According to the latest Microsoft Work Trend Index Annual Report, 46% of leaders say their companies are already using agents to fully automate workflows or processes.
At Cloudwell we’re seeing that shift happening in real time and Cloudwell’s Senior Engineer, Owen Harvey, is right at the center of it, exploring how low-code agents and Copilot Studio can solve real-world challenges and reshape everyday workflows.
In this edition of Cloudwell Conversations, Owen shares his own experience of building an HR Agent, explains how to spot when a low-code agent has reached its limit, shares common questions customers are asking about agents, and takes a practical and candid look at how AI-driven automation is reshaping the modern workplace.
Kelvin Helmholtz (KH): You’ve been working on an HR Agent. What inspired you to build it, and what business challenge were you aiming to solve with it?
Owen Harvey (OH): When I first started exploring Microsoft Copilot Studio, I was looking for a productive use case that would drive employee engagement for further interaction and testing. I wanted to get familiar with building agents by working on something that would see real-world use.
Over the years, I’ve spent a lot of time helping with, and answering questions around, HR policies. It became clear that the most impactful use case for our first internal agent would be Human Resources. I saw an opportunity to streamline the question/answer loop with a conversational agent that could handle FAQs, initiate PTO requests, and even auto-update our Out of Office calendar.
KH: How do you identify where a low-code agent can make the biggest impact within an organization?
OH: The most important first step for identifying where low-code agents can make the biggest impact is discovery. Look at your company’s current workflows and procedures. Identify procedural pain points, likely around high overhead, low effort “busy work” for lack of a better term.
The sweet spot is where interactions are frequent, policy-driven, and grounded in structured data. IT support, HR, and general repetitive business operations are all great places to start. These are areas where automation can return hours to the business without needing deep custom development.
You also want to keep in mind that if a task consistently requires human intervention, it’s probably not a great fit for low-code unless the workflow can be redesigned first.
KH: When does a low-code agent reach its limits, and how do organizations decide it’s time to bring in external expertise?
OH: As soon as you hit complex workflows, custom security models, or deep integrations with legacy systems you start to feel the edges of the limits for low-code solutions. That’s when organizations should bring in external expertise to help architect and likely build a more complex solution.
At Cloudwell, we help clients navigate these boundaries. We know when to extend with the Microsoft 365 Agents SDK, when to orchestrate with Azure Functions, and when to ground responses using enterprise data via Microsoft Graph.
We also help with Responsible AI practices, ensuring agents behave predictably and securely across environments.
KH: What higher-level features or integrations in Copilot Studio have you found most powerful as you’ve continued developing the HR Agent?
OH: For the HR Agent, some of the most powerful features came from Copilot Studio’s extensibility. Actions and connectors let us integrate with our internal payroll, work/life sites, calendar systems, and org directory.
We used Microsoft Graph to pull reporting lines and skills data, and SharePoint to manage policy content. The agent could not only answer questions but also perform tasks, like automating the submission of a PTO request, kicking off an approval flow, notification framework, and calendar/out of office notice automations, all without our employees leaving Teams.
Leveraging tools and response flows helped to bring the agent into the realm of doing rather than simply responding. Additionally, the built in telemetry and feedback loops in Copilot Studio helped to refine prompts and improve accuracy over time.
KH: What are the most common questions Cloudwell customers are asking about Copilot Studio and agents right now, and what patterns are you noticing in those conversations?
OH: Customers want to know how Copilot Studio fits with their existing systems, how secure it is, and whether it’s ready for enterprise scale. Emerging interest in HR and IT use cases, curiosity about grounding and hallucination risks, and the importance of data and usage governance are all at the forefront of discussions.
Many are surprised to learn that Microsoft Purview can be used to govern agent content and interactions, and that Power Platform DLP policies apply here too. With Entra ID, role-based access control, and audit logging all a part of the equation, security is baked in.
KH: Agents have been described as the ‘new apps’. Do you think this is the case and what does that mean practically?
OH: Practically, that means we’re shifting from static interfaces to dynamic conversations. Instead of building a form or dashboard, we’re designing a dialogue. That changes how we think about UX, content governance, and even compliance.
Agents need to be accurate, grounded, and aligned with company policy. They also need to be versioned, monitored, and continuously improved. It’s not a “build and forget” model, it’s an operational capability.
And that’s where Cloudwell’s managed services come in: we help clients run these agents like products, not projects.
KH: How should organizations measure the value of AI agents? What metrics or outcomes really tell the story of success?
OH: Measuring the value of AI agents is critical, but sometimes, can be quite difficult. Generally, we want to look at deflection rate, time-to-resolution, SLA adherence, and employee satisfaction.
For the HR Agent, we tracked how many interactions were resolved without human intervention, how quickly requests were processed, and how employees rated the experience. We also monitored grounding accuracy, like how often the agent pulled the right answer from the right source. Adoption metrics like monthly active users, and the trends around usage, helped to determine what changes we made had the most impact.
Metrics, when done correctly, can tell us whether the agent is helping people and saving time.
KH: You’ve been with Cloudwell for more than 11 years. How has your role evolved during that time, and what changes have you seen in how clients approach technology and workplace transformation?
OH: Over the past 11 years, my role has evolved from traditional infrastructure and cloud architecture buildouts, to including AI strategy into business transformation.
What’s changed most is how clients approach architecture for addressing business needs. It used to be about what tools to use, but now it’s about how best to reach the desired outcomes.
Clients want to know how technology will improve employee experience, reduce costs, and drive agility. They’re more open to experimentation, but they also expect results. That’s why we focus on time-to-value: we help clients move fast, but with purpose. We don’t just deploy, we design, integrate, secure, and support.
KH: Looking ahead, what excites you most about the next stage of Copilot Studio’s evolution, and where do you see Cloudwell adding the most value for customers as the platform matures?
OH: Looking ahead, Copilot Studio is entering an exciting phase. Microsoft is investing heavily in Copilot extensibility and platforms like Azure AI Foundry and Copilot in Azure are opening new doors for enterprise-grade orchestration and retrieval.
We expect tighter integration with Dynamics 365, more advanced grounding capabilities, and richer plugin ecosystems. Cloudwell will continue to add value by bridging strategy and execution, helping clients identify high-impact use cases, design secure architectures, and deliver measurable outcomes.
We’ll also support change management, training, and adoption to ensure agents don’t just launch; they thrive.
Could your organization use support building or scaling low-code agents?
Reach out to the Cloudwell team and let’s get started.