AI does more.
People stay accountable.
Our engineering principles for useful automation, reviewable decisions, and clear boundaries.
OUR POINT OF VIEW
AI-first means
workflow-first.
Start with the work: what context is needed, which actions are allowed, how quality is measured, and who owns the outcome. Then use AI to connect the stages and reduce repetitive effort.
For our in-development products, the principles below describe intended architecture. They are a basis for discussion and evaluation, not security certifications or claims of independently audited controls.
Context before action
Use relevant repository, service, and operational context. Distinguish supported evidence from inference and make uncertainty visible.
Scope the tools
Separate read access from write authority. Define the allowed environment, inputs, and actions before an automated run.
Review the consequential step
Respond requires human approval before remediation. Software Factory is intended to require human review before merge and release. Portal workflows should follow platform policy.
Evaluate against the work
Use acceptance criteria, test evidence, and failure cases that represent the real workflow. Passing an evaluation is evidence to inspect, not a blanket guarantee.
Leave an understandable record
Make a plan, its revisions, approvals, execution results, and unresolved issues available for review in the intended workflow.
BEFORE AN ENGAGEMENT
Get the boundaries
on the table.
We discuss the requirements that determine a viable design before selecting tools or making deployment commitments.
- What data can be accessed, and what must remain isolated?
- Which model providers and deployment options are acceptable?
- Who authorizes write actions, merge, and release?
- What evidence must accompany a change?
- What retention, confidentiality, and security requirements apply?