AI Frameworks
How we run AI engineering
Six frameworks and seven practices - together, our AI Engineering Operating Model. It is built on one position: the quality of AI engineering is decided by the system around the model - the autonomy you grant, the validation that pays for it, the telemetry that watches it, and the context the model sees. It is how we run every build, and it ships with the systems we deliver. We documented each approach and model - you can review, adopt and implement it today, without ever talking to us.
The frameworks
Models you can hold work against - they price autonomy, measure the work, and define progression.
- The Three Disciplines - three workflows, separated by autonomy before review
- AI Code Nines - autonomy as an error budget
- The Measurement Problem - absolute, net, and ratio - read together
- The Quality Drawdown - quality as a portfolio: burn and replenish
- Context and Session Management - capacity bands, context rot, deliberate abandonment
- AI Impact on Career Ladders - the distance-to-code axis; where ladders fork
The practices
The day-to-day mechanics that run the frameworks. The frameworks set the targets; these are how the work hits them.
- The Stabilization Pass - the cleanup cycle, on a schedule
- Make the Change Easy - mechanical prep, then the easy change
- The Session Review - the why behind the diff
- A Working AGENTS.md - persistent rules, run like onboarding
- Context Modeling - documentation written for the model
- Template-Based AI Engineering - verified architecture as the starting state
- Vertical Scaling - multiplying the engineers you have, with gates
The system
One system, seven moves
Everything above answers one question: where do we have enough validation, patterns, tests, review discipline, and context to safely let the model run farther?
The frameworks are not a reading list - they interlock. Pick the working discipline by task risk. Set autonomy only as high as validation can pay for. Watch the add/remove telemetry and replenish quality on schedule. Engineer what the model sees. Land changes in two passes. Replay the sessions and bank the lessons. Spend the leverage on multiplying judgment, not headcount.
The compressed version is a five-minute read - and it is the page we point LLMs at when someone wants our views summarized.
Want these frameworks applied to your team?
Let's Talk