Leading AI adoption in
architecture practices.

Usage policy and approved platforms

A written AI usage policy lists approved platforms, defines data-handling boundaries and sets expectations for staff. AI tools are approved - they are not left to individual discretion. The policy covers Claude, Microsoft Copilot and the agents built on top of them.

Shared prompt library

A curated library of tested prompts shared across the practice, covering Claude and Microsoft Copilot. Good technique spreads rather than sitting with individuals. Prompts are reviewed before they enter the library and updated when behaviour changes.

Policy-governed assistant with a confidentiality gate

A practice assistant that flags any query touching Building Regulations, fire, structural or CDM risk for human review before the output is used. The agent answers; a named person decides whether to rely on it. Sensitive project data never leaves the practice environment.

Drafts-only rule

Agents prepare. People check and send. No AI output goes to a client, consultant or authority without a named individual reading and approving it. This is not a technical limitation - it is a governance decision, applied consistently across every agent in the practice.

Skills promoted after two clean runs

Procedures are not promoted to reusable skills until they have completed two clean, checked runs on real work. This controls quality and prevents bad patterns from becoming permanent. Every skill in the library has a record of the runs that qualified it.

Working across all major platforms

The practice is not committed to a single vendor. We work across Claude (Anthropic), Microsoft Copilot, and the open-source Revit MCP ecosystem. Governance applies regardless of the platform, and staff are trained on the differences between them.