AI strategy for architecture firms

ALPA is a boutique design & technology studio for the built environment. Through AI Strategy & Adoption, we help architecture firms turn scattered experiments into a capability the whole practice owns.

Why an architecture firm needs an AI strategy

Most architecture firms are not short on AI tools. They are short on structure.

One person uses a general chatbot for meeting notes. Another tries image generation for concept work. Someone else keeps a spreadsheet full of prompts. Each experiment might be useful. Together they don’t make a capability: the workflows are private, they don’t carry over when people move between projects, and nobody can say what client material has gone where.

Without a strategy, the tools end up deciding for you. Each one comes with its own interface, its own memory, its own permissions and its own idea of what an architecture firm needs. Project context gets copied into five systems. Product decisions get made again from scratch. Over time the firm’s most valuable assets end up inside other companies’ products, one subscription at a time: its project history, its material decisions, its standards and its office methods.

An AI strategy is how a firm decides, on purpose:

• Which work AI should do, and which work stays with people.
• Where the firm's knowledge lives, so it stays portable and keeps getting more useful instead of being locked inside one vendor's workflow.
• What it pays for and what it owns. Most AI tools sold to architects are a prompt and an interface on top of a model your firm can already use.
• Who reviews what, before anything reaches a client.

Models will keep changing. A firm with a strategy gets more out of each new model, because the structure underneath is already in place. A firm without one starts over every time.

Read more: how to build your firm’s AI stack.

Who this is for

Principals and leaders of small and mid-sized architecture, engineering and construction firms whose teams are already experimenting with AI and who want those experiments to become part of how the studio works.

We partner with architecture, construction and product teams to build better tools, processes and platforms for the built environment. AI Strategy & Adoption is where that work starts.

What AI Strategy & Adoption solves

“Everyone uses AI differently, and none of it sticks.”

We map the recurring jobs in your practice where AI earns its place, such as project intake, meeting follow-up, precedent and material research, specification preparation and consultant coordination. Then we define a small number of repeatable workflows around them, so the work no longer depends on one person remembering the right prompt. See what agentic AI actually changes for practice.

“We keep getting pitched AI tools and can't tell which ones are worth it.”

We apply a simple test. If the tool holds proprietary data or performs validated computation, it may be worth buying. If its value is mostly prompts, templates and access to your own documents, your firm can own that workflow instead of renting it. Here's how to evaluate AI tools.

“Our knowledge is scattered across tools we don't control.”

We lay out an adoption roadmap in layers: model access, an agent harness, domain agents and skills, project memory, product and material memory, and governance and review. You don't have to build every layer yourself. The roadmap shows where your project history, standards and material decisions should live, so they stay with the firm.

“Someone wants to train our own model.”

Before anyone budgets for it, we sort out which kind of "training" is actually on the table and which approach fits the problem. More often than not, the answer is to give a model the right documents at the right time, not to change the model. Can your firm train its own model?

“We don't know who signs off on AI output, or what can be shared.”

Governance and review are part of every stage of the roadmap. That means clear guidelines on what can be shared, permission boundaries, source traces, and review states that show when work is draft, reviewed, client-ready or approved. The goal is