Eight standards. Clear responsibilities.
Thirty years of lessons, applied to infrastructure, security, software and AI. We agree the scope, owners and any exceptions before go-live.
- 01
Plan the way back.
A change needs a tested rollback, recovery or compensating plan. Irreversible steps and their consequences are agreed before release.
Learned from Automation & DevSecOps, 2013 to 2022 - 02
Time the restore.
Representative restores are timed against recovery objectives, and recovered services are checked with the business. Test coverage and remaining gaps are recorded.
Learned from What a DR test should actually prove - 03
Name the access owner.
Access has a named owner and is granted at the narrowest point that works. Reviews remove permissions that are no longer required.
Learned from What NetWare taught us - 04
Design the boundaries.
Isolation, identity and trust belong in the design. Existing gaps are assessed and the work needed to address them is agreed.
Learned from Virtualisation & security, 2000s - 05
Define the interfaces.
Interfaces define inputs, outputs and failure behaviour. We test the agreed contract and make compatibility and version changes explicit.
Learned from Software engineering, from 2011 - 06
Keep an operator in control.
A named operator can stop new agent actions and revoke access. Logs support investigation; actions already committed may need a separate recovery plan.
Learned from Least privilege for AI agents - 07
Make evidence traceable.
Answers based on retrieved information link to supporting sources. Unsupported claims and uncertainty are made clear so a person can check the result.
Learned from Production AI, 2023 onwards - 08
Test credible failures.
Capacity and recovery tests cover agreed peak loads and credible failures. Residual risks, dependencies and operator responsibilities are documented.
Learned from Capacity planning
See all eight standards applied end to end in application modernisation.