The Team They Never Hired

The Gap Nobody Closes

A medical practice in Florida needs a patient portal that handles PHI, manages appointments, and doesn't cost $40,000 in enterprise HIPAA add-ons. A PEO in Texas needs to automate employee onboarding across 40+ state W-4 forms without buying Workday. A claims management firm needs lead capture, CRM, and document automation without Salesforce.

These are not AI problems. They are business problems that AI makes solvable by a team of one or two people instead of a team of fifteen. But that team of one or two needs to understand the industry, not just the technology. The models are commodities. The context is the moat.

Stripe built Radar to catch fraud across millions of transactions. LinkedIn routes payments via ML to pick the best gateway per transaction. DoorDash upgraded a heuristic to ML and saved thousands of canceled orders. These are the case studies that prove the pattern. But none of those companies built their AI teams by hiring data scientists off a job board. They built them by embedding engineers who understood the business deeply enough to know which problems were worth solving.

The Play

Here's the model that works. I've been running it for two years through Prospyr 305, the development arm that grew out of All Lines Business Solutions.

1. Pick one industry. Not “SMBs.” Not “healthcare.” Something specific enough that you can learn the vocabulary, the compliance landscape, the workflow pain points, and the software they already hate. Medical practices. PEOs. Auto body shops. Claims management. The narrower the better. You're not competing with McKinsey, you're competing with the guy who set up their QuickBooks in 2014 and never called back.
2. Learn how it actually runs. Not from a textbook. From the owner. Sit with them. Watch what they do for a week. The first thing you'll notice is that the real workflow looks nothing like the documented workflow. The second thing you'll notice is that 60% of their day is automatable with tools that exist today.
3. Build the AI layer. Not a chatbot. Not a “copilot.” Actual systems that execute work: intake forms that create leads automatically, scheduling that writes to two Outlook calendars simultaneously, document pipelines that extract data and generate compliance PDFs, agents that monitor and escalate. The technology is solved. The integration is not.
4. Price for the upside. This is the part nobody talks about. A retainer caps your earnings at the hours you bill. But if you're lifting a client's margins by 25% through automation, the value you create dwarfs the retainer. For startups, take equity. For established businesses, structure performance milestones where you get a bonus tied to the margin improvement you can measure. Don't leave the upside on the table, but don't walk into a doctor's office and ask for ownership. Read the room.

What This Looks Like in Practice

All Lines Business Solutions is the parent. Accountancy, business consulting, administrative services. The work that pays the bills and teaches you how businesses actually run. Prospyr 305 is what grew out of it: the development and AI infrastructure arm that builds the systems ALBS clients need but no off-the-shelf software provides.

Through ALBS, I sit with practice owners and learn their workflows. Through Prospyr 305, I build the systems that replace them. A medical practice gets a HIPAA-ready patient portal. A PEO gets an onboarding platform that generates state-specific W-4 PDFs. A claims firm gets lead capture, CRM, and document automation. Production software, deployed, integrated with their existing tools, running 24/7.

Behind the scenes, Prospyr 305 runs on four AI agents of its own: one handles strategy and client relationships, one orchestrates operations, one runs research and infrastructure, one manages daily execution. They coordinate via a shared memory system, maintain an Obsidian knowledge graph, and produce competitive intelligence overnight. The agents aren't a product we sell. They're how we deliver the work.

Why This Works Now

Three things changed in the last 18 months that make this model viable for anyone willing to do the work.

• Models are good enough. GLM-5.2 hits 280 tokens per second on optimized inference. DeepSeek R1 reasons through complex problems. Qwen writes production code. You don't need GPT-4 at $20/1M tokens. You need a local model on a $20/month box that handles 80% of the work and a fallback to a frontier model for the 20% that matters.
• The tooling is free. OpenClaw, Ollama, Supabase, Vercel, Next.js. The entire stack to build, deploy, and operate AI agents costs less than a phone bill. The barrier isn't technology access. It's knowing what to build and which industry to build it for.
• The market is desperate. Small businesses are drowning. The good ones have more work than they can handle. The bad ones can't afford help. Both need the same thing: systems that run the busywork so the owner can do the work that actually generates revenue.

The Compounding Effect

Here's what happens when you do this long enough: the work compounds. Each client teaches you the industry better. Each system you build becomes a template for the next. Each workflow you automate generates data that makes the next automation smarter. After two years, you're not building from scratch. You're adapting patterns from a library of solved problems.

That library is the real asset. Not the code, not the clients, not the retainer revenue. The pattern library. The accumulated knowledge of what works in medical practices, what fails in PEOs, what matters in claims management, what nobody in accounting has figured out yet. That's the moat. And it only deepens with every engagement.

The VCs funding agent startups at Sequoia, a16z, and Kleiner Perkins are betting on companies that build AI for everyone. That's a hard business. The easier business, the one that compounds faster, is building AI for someone specific. One industry. Deep context. Real problems. Measurable outcomes.

You don't need a CS background. You don't need funding. You need a domain expert who knows how the business works, and you need the patience to build the system that replaces the one they hate.

The models are commodities. The context is the moat. The industry knowledge is the product. Everything else is infrastructure.