Agents move fast.
Parallel work compounds quickly. So do conflicts, stale context, and unintended changes.
Waymark · In active development
Let AI agents do the work.
Keep humans in control of what matters.
A system for governing AI coding agents: isolated workspaces, credentials held outside the agent's reach, and human approval where it counts. In active development, and used to build CADMedic.
Overview
Product Lab
One decision needs youA release is ready for review
The exact production plan and validation evidence are waiting.
Live workOpen Work
Harden release recoveryProduct API
RunningRefresh product readinessWaymark Studio
QueuedReview field sync evidenceiOS Client
QueuedNeeds you1
Review production plan
Artifact, target, checks, and rollback proof are ready.
Capacity and readinessOpen Settings
Waymark Engine Ready
Repository proof Ready
Agent harnesses Available
Work queue 3 waiting
Isolated worktreesEvery agent works in a clean, bounded workspace.
Broker-owned credentialsAgents never receive your production secrets.
Verified deliveryChecks and evidence travel with every change.
The new delivery problem
AI agents are getting faster.
Production is not getting safer.
The answer is not another agent. It is a trusted boundary between agent intent and privileged action.
Credentials change the stakes.
A productive coding agent should not need the keys to your repositories or infrastructure.
A green check is not delivery.
Safe release work needs exact artifacts, scoped authority, rollback proof, and durable evidence.
One governed path
Let agents do the work.
Let Waymark hold the boundary.
Every run follows the same visible path—from scoped intent to verified delivery.
IntentScoped
AgentsIsolated
BrokerPolicy
ChecksVerified
DeliveryControlled
Agents propose
Plan, build, test, and review in isolated workspaces.
Broker proves
Policy, credentials, commands, evidence, and recovery stay outside agent reach.
Your policy decides
Choose where work stops, where a human approves, and what can proceed automatically.
Native mission control
See the whole system—not another stream of agent output.
Studio turns issues, agent runs, checks, approvals, and releases into one operational picture on your Mac.
Explore StudioProduct Lab Engine connected
Harnesses4available
Active run1governed
Approvals1needs you
Issue assignedWork isolatedChecks runningDelivery
Clean work by construction
Managed worktrees isolate every run from your checkout and from other agents.
Policy before privilege
A native Broker evaluates every sensitive action before it can execute.
Evidence, not optimism
Validation, approvals, release identity, and outcomes remain connected.
Stop means stop
Stale proof, scope drift, unsafe commands, or cleanup failures halt the path.
Built with Waymark
CADMedic is not a demo. It is the proving ground.
I use Studio to coordinate real CADMedic backlog work, agent implementation, independent review, validation, and privileged delivery preparation. CADMedic is built and in internal testing, and it is where I find out which parts of Waymark hold up.
CADMedic Built with Waymark
EMS operations platform · in internal testingA full software delivery system
Real backlog · multiple repositories · production boundaries
Issues become agent workGovernedMultiple review layersGovernedFull validation runsGovernedBroker-owned PR and deliveryGoverned
One continuous governed pathIssue → agents → reviews → tests → PR → merge → delivery
The AI never receives the keysTokens, servers, and production authority remain behind the Broker wall.
Designed for many agents
One control plane.
Every coding agent.
Use the models and harnesses that fit the work. Waymark keeps privileged execution, policy, and evidence consistent around them.
PlannerBuilderReviewerLocal model
Waymark BrokerPolicy boundary
Scope Checks Evidence Verified delivery
01Agents propose
Specialized roles work in isolated lanes without sharing a privileged execution surface.
02Broker proves
The native boundary owns credentials, commands, policy, and durable evidence.
03Policy decides
Automate routine reversible work. Stop safely when the evidence no longer matches.
Architecture you can inspect
Trust should have a technical explanation.
Explore the module boundary, data tiers, process sandbox, RepoSeatbelt recovery, and fail-closed delivery model behind Waymark.
Agent-facingWaymark CoreIntent · plans · safe context
Compile-time boundaryPrivilegedWaymark BrokerCredentials · git · delivery
In active development
Talk with me about Waymark.
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.