Beyond features Platform & capability work May 2026

AI work that isn't tied to any single feature.

The FDLC describes how features get built. The work below is the platform that makes that workflow possible. Constitutions, companion files, tooling stack, governance cadences, and enablement infrastructure. These don't ship with a feature. They're what every feature is built against.

Context layer

Constitution before autonomy.

Agents operating without a defined set of rules produce plausible outputs that subtly violate constraints. The context layer is a three-stream infrastructure (design, platform, engineering) that converges at generation time. Companion files extend it with domain-specific context the team owns.

Org-wide Owned by DS Team

Textura design constitution

The rule set agents use at generation time for design output. Accessibility floors, design system coverage, UX standards, and visual contract enforcement. The first thing any design-generating agent loads.

  • Accessibility, DS coverage, and UX standards encoded as machine-readable rules.
  • Versioned and amended through the AI FDLC Review, not edited ad hoc.
  • Backed by Textura MCP, which provides live component and token state at generation time.
Cross-domain Owned by DS Team

Textura platform constitution

Layout-sensitive and multi-domain composition rules. Used where features span product surfaces or where composition correctness matters beyond a single domain's design system.

  • Cross-domain layout rules and composition patterns.
  • Loaded only when the feature is composition-heavy or multi-surface.
  • Promotion path: domain pattern candidates can graduate up to this layer.
Org-wide Owned by Architecture

Engineering constitution

Quality, security, architecture, and service contract rules. Loaded with every code-generating run. The reason AI-generated code can be reviewed against a stable target rather than judged in isolation.

  • Code quality, security, and architectural floors enforced at generation time.
  • Service contracts and API guidelines encoded so cross-team breakage is caught up-front.
  • Amended through Architecture Review and the AI FDLC Review.
Planned Cross-domain

Engineering platform constitution

Cross-domain API contracts, integration event standards, and shared service patterns. Held in the deferred register until pilot signal confirms what platform-level rules need to be enforced at generation time.

  • Cross-domain API and integration patterns.
  • Activation gated on pilot evidence about real platform-level failures.
  • Bundles with portal platform decisions in the deferred architecture register.
Companion files

Domain context the team owns.

Constitutions are org-wide. Companion files are how each domain extends them with the specifics agents need to generate good output: the data shapes, the state list, the design intent, the spec sufficiency bar. Authored at the domain level, refreshed continuously.

Per domainDesign lead

Domain design companion

Domain-specific design context, layered on top of the Textura design constitution. Maintained by the domain design lead. Stale companion files are a leading indicator of generation quality decline.

Per domainEngineering lead

Domain engineering companion

Domain-specific engineering context: data shapes, state machines, service hooks, recurring patterns. Authored by the domain engineering lead. Used in every per-unit-of-work generation in that domain.

Org-widePM Leadership

PM spec guide

Defines spec sufficiency standards by feature type. The minimum viable spec is different for a small UX refinement vs. a new service integration; the guide makes that explicit so PMs are not guessing the bar.

ContinuousDS Team monitors

Companion file health

Staleness is monitored, not just authored once. The DS Team and per-domain leads watch for drift between companion files and shipping reality. Context quality is the generation quality ceiling.

Tooling & tracks

Two tracks. One workflow.

Teams operate the same workflow against two different tool tracks. The choice depends on workload, context-loading depth, and cost-per-outcome signal. Track A is widely available today; Track B is the path we are licensing for.

Track A · Available today

GitHub Copilot with structural discipline

Used with explicit context loading, constitution grounding, and the four mandatory human checkpoints. The lowest-friction track for teams already operating on the GitHub stack.

  • Constitution and companion files loaded as repo context.
  • Activity Matrix and prompt library guide which mode (agent-led, agent-assisted, human-led) per activity.
  • Best-fit for shorter generation units and incremental refactors.
  • Cost profile is well understood; budget envelope is mostly seat-based.
Track B · Pursuing enterprise license

Claude Code with persistent context

For deeper context loading, multi-step agentic work, and constitution-enforced generation across larger units of work. Requires CI/CD pipeline setup and a token budget governance model owned by EngOps.

  • Persistent context handles spec, constitution, and companion files in one session.
  • Multi-step plans against the spec; checkpoint integration with the gate structure.
  • Best-fit for substantial generation units, complex refactors, and agentic workflows.
  • Cost-per-outcome is what the pilot is calibrating.
AI Agent Landscape

Beyond IDE agents.

The agent landscape is the set of horizontal agent capabilities the org maintains beyond per-feature work. Some are live, some are gated on pilot signal, some sit in the deferred register pending evidence that the work is real.

Live

VoC + behavioral synthesis

Continuous customer signal: call recordings, escalation patterns, behavioral telemetry, synthesized into briefs PM dispositions at Brief (G0).

Live

Textura MCP

Live component and token state at generation time. The bridge between the design constitution and what the agent can actually compose against.

