TASK-FIT SELECTION
Tools were assigned where they fit: primary Sites engineering in one stream, isolated research and parallel batches in others—coordinated through briefs, handoffs and the same review gates.
LAB / BUILT WITH CODEX / FIRST-PARTY ENTRY
Codex accelerated the build. Aevrion directed the system.
A factual account of how Aevrion used Codex—an AI coding tool—inside a directed, reviewed and tested engineering process to build Aevrion Ops 2.0: what the tool was used for, what it did not own and how the checkpoints worked.
Codex is an OpenAI product, named here factually as a tool Aevrion used. Nothing on this page implies OpenAI endorsement of Aevrion, a partnership, or that any tool independently designed or approved this system.
01 / WHAT CODEX WAS USED FOR
Directed implementation, at production standard.
Per the project's phase reports and history, Codex carried the primary Sites engineering of Aevrion Ops 2.0—always from briefs and blueprints, always into review.
- 01
SITES ENGINEERING + IMPLEMENTATION
Primary implementation of the production site's components, routes and styling, working from Aevrion briefs and approved blueprints.
- 02
CONTINUUM IMPLEMENTATION
Realizing the approved Computational Continuum candidate—orchestration field, editorial systems and the seven-state build interaction—as working code.
- 03
MOTION + INTERACTION ENGINEERING
Implementing and repairing the homepage motion systems, including corrective passes recorded in the project history.
- 04
RESPONSIVE ENGINEERING
The mobile-integrity work that restored intentional composition across viewports after homepage promotion.
- 05
IDENTITY INTEGRATION
Producing and integrating the Logo 2.0 asset system into the running site under the approved specification.
- 06
ROUTE + CONTACT FOUNDATIONS
The internal-page system, route architecture, Contact intake foundation and the AI Build route family.
- 07
TESTS + QA SUPPORT
Rendered-route test suites, QA checklists and evidence capture that gate each phase's acceptance.
Each category above is traceable to versioned phases and reports in the project record— this list is documentation, not marketing recall.
02 / WHAT CODEX DID NOT OWN
The tool never held the pen on decisions.
The boundary is what makes the acceleration trustworthy. Five things stayed human throughout—visibly, in the project record.
- 01
BUSINESS STRATEGY
What Aevrion is, who it serves and how it is positioned—decided by Aevrion before implementation briefs existed.
- 02
BUSINESS ARCHITECTURE
The operating model, route responsibilities, content architecture and copy decisions—defined in owner-reviewed documentation.
- 03
CREATIVE DIRECTION
The art-direction decision came from an evidence-based lab process—multiple isolated concepts, synthesis, review boards—selected and approved by human direction.
- 04
ACCEPTANCE AUTHORITY
Every phase ended in human review against recorded criteria; corrective work in the history shows acceptance being enforced, not assumed.
- 05
PRODUCTION AUTHORITY
Promotion, indexing, activation, deployment and cutover decisions remain owner-controlled—several are still deliberately pending.
03 / WORKING IN CHECKPOINTS
Fast inside the phase. Gated between phases.
The project moved through versioned phases with recorded checkpoints. Inside a phase, implementation was fast; between phases, nothing moved without validation and an explicit decision.
- 01
BRIEF
Each phase started from a documented brief or approved blueprint—scope, constraints and acceptance expectations stated first.
WHAT IT ENFORCEDA defined task instead of an open prompt. - 02
BUILD
Codex produced the implementation for that phase inside an isolated branch or route.
WHAT IT ENFORCEDA reviewable candidate, quarantined from production. - 03
VALIDATE
Tests, rendered-output checks and QA passes ran against the candidate.
WHAT IT ENFORCEDEvidence, not impressions. - 04
REVIEW
Human review against the brief and the record—design, behavior, semantics, responsiveness.
WHAT IT ENFORCEDFindings on record. - 05
ACCEPT OR CORRECT
Accepted work advanced; defects produced corrective passes—the project history shows both.
