Humans lead. Agents execute. Skills compound.

Humans, orchestration, agents, Skills — layered.

Foundry is a layered hierarchy with humans at the top. It has two human role types and three technology layers:

• Customer & solution leadership — holds direction, taste, prioritisation, quality judgment, customer dialogue. Architects, service designers, owners of backlog, budget, and time.
• System builders — build and maintain the engine itself: Terrain, Signals, Foundry, including compute, identity, memory, tool access, billing, orchestration, agentic development, account management, skills, Shield. They don't deliver customer projects directly. They build the motor that makes delivery possible.
• Orchestration — planners who break work down, route it to the right agents, set boundaries, sequence, validation rules.
• Agents — specialised AI for code, research, testing, deployment. Organised into teams that mirror a human delivery team — discovery lead, delivery lead, research, architecture, developer, E2E tester — but parallel, faster, and always inside the frame the orchestration layer sets.
• Skills — the reusable craft agents use. Two types: Seven Peaks Skills (shared across clients) and Client Skills (built for you, owned by you). Prompts are disposable. Skills accumulate.

Identical AI model, nearly 2× performance inside a well-designed delivery system. The system is the product.

Engineering inside Foundry.

What a classic consultancy sells as utvikling — frontend, backend, API, mobile, integrations, QA, DevOps — is how Foundry produces working software. Same crafts, accelerated by a governed AI-native pipeline.

Inside Foundry, every engineering discipline gets a dedicated agent team operating inside its own Skills:

• Frontend — component generation validated against the design, a11y, performance, and your design system
• Backend & APIs — service scaffolding, contract tests, migrations, data-layer changes — every diff reviewed against Shield Layer 1
• Mobile & multiplatform — Kotlin Multiplatform and equivalent patterns where a single codebase serves iOS + Android + web
• Integration — adapters for ERP, CRM, billing, identity, whatever your stack actually uses
• Testing — unit, integration, E2E, visual regression, accessibility — generated and maintained by a dedicated test agent
• DevOps & deployment — infrastructure-as-code, CI/CD, observability, rollback — audit-trailed in Bridge

You still get engineering. You get it faster, with better coverage, and with every change visible and reviewable. Development is the output, not the offering.

AI is the medium, not a service line.

We don't sell "AI" as a separate product. AI is how Foundry operates — in every agent, every Skill, every Shield check. What a classic consultancy calls kunstig intelligens is available inside Foundry in three concrete ways:

• AI that builds — the whole of Foundry. Agents generating, reviewing, testing, and shipping code under human-named approvers
• AI inside your product — LLM features, RAG, domain-specific models, Client Skills trained on your data (which you own outright)
• AI operating on your operations — Terrain's continuous context build, Signals' market-moves interpretation, agent workflows running inside your business systems via MCP and direct integrations

Examples of Client Skills we've built inside real engagements: brand-voice skills tuned to specific tone constraints, domain-accuracy skills that catch regulatory or technical errors before Shield even sees them, integration skills connected to a specific customer's identity and billing systems. These are owned by the client. They work better in month six than they did in month one.

An AI model that fails 97% of the time without context succeeds nearly 100% of the time with it. Foundry supplies the context.

End-user first, at Foundry speed.

For 16 years Seven Peaks has held a clear end-user-first position: what's built must work for the humans who actually use it. Foundry preserves that. What changes is the speed from insight to working interface.

• Service design — end-to-end service journeys, customer process mapping, cross-channel design
• UI/UX — interaction design, prototyping, usability testing, component systems, visual design
• Design systems — tokens, components, documentation; Client Skills that keep agent-generated code aligned with the system without humans policing every pull request
• Brand & motion — visual identity work, motion design, presentation design
• Accessibility — WCAG 2.1 AA as a floor, not a ceiling, validated by Shield

In Foundry, design work feeds directly into agent-generated code. A designer's decision arrives at the frontend agent as structured Skills input, not as a Figma hand-off wishlist. The loop tightens.

A bridge has two sides.

Bridge is Foundry's customer-facing surface. Not a customer portal — you're not a guest looking into our system. You stand on your side, we stand on ours, and the surface is owned together. Contract, backlog, spend, Signals, decisions, audit trail, security status — one surface, equally visible to client and crew.

Bridge contains:

• The contract — the formal relationship, visible and auditable
• Cost management — spend vs. estimate, invoices, token usage in AI delivery, remaining budget
• Backlog and progress — kanban, epics, priorities, timeline, blockers
• Chat — traceable dialogue between Seven Peaks and client
• Signals and actions — proactive recommendations with draft actions ready for approval
• Upsell and continued development — offers on new features and follow-on work, presentable and signable directly in the surface
• Traceability, security, integrity — logs, audit trail, security status, compliance — as a live state, not an after-the-fact report

Bridge is where the three promises stop being slides and become daily practice. Ownership becomes visible. Acceleration becomes measurable. Capability transfer happens by watching.

Speed without recklessness.

Shield is a configurable safety layer built into Foundry — not bolted on. Three layers:

• Layer 1 — Seven Peaks standard — code review, dependency scanning, vulnerability detection, privacy audit, spec validation, Customer-DNA consistency check
• Layer 2 — client-specific — industry standards and compliance, client's internal security requirements, regulatory logic (GDPR, EU AI Act from 2 August 2026)
• Layer 3 — human approval gate — named approver for critical decisions; client sets their own gates

Some safety layers are permanent regardless of how capable the model gets. Access control is business rules. Human approval is risk management. Regulatory logic is compliance. These survive any model upgrade because they're not about what the model can do — they're about what must be true.

How an engagement with Foundry runs.

Every engagement has a named solution lead and a named project lead from our side. No anonymous account managers. Delivery happens in short iterations — typically 8–14 weeks per meaningful milestone — with weekly Bridge reviews and continuous Shield checks.

What the customer owns at the end of an engagement:

• All source code, documentation, and artefacts produced
• All Client Skills built for you — brand, domain, integrations
• Your Customer DNA and Client Memory — the structured context
• Full audit trail of decisions, spend, and security state via Bridge

What Seven Peaks owns: the method, Foundry as an engine, Seven Peaks Skills, security standards. The engine is ours. Everything it produces for you is yours.