Live

Docs continuity agent

Flags drift between shipped behavior and published documentation. Lets TC focus on drafting and judgment, not change detection.

Deferred architecture register

Ten items are deliberately held until pilot signal proves the work is needed. Activation is gated on evidence, not enthusiasm. A sample of the register:

DAR-01
Textura MCP Phase 2 — live design system mutation API
Deferred
DAR-02
Textura MCP Phase 3 — multi-domain composition introspection
Deferred
DAR-03
Portal platform agent surface
Deferred
DAR-04
Context governance scaling (versioning + amendment workflow at org scale)
Deferred
DAR-05
Engineering platform constitution authoring & rollout
Deferred
DAR-06
Cross-product agent surface for support & CS workflows
Deferred
Governance cadences

Four standing reviews. One signal loop.

Constitution amendments, pattern compounding, and workflow calibration all need signal from deliberate practice. Four standing cadences turn what teams see into decisions that flow back down. Monthly during pilot and active rollout, quarterly at scale.

Cadence 01

Domain sprint cadence

Pod-owned, per-domain review of in-flight capture: friction patterns, generation quality signals, gate exit data. Feeds the AI FDLC Review with domain-grounded evidence.

Cadence 02

Standing design system review

DS Team lead. Approves pattern promotions from feature-level candidates into the design system; ratifies constitution amendment candidates surfaced during Build.

Cadence 03

Standing architecture review

Cross-domain contract issues, deferred register activations, engineering constitution amendments. Engages on triggers, not on a fixed schedule of every feature.

Cadence 04

AI FDLC Review

Central, cross-functional decision and ratification body. Advancement decisions, rollout calibration, constitution amendments, sequencing for new teams. The signal loop closes here.

Mechanism

In-flight capture at gates

Embedded in gate checklists, not a separate task. Cycle time, rework rate, handoff clarification burden, implementation fidelity, cost-per-outcome by track.

Output

Rollout calibration package

What the AI FDLC Review produces: advancement decision, team sequencing model (by readiness, not seniority), updated cost governance, ratified amendments.

Decisions flow back down to teams; signal flows up from the Pod cadence. The signal your teams produce is what shapes rollout for everyone. Governance principle, AI-Enhanced FDLC
Enablement infrastructure

The work that keeps teams unblocked.

If the workflow is the road, the enablement layer is what teams need so they can actually drive it: prompt libraries, onboarding paths, templates, and sequencing logic that treats readiness as the variable, not seniority or priority.

Resource 01

Prompt library by Activity Matrix

Role-based starting points: spec drafting, AC generation, variant generation, code review, test scaffolding. Curated, versioned, and shared across the org rather than re-invented per team.

Resource 02

Onboarding & tooling setup guides

Setup lift lands in teams (repo provisioning, tooling, system access). The guides exist so the lift is small and predictable, not an exploration project.

Resource 03

Spec, constitution & companion templates

Templates for the artifacts that drive generation. PM spec template by feature type, design constitution scaffolding, engineering companion file shape.

Resource 04

Team sequencing model

Defines which teams onboard first, based on readiness criteria (companion files ready, spec sufficiency demonstrated, leadership support). Set by the calibration package, not pre-committed.

Resource 05

FDLC documentation

Practitioner-facing documentation of the workflow, co-authored by TC and the AI working group. This site is one entry point; the deeper reference lives alongside it.

Resource 06

AI Ways of Working group

Cross-org working group that surfaces friction, recommends adjustments, and feeds survey signal into the AI FDLC Review. Open contributor model.

Experiment & contribute

The workflow evolves with the teams running it.

The FDLC isn't a static document handed down from a working group. Teams running it daily see friction, gaps, and opportunities first. The contribution loop turns those local experiments into ratified amendments that ship to everyone.

Step 01

Experiment locally

Pods can pilot variations — a new prompt shape, an adjusted gate threshold, a companion file format, a different review cadence — inside their own domain. No central approval needed to try something for a sprint or two.

Step 02

Capture evidence

Document what changed, what it replaced, and what shifted: cycle time, rework rate, generation quality, gate exit data, friction patterns. Evidence is what separates a proposal from an opinion.

Step 03

Submit through the loop

Route the proposal to the AI Ways of Working group for shaping, then to the AI FDLC Review for ratification. Constitution amendments, workflow changes, and new templates all flow through the same standing cadence.

Step 04

Ship to everyone

Ratified amendments update the constitutions, companion templates, or workflow definitions. The change is versioned, dated, and announced — so every team picks it up the same way features ship to customers.

Decisions flow back down to teams; signal flows up from the pods. The same loop that calibrates rollout also accepts the changes teams want to make. Contribution principle, AI-Enhanced FDLC
Addenda

Deeper reading.

Companion references that go beyond what fits on this page. Each one is a focused deep-dive on a specific facet of the AI-enhanced workflow — the activity matrix, the agent inventory, cost discipline, a worked example, and the year-long rollout plan.