Why I built Waymark
CADMedic is a complex EMS operations platform spanning operational web software and native field apps. It is built and in internal testing. It gave me a real problem to solve: how can one developer direct more agentic work without becoming the bottleneck—or handing agents the keys to delivery?
From hand-coded to agent-orchestrated
Waymark grew out of the work required to turn raw agent output into dependable product progress.
CADMedic began as a 100% hand-coded product. I wrote it myself, drawing on my own work as a paramedic in the operational and clinical workflows it supports. That produced a strong foundation—but every new feature, fix, review, and release still passed through one pair of hands: mine.
Agentic coding made it possible to explore and implement more work. It also created a new queue: preparing context, separating work, watching runs, checking diffs, rerunning tests, resolving collisions, and moving approved changes through Git and deployment.
Waymark turns CADMedic issues into governed runs. It creates isolated workspaces, gives specialized agents bounded jobs, gathers independent review and test evidence, and asks the developer for decisions only at the boundaries they choose.
Why CADMedic is the test
A seemingly small change can touch dispatch, live unit context, mapping, clinical workflows, billing, fleet, or native field apps. Waymark keeps the issue, agent work, review evidence, and delivery decision connected across that complexity.
One governed run
CADMedic gives Studio the complexity a real AI CI/CD control plane has to handle.
CADMedic work starts as product work—not a special Waymark script. Studio syncs the issue, repository scope, risk, and required finish line.
Planning, implementation, testing, and review can be separated across agents. Each works in an isolated Broker-managed workspace.
Repository checks, secret scanning, deterministic review, disposable runners, and UI evidence test the same change from different angles.
After approval, Studio owns the commit, branch push, provider status, pull request, and merge path. The agents never receive those credentials.
CADMedic is the dogfood target for exact production identity, staging, rollback, and cutover. That final generalized path remains under active validation.
What changed for CADMedic
Waymark does not replace the judgment that built CADMedic. It removes repetitive coordination around that judgment, so more carefully bounded work can move through one governed path.
Separate agent roles can plan, implement, test, and review without sharing one fragile working copy or receiving privileged credentials.
Validation commands, independent review, and UI evidence are attached to the change before it reaches the developer.
Waymark—not the coding agent—holds the credentials and enforces the chosen approval boundary for pull requests, merges, and production.
The result is not less human control. It is more useful work reaching human judgment.
What this proves
CADMedic is how I test Studio against real product work. Repository automation is in active use on CADMedic; a generalized production cutover is the next thing to prove.
In active development
Waymark is in active development. If your organization is working out how to let AI agents do real work safely, I'd like to hear about it, whether to compare notes or to discuss Waymark itself.