Rolling Out Your First AI Agent

AI agents, task-specific tools capable of decision-making, automation, and interaction, are becoming essential components of modern business operations. But there’s a catch: while many organizations are easer to explore AI, very few have agents successfully deployed in production.

Key Takeaways:

Governance and process discipline are non-negotiable. Agents will punish poor documentation, weak data quality, and ad hoc workflows.

Think of AI agents as digital employees. Like humans, they need onboarding, training, and supervision to deliver consistent value.

Avoid rushing straight into instructions—design first. Visual process mapping is key to building reliable, auditable agents.

Start with low-risk, internal use cases. Early success builds trust and helps organizations develop the skills and governance needed to scale.

Why Most AI Agents Never Go Live

Gotts began the session with a striking statistic: for every 10 AI agents built, only one typically reaches production. The reason? It’s not a technology issue, it’s a confidence issue. Many organizations hesitate to deploy agents broadly because they don’t fully understand or trust how they work.

Agents can feel like magic when they generate a poem from a two-line prompt. But in business settings, where accuracy, reliability, and traceability matter, that same ambiguity becomes a liability. Gotts compared working with agents to instructing a precocious but literal-minded 12-year-old: if you’re not specific, things will go wrong in creative and unpredictable ways.

The Anatomy of an AI Agent

Across platforms, whether Salesforce, Microsoft, or custom-built tools, agents consist of two core components:

1. Instructions: What the agent is supposed to do, usually written in natural language or structured prompts.
2. Actions: The workflows, APIs, or code that enable the agent to execute those instructions.

Unfortunately, many organizations rush into instruction-writing without first clarifying the agent’s purpose. This leads to confusion, bloated logic, and inconsistent behavior.

The Role of Process Mapping in Agent Design

Gotts emphasized the importance of process diagrams as a foundational design tool. Before typing instructions, teams should visually map out what the agent needs to do step by step, including alternate paths, guardrails, and required data.

This diagram acts as:

• A communication tool for business stakeholders.
• A blueprint for agent logic.
• An audit trail for governance and compliance.
• A troubleshooting guide when something breaks.

With this process-first approach, his team has moved agents from idea to production in as little as 37 minutes, though more complex agents can take up to a week.

Think Like a Manager, Not a Coder

To demystify the development cycle, Gotts proposed an analogy: treat AI agents like employees. Just as you wouldn’t drop an intern into a customer-facing role without training, you shouldn’t deploy agents without careful onboarding.

The life cycle of an agent includes:

• Recruiting (ideating and selecting the use case),
• Onboarding (designing and testing the agent),
• Monitoring (tracking behavior and feedback),
• and Performance Management (iteration and improvement).

By framing AI this way, teams can focus on clarity, alignment, and incremental improvement rather than chasing perfection from day one.

Pick the Right First Use Cases

Another critical insight: don’t aim for a grand slam with your first agent. Many organizations get stuck trying to automate high-risk, customer-facing scenarios too early. Instead, start with internal agents that are:

• Narrow in scope
• Data-accessible
• Low-risk
• Easily monitored for feedback

Examples shared during the session included:

• Vacation request agents that connect to internal HR systems
• Lost-and-found bots for transit companies
• Appointment scheduling assistants for legal practices
• Support tools for crafting customer success emails

Notably, these agents reduce cognitive load and improve user experience even when they’re just a smarter interface for existing workflows.

Governance Isn’t Optional

AI agents amplify both strengths and weaknesses in business systems. If your organization lacks clear processes, structured metadata, and quality data governance, agents will make that painfully obvious.

Gotts shared the example of a frequently asked questions (FAQ) agent that began pulling policy answers from Canadian government websites—not because of a technical failure, but because of vague prompts and insufficient guardrails. Process diagrams helped pinpoint the issue quickly, but the experience highlighted a broader lesson: ambiguity is the enemy of reliability.

Well-governed systems, in contrast, make agents exponentially more effective. Regulated industries like banking and healthcare are at an advantage here, process rigor is already embedded.

Confidence Comes from Structure

To gain organizational buy-in, Gotts encouraged leaders to build simple, explainable agents first. Visual tools and precompiled logic make it easier to:

• Demonstrate how the agent works,
• Manage feedback loops,
• Satisfy security and compliance requirements, and
• Iterate quickly when improvements are needed.

He also reminded attendees that every agent is a chance to build “muscle memory”—the skills, documentation, and confidence that allow future agents to scale more quickly and safely.

The Future: AI Will Punish Mediocrity

The session ended with a powerful call to action: while agents are easy to build, they demand discipline. Poor workflows, bad data, and informal development practices will be exposed instantly.

“Agents punish mediocrity,” Gotts said. “If you’re good, they’ll make you better. If you’re bad, they’ll show it.”

To thrive in this new landscape, organizations need more than technical skill. They need business analysts, change leaders, and operational strategists who can guide teams through the messy work of process design, data governance, and stakeholder education.

Because in the age of AI agents, the winners won’t be the fastest coders…they’ll be the best communicators.