WHAT IT ENFORCEDAn explicit decision every phase. - 06
PROMOTE
Only accepted candidates were promoted—the homepage itself lived as an isolated candidate before promotion.
WHAT IT ENFORCEDProduction moved on acceptance, never on momentum.
The history includes corrective passes—motion restored after a precision pass, mobile integrity restored after promotion. That is the review system working, and it is part of the evidence.
04 / EXAMPLES FROM AEVRION OPS 2.0
Seven checkpoints from the actual record.
Selected phases where Codex carried the implementation under direction—chosen for what each demonstrates, not as an exhaustive changelog.
- FOUNDATION
- The v0.1 production foundation: framework, tooling, testing harness and route skeleton.
- DISCIPLINE FROM DAY ONE
- SEMANTIC STRUCTURE
- The v0.2 homepage narrative built in semantic HTML before visual design.
- CONTENT BEFORE DECORATION
- CONTINUUM CANDIDATE
- The Phase 3.3 final art-direction candidate, integrated in isolation and refined through precision and motion passes.
- QUARANTINED UNTIL ACCEPTED
- LOGO 2.0
- Production identity assets built to specification and integrated with an accessibility-checked lockup.
- SPEC-DRIVEN
- HOMEPAGE CONTENT SYSTEM
- The 13-section production content architecture implemented against the approved copy deck.
- BLUEPRINT-GOVERNED
- CONTACT FOUNDATION
- Validated, fail-closed intake with development/production separation—behavior still verified by tests today.
- FAIL-CLOSED
- AI BUILD ROUTES
- The first two service routes, implemented from approved Phase 3.6A blueprints as review-gated candidates.
- REVIEW-GATED
05 / MULTI-AI ENGINEERING
More than one tool. One owner.
The project record also includes Claude Code: isolated creative research that informed the final art direction's interaction logic, selected motion work and parallel review-gated route production—each documented in internal ledgers.
ONE ACCEPTANCE AUTHORITY
Whatever produced the work, the same human direction defined it and the same acceptance standard judged it. Provenance is recorded per contribution; no percentages are published, because the point is the system, not a scoreboard.
This is not a tool comparison, and it is not a contest. It is the tool-independent principle applied to engineering: the requirement picks the instrument; the owner keeps the authority.
06 / WHAT THIS PROVES — AND WHAT IT DOES NOT
A real result.
Honestly sized.
What this build demonstrates: AI-assisted engineering can accelerate disciplined implementation substantially—when the architecture is explicit, the briefs are real, the output is tested and a human owns acceptance at every gate.
What it does not demonstrate: that AI can independently run a production engineering organization; that every project should use the same tools; or that any tool guarantees quality. The discipline produced the quality. The tools changed the speed at which discipline could work.
Remove the checkpoints from this story and it becomes a different story—one Aevrion would not publish.07 / FROM LAB TO PRODUCTION
How tool-assisted work earns promotion.
Whatever produces a candidate—Codex, Claude Code or a person typing alone—the same six gates stand between it and production.
- 01
TESTS
Rendered-route suites lock status, metadata, structure and content order before anything advances.
- 02
REVIEW
Human review against the brief, the record and the protected systems around the change.
- 03
ACCESSIBILITY
Semantic hierarchy, keyboard behavior, focus, contrast, forced-colors and reduced-motion checks.
- 04
RESPONSIVE QA
Nine-viewport audits with zero-overflow requirements and intentional mobile composition.
- 05
PERFORMANCE
No heavy frameworks, no speculative dependencies, reported CSS/JS impact per change.
- 06
PRODUCTION READINESS
Review gating, honest robots states, sitemap control and owner acceptance before any activation.
The gates are visible in this system today: review-gated routes, noindex states, per-batch evidence and tests that fail loudly. Promotion is a verdict, not a default.
08 / THE PRINCIPLE, APPLIED TO YOUR SYSTEM
Use the tool. Keep ownership of the system.
The same model applies to client work: AI acceleration where it earns its place, inside an architecture you own, with acceptance that stays human. Bring the business problem—the tooling follows the requirement.
Start a project