AI writes the code.
I deliver the software.
I architect systems, write the specs, and direct teams of AI coding agents through every build. The result: production software with real users, running real businesses today.
18+ mo
in production
3 yrs
directing AI agents
24/7
systems running
100%
agent-built code
Selected Work
Production systems for law firms, operations teams, and my own infrastructure — each one specified by hand and built by directed AI agents.
CHP Report Automation
PI law firms need traffic collision reports, and chasing them wastes staff hours every week. This system runs the entire process: monitors email intake, identifies missing data, calls CHP offices with voice AI, and works the government portal end to end. Specified, directed, and maintained in production for over 18 months.
0 firms · 0 users · 0+ requests
Runs on: Next.js · Convex · Playwright · VAPI · Twilio
→ Read case study
Knowledge Extraction Engine
Turns an expert's operating knowledge into structured rules a machine can act on. AI conducts the interviews; the extraction pipeline and human review workflow are mine. 279 rules captured from a single domain expert so far.
0 rules · 0% coverage · 0 categories
Runs on: Next.js · Convex · Claude API
→ Read case study
Multi-Agent Case Analysis
Multiple AI agents analyze different parts of a legal case in parallel, with results aggregated and ranked by severity. Routing logic and aggregation architecture fully specified; currently in detailed design.
0x throughput · HIPAA-compliant · hybrid cloud+edge
Runs on: Next.js · Convex · Claude API · Filevine API
Always-On Agent System
The infrastructure behind everything else. A coordinator agent on dedicated hardware delegates to specialist agents for coding, research, and analysis — running around the clock across Discord, Telegram, WhatsApp, and iMessage.
Runs daily · multi-channel · persistent memory
Runs on: Mac Mini M4 · Multi-Provider AI
→ Read case study
Capabilities
The work sits at the intersection of engineering judgment and AI direction. These are the disciplines behind every project above.
Agent Orchestration
Directing fleets of AI coding agents as a coordinated team.
- Claude Code
- Codex CLI
- Cursor
- Multi-agent systems
- Model routing & selection
System Design
Turning business problems into specs agents can execute.
- Product requirements & technical specs
- Architecture planning
- Task decomposition
- Phased delivery plans
AI Operations
Keeping quality high and costs predictable at scale.
- Prompt & context engineering
- Output review & QA
- Cost optimization
- Multi-provider strategy
Domain Expertise
Deep familiarity with regulated, deadline-driven work.
- Personal injury workflow automation
- Government systems & CHP processes
- API integration
- HIPAA & legal compliance
Infrastructure
Production deployments that stay up and stay maintained.
- Vercel & Fly.io deployment
- Always-on dedicated hardware
- Scheduled automation
- Multi-channel messaging
Industries
Personal Injury Law
Thousands of decisions per week, buried in paperwork. I design AI systems that track cases, hit deadlines, and flag problems. I've built production tools for real PI firms handling real cases.
Operations & Field Sales
Reps in the field, leads going cold, managers drowning in data. I build the intelligence layer that keeps your pipeline moving. Same process: specs, agent execution, human review.
Admissions Consulting
Hundreds of students, dozens of deadlines, zero margin for error. I design systems that keep track of everything so consultants can focus on the students.
How I Work
From understanding your problem to shipping production software.
Understand
I learn your business. The decisions, the exceptions, the edge cases that only your best people know.
Spec
Detailed specs that coding agents can execute. PRDs, data models, constraints, phase-by-phase prompts.
Build
Agents write the code. I review every output and iterate until it works the way it should.
Review
QA, testing, polish. Nothing ships until it handles the edge cases your business actually hits.
Ship
Production deployment with documentation. I stick around to make sure it holds up.
About
Manager Matt is a one-person studio with an unusual build process. I write the specifications, design the architecture, and direct teams of AI coding agents — Claude Code, Codex, Cursor — through every line of the build.
The process isn't theoretical. It has shipped production software that law firms and operations teams depend on every day, including a system that has run continuously for over eighteen months.
Three years of daily work across every major model taught me what agents do well, where they fail, and how to structure a build so quality holds at speed. That judgment is the product — the agents are just very fast hands.
— Matt Sanchez, Founder