuity to spot opportunities and the hands-on craft to build working agents.
The second insight involved cultivating confidence in junior talent. Interns naturally hesitate to speak up in traditional lectures. Active learning unlocked their engagement: live quizzes doubled as instant diagnostics, highlighting knowledge gaps immediately so they could be addressed on the spot. For early-career professionals, real mastery comes from doing: testing, failing safely with expert guidance, and iterating. Placing hands on the keyboard early turned hesitation into lasting autonomy.