Calling 911 Isn't Always the Solution

Christopher “CP” Richardson is the AI Insights and Engagement Lead at the Project Management Institute and Secretary of the Board of Directors for PMI Agile Alliance. A former firefighter and first responder, Richardson moved into information science, earned a master’s degree in project management, and built a career spanning organizations including Capital One, NPR, and The Washington Post. He is also the co-founder and head of operations at Vin Noir Explorers & Importers, North Carolina’s only minority-owned wine importer and distributor. Connect with CP on LinkedIn.

In this episode of Building For Others, CP joined Sachin Kamdar and Doyle Irvin to discuss what it means to build AI in service of other people. The conversation moved from a proactive community-paramedicine concept inspired by his firefighting experience to AI governance at PMI, lessons from fraud detection at Capital One, and the autonomous sales agent he built for his family’s wine business. Across those examples, CP returned to a consistent idea: AI is most useful when it understands the operating context, takes care of coordination work, and gives people more room to focus on one another.

A career built around giving other people a chance

Richardson’s work across Capital One, NPR, The Washington Post, public safety, and now the Project Management Institute shares a consistent purpose: helping someone else get a leg up.

Early in his career, someone looked past the fact that he was answering phones during the 2009 financial crisis, recognized his computer science background, and gave him an opportunity. That experience shaped how he approaches technology today. Whether he is helping a global community of project managers understand AI or exploring better ways to deliver public services, he begins with the outcome for the person being served.

AI can coordinate help before an emergency begins

Would you call 911 if you missed an appointment? Many people don’t have any other option. During his four years as a firefighter, Richardson saw people call 911 for problems ranging from heart attacks to missed dialysis appointments and difficulty managing medication.

In many communities 911 becomes the default because people do not have a clear route to other services. Richardson emphasizes that proactive AI can detect patterns in non-emergency calls, identify communities that need additional support, and connect residents with transportation, medication assistance, or community paramedics before another emergency call becomes necessary.

The agent would handle the coordination work so firefighters, paramedics, and care providers could spend more of their time helping people face to face with the critical problems appropriate for their skillsets and tools.

Project managers are becoming AI’s coordination layer

As AI changes how projects are planned and executed, Richardson is hearing project managers ask two different questions.

First, how can AI help them personally manage calendars, priorities, stakeholders, and administrative work?

Second, how can AI improve the performance of an entire project or portfolio across cost, quality, and speed?

PMI is helping its community move from individual productivity experiments toward frameworks for managing AI across projects, programs, and portfolios. Richardson himself thinks project managers are immensely well positioned to lead that the AI transition, because their work already involves coordinating stakeholders, navigating stage gates, and turning plans into actual outcomes.

Good governance means faster organizations with more effective experiments

Governance becomes a bottleneck when builders create an AI system in isolation and ask risk, legal, and compliance for approval only after the work is finished.

Richardson argues that those teams should be involved early enough to understand the system and shape it as it develops. Shared context makes reviews faster and gives employees clear boundaries within which they can build. Done well, governance creates an approved path for experimentation, the kind of experimentation that might lead to career growth as people stretch their former boundaries.

The two-person wine company that built its own sales agent

Richardson and his wife run Vin Noir Explorers & Importers, a wine importing and wholesaling business that handles everything from compliance and logistics to marketing and sales. To make statewide outreach manageable, Richardson built an agent that reviews restaurant and retailer websites, scans wine menus, identifies gaps, and matches those gaps against a catalog of roughly 250 SKUs. It also finds relevant owners or beverage managers, drafts tailored outreach, and gives Richardson the account details he needs for an informed in-person follow-up. The result is cold outreach that reflects genuine research rather than a generic sales template.

CP’s advice for builders

Keep a place for the ideas that sound a little crazy

Richardson’s closing advice to builders was simple: try the idea. He keeps a running “crazy idea list” in Codex, adds thoughts as they occur to him, and returns to them later to decide which are worth pursuing.

AI has dramatically reduced the effort required to test a concept and get something concrete back. Not every experiment will become a production system, but builders no longer need to settle every question before making the first version.

Bias can enter through the system surrounding the model

Even when an instruction does not explicitly favor a particular result, the data and choices around an agent can shape its recommendations.

Richardson’s outreach agent selects wines from a portfolio chosen by him and his wife, so their preferences for French and Italian wines influence what the agent can recommend. The same issue appears in enterprise systems through source data, available tools, reviewer expertise, permissions, and business rules. Evaluating an agent therefore requires looking beyond its prompt and model to the entire ecosystem in which it operates.

The most valuable context often comes from experience

The wine agent’s effectiveness depends on details that a general model would not know on its own. It prioritizes businesses the company can realistically serve, accounts for licensing information, avoids contacting restaurants during the rush before dinner service, and recognizes that Monday or Tuesday morning may be a better time to reach an owner. Because the company operates in the Outer Banks, the workflow also considers hurricanes, fires, and other disruptions before sending a sales email. These operational lessons are what turn a capable model into a useful business system.

“It works” is just the start of an AI project

Richardson connected today’s agent-building boom with his earlier fraud-model work at Capital One. Those fraud systems used 34 individual features, and new transaction types or inputs had to be weighed, scored, and evaluated.

Agents require the same ongoing attention. A model change, new integration, altered data source, or additional feature can produce different behavior even when the surrounding workflow appears unchanged. Production agents need clear ownership, regular evaluations, and a process for reassessing performance before customers discover a problem.