The last six months I've been heads-down on agentic frameworks, multi-agent orchestration, and whatever prompt engineering tricks didn't make me roll my eyes. Reading is fine. Toy projects are fine. At some point you have to put the work in front of a real business with real customers and real consequences. I just hit that point.
I'm working with two of my oldest friends, Tia and Brad, who run a consulting firm called Dig Deep. They help smaller municipalities navigate grant funding, logistics, governance, and the absolute maze of working with state and federal agencies on critical infrastructure projects. It's the kind of work that actually matters. Water systems, roads, the stuff that breaks when nobody's paying attention. My job is to build them an agentic system that fits how they already work, not a shiny demo that looks great on LinkedIn and dies in production.
Why I'm Doing This for Free (and Why That's Not a Humblebrag)
Let me kill the nobility angle before anyone reaches for it. I'm not doing this pro bono because I'm a great guy. I'm doing it because what I'll learn is worth more than any invoice I could send. You can read every Anthropic doc and watch every LangGraph tutorial on YouTube, and you still won't know what you don't know until a real client says "that's not how our workflow actually works" and you have to rip it apart and start over.
Also, Tia and Brad have spent twenty years giving me advice, support, and the occasional reality check I badly needed. Charging them would be weird. This is closer to even.
The Real Problem Isn't the Tech, It's the Gap
Here's what nobody talks about enough: most small businesses know they need AI somewhere in their workflow, and most of them have no idea how to get there. Not because they're not smart. Tia and Brad are two of the sharpest people I know. The gap between "I use ChatGPT sometimes" and "I have a multi-agent system that handles grant research, drafts compliance docs, and routes work intelligently" is enormous. And the people who could close that gap charge enterprise rates and ship enterprise solutions that don't fit a five-person consultancy.
Running one LLM in a chat window is not the same problem as building agents that talk to each other, share state, and don't hallucinate their way into a federal compliance violation. Small businesses don't have the time, the budget, or honestly the patience to figure that out themselves. I get it. I've been at this for two years and still get humbled weekly.
What I'm Actually Going to Stress Test
This project is going to push on parts of my skill set I haven't used in a while. PM muscles, mostly. Scoping for a client who isn't technical. Translating "we spend too much time on grant research" into an actual system design. Figuring out what to automate and what to leave alone because human judgment is the whole product. That last one is where I expect to screw up first. My instinct is to automate. Their business runs on judgment calls. Those two things are going to fight, and I'll have to pick.
The fun part: I'm already eating my own dog food. This post was sketched by me and refined with AI tooling I built into my own blogging platform from scratch. The thoughts and edits are mine. The polish came from a system I built because I got tired of context-switching between five different tools just to publish one post. If I can't make my own workflow work, I have no business designing one for a client.
More to share as it matures. Right now nothing has broken yet, which probably just means I haven't pushed hard enough. Give it two weeks.