I've owned a government consulting company for 18 years now.
One of the banes of any govcon's existence is poring through all the opportunity notices coming from different sources, qualifying them, researching them, and doing the thousand steps to get to an actual RFP.
You spend hours a week qualifying each one. It's monotonous and everyone sticks it on the junior person.
Then once you qualify the opportunity, you have to research it. Who's doing the work now? How are they doing? Who else is in the space? What's the policy and legislative activity around this area? And on and on.
The faster you can get all this data together and start working your network to build a team, the better.
What I Built in 48 Hours
Using Claude Code (Anthropic's AI coding assistant) with Microsoft's Azure AI system, I built a complete agentic CRM system for my company. It has six specialized AI agents:
Qualification Agent — scores opportunities against our capabilities
Research Agent — pulls market intelligence from government and news sources
Competition and Teaming Agent — finds companies doing the work now or working in adjacent areas
Capture Coach — suggests win strategies and next steps based on playbooks
Pipeline Analyst — tracks trends and flags at-risk deals
Antagonist Agent — challenges every recommendation to prevent blind spots
Because I built with Microsoft, it came with the authentication and access controls we already had in place. I needed a database for the opportunities, so I was able to seamlessly plug in MS Dataverse.
Then I needed a way for people to interact with the system, so each agent can be called through a Teams chat, and you can email new opportunities to an inbox for initial qualification.
Total cost: ~$25/month in Azure hosting, $20 or so for Dataverse, and variable Claude API usage. In other words, it's cheap.
The Real Shift: Architecture Over Infrastructure
Here's what changed that made this possible: you don't need to build from scratch anymore.
Microsoft's Agent Services. Google's Vertex AI Agents. Anthropic's Claude API. AWS Bedrock.
These platforms give you the primitives:
Authentication and security
Hosting and scaling
Integrations with existing systems
AI model access without ML expertise
Your job is now architecture and orchestration, not infrastructure.
I didn't write database code. I didn't build authentication systems. I didn't configure servers.
I described business problems to an AI coding assistant, and it built the solutions.
You know AI matters. You just don't have time to figure it out.
That's why I built this course. In 6 weeks, you'll go from "I should really learn this" to deploying real AI tools in your business—automations, custom GPTs, even AI agents that work while you sleep.
No coding. Live sessions. Limited to 15 owners so I can actually help you get unstuck.
February cohort is open. First one sold out.
The Hard Part
I'm not saying building agents is a snap. It's much easier than it was, but it isn't off the shelf. Nor should it be. You need agents tuned to your systems and processes, not some generic setup.
The hard part was testing, adjusting, and iterating. That’s still ongoing and will be as long as we use the system. You never build everything perfectly the first time and you learn as you use the system..
Is the opportunity filter too restrictive or permissive? Are there variables we missed? Is the research overkill, or is it costing us too much? Where can we plug in a cheaper model to reduce token costs without losing quality?
These are the questions you need to answer. Then as you use the system, you need to create a feedback loop. What worked and what didn't? How do we upgrade the agents based on that?
What This Means for Me
The barrier to building intelligent AI and automation just dropped, but it didn't drop to zero.
The more experience you have with agents, the better you get at building them. You have to understand where they work well and where they have limits. You need to constantly test each model to understand what it's best at and make cost/quality trade-offs. Things break, and you can grind your business to a halt by over-relying on flaky MCP connectors.
Over that same week while I was building the agentic CRM, three different business owners asked if I'd help them build agents. Initially I balked. I just stepped down from running a consulting firm. Why would I start another?
It's messy (difficult clients, HR issues, etc.), and I've worked mostly with small to mid-market companies. Could they even pay for this?
But as I’ve always said, the first time a potential client asks me for something new I say, “No.” The second time I say, “We’ll think about it,” and the third time I start a business.
I've stuck mostly to teaching owners how to do this themselves, but in the end that's not going to be enough. All owners need to know this technology but they are too busy to manage it all themselves. They can't stay on top of how fast the space is changing. They can't go deep enough.
So I'm considering having OwnerRx start a consulting practice in addition to our training classes for companies who want to make the AI agent transition. I'd love to get your feedback on the idea. I know it's needed, but will owners pay for it? What are their expectations? Are companies ready to commit to an iterative process that requires transforming how they work?
These are the questions I need answers to and the ai agents can’t do it all for me. Email me at [email protected] if you have any thoughts.
That’s it for this week,
Alan