Cases
Decisions, constraints, and results from work inside live systems.
These cases cover delivery recovery, AI engineering control, AI product risk, and resilience in live systems. Client details stay anonymous where permission or an NDA requires it.
Selected cases
Feature Delivery Recovery
Delivery Recovery
A leadership handoff left no reliable map of the inherited platform.
- Decision
- Redesign the system boundaries instead of relying on local fixes.
- Result
- The team got a shared map of system boundaries, ownership, and the path to release.
- Role
- Chief Technology Officer
System Digital Twin
AI Engineering Control
An undocumented platform was too large to inspect manually in time.
- Decision
- Build a textual system twin with progressive context for agents.
- Result
- The team got an inspectable system map for further engineering work.
- Role
- Chief Technology Officer
Redis P0 Before First Users
AI Product Risk & Safety
A pre-launch audit examined session storage as an attack surface.
- Decision
- Remove Redis bearer tokens; check storage contents with a deterministic gate.
- Result
- Bearer tokens were removed from session storage, and every build gained a deterministic check.
- Role
- Founding AI Product Architect
Layered DDoS Defense Beyond the CDN
Resilience & Security
A live platform faced attacks that imitated application behavior.
- Decision
- Keep Cloudflare for volume; add a separate application-behavior defense layer.
- Result
- Traffic volume and application behavior were handled by separate defense layers.
- Role
- Chief Technology Officer
More cases
How the initial diagnosis changed the plan, and what adoption cost.
First Day After the CTO Exit
Delivery Recovery
A new CTO inherited production pressure and no usable handover.
- Decision
- Map incidents and ownership before promising acceleration.
- Result
- The team got a shared map of incidents, system boundaries, and ownership.
- Role
- Chief Technology Officer
Agent Library Adoption by Pull
AI Engineering Control
Formal AI enablement created compliance theatre instead of habit change.
- Decision
- Leave a useful agent library in the shared repo and let adoption happen by pull.
- Result
- Engineers began adopting the shared agent library after seeing it work in real reviews.
- Role
- Chief Technology Officer
Bonus Abuse as an Account Factory
AI Product Risk & Safety
Fraud recurred through linked account creation rather than isolated bad actors.
- Decision
- Treat the operation as a factory graph instead of a queue of separate accounts.
- Result
- The team shifted fraud investigations from individual accounts to linked operational behavior.
- Role
- Chief Technology Officer
GCP to Bare Metal: Cost and Control
Resilience & Security
Cloud cost and network-control limits became part of the same operational problem.
- Decision
- Move the platform to infrastructure where the team could own the critical network boundary.
- Result
- Infrastructure costs fell by 30%, while the team regained direct control of the network boundary.
- Role
- Chief Technology Officer
The problem behind each case.
Each problem page covers the first warning signs and where this kind of intervention fits.
Next step
Facts and context stay more useful than a polished brief.
The Facts page has the concise public record. When the problem is clear before the engagement format is, send the situation first.