Agent failure was not a Git change
An agent workflow stopped working, but recent repository changes did not explain the failure.
- Looked like
- Code regression
- Actually
- Outdated toolchain version
- Layer
- Agent runtime / Toolchain
Systems debugging for production reality.
I help AI, IoT, and edge teams debug failures that sit between application code, runtime dependencies, networks, devices, and customer environments.
Failure boundary: app code / runtime / toolchain / network / device / customer site
Premise
Production failures rarely stay in one layer.
Recent code changes are not always the cause. Tunnel instability may be a LAN problem. A database startup failure may be a runtime dependency problem. An agent failure may be a toolchain compatibility problem. The work starts by finding the boundary everyone assumed was already known.
Work
How I usually help
Most work starts with a focused diagnostic: define the failure boundary, test the highest-signal hypotheses, and turn the result into something the team can reuse.
Case Notes
Anonymized failure patterns
These notes are anonymized, but the failure patterns are real: toolchain mismatch, replaced runtime dependency, LAN-level IP conflict, and SSH ACL rollout mismatch.
Database startup failure was not database configuration
A customer-site MySQL service failed to start, and normal database-level checks did not explain the issue.
- Looked like
- MySQL configuration issue
- Actually
- Replaced shared library
- Layer
- Linux runtime dependencies
Intermittent connectivity was not tunnel instability
An edge device appeared unstable through remote access tooling, but the actual failure was inside the local network.
- Looked like
- VPN / tunnel instability
- Actually
- LAN IP conflict
- Layer
- Network / Customer site
Tailscale SSH failure was not node reachability
An edge node was reachable over the tailnet, but Tailscale SSH was denied by a separate SSH ACL path.
- Looked like
- SSH or network reachability failure
- Actually
- Tailscale SSH ACL mismatch
- Layer
- Tailnet access policy
Debugging surface
Where assumptions usually break
App code / Runtime / Toolchain / Network / Device / Customer site
Contact
Have a production issue that does not fit the usual checklist?
Send the symptom, environment, and what has already been tried.