StackFoundry AI

StackFoundry AI is an agentic, “agency-in-a-box” workflow that moves legacy front-ends to Sitecore XM Cloud 50–90% faster. It is live as an internal AI-powered pipeline today (already used on two client projects, including CWS.com EU) and is being prepared for public release. We build on LangChain, coding agents, Sitecore MCP, custom RAG, and an artifact store to keep humans in control while the system does the heavy lifting.

What it delivers (at a glance)

• Discovery-to-migration copilot that understands pages visually and structurally (DOM + screenshots + network traces).
• A canonical Component Matrix that aligns specs, code, and XM Cloud renderings/templates.
• Next.js-first codegen wired to XM Cloud placeholders, layouts, and data templates via MCP.
• Automated content mapping, creation, and page assembly with traceability back to source URLs.
• Guardian QA: visual diffs, accessibility/SEO checks, tests, and human approval gates.

Five capability areas

1. Discovery & Component Modeling — crawl, segment, and cluster components into a clean matrix.
2. UX Alignment (optional) — apply or propose design tokens and variants; stay faithful or re-skin.
3. Component Library & XM Cloud Integration — generate Next.js components, renderings, placeholders, and data templates.
4. Automated Content Migration & Page Assembly — map content into fields, create items/pages, preserve SEO.
5. Governance & Human-in-the-Loop — approvals, tests, visual diffs, and rollback-friendly MCP runs.

Agentic architecture (MVP state)

How it works (human-steered, AI-heavy)

• Crawl & see: snapshot pages (HTML + network + multi-breakpoint screenshots) and segment components with a multimodal model.
• Matrix-first: cluster components, dedupe variants, and map them to XM Cloud renderings, placeholders, and data templates.
• Design-aware: optionally re-skin via design tokens; otherwise, faithfully lift existing look and feel.
• Generate and wire: produce Next.js components + field contracts; configure XM Cloud via MCP with matching schemas and placeholders.
• Migrate content: map page content into component fields, create items/pages, keep source → target traceability.
• Guardrails: run tests, a11y/SEO checks, and visual diffs; publish only after human approval.

Feasibility and scope

• This works today with top-tier multimodal models when positioned as an agentic copilot.
• Known limits: JS-heavy or highly bespoke interactions still need a Sitecore architect in the loop.
• Strengths: Component Matrix grounding keeps mapping accurate; MCP gives safe, auditable CMS writes; design tokens prevent drift.

Why it matters

A formalized, codified framework that turns AI-assisted migration into a repeatable product, not a one-off script. It mirrors emerging patterns (e.g., Sitecore XM Cloud Shift AI and other MCP-driven agents) but extends them into a multi-platform, design-aware offering with strong human controls.

Where we’re going

• Public release in 2026 in partnership with 7kingscode; foundation of the 7kingscode Sitecore practice (led by me as CTO).
• Multi-CMS adapters after XM Cloud; multi-head outputs (Next.js first, Flutter/mobile next).
• Progressive automation: start as supervised copilot, expand auto-approval for low-risk sections over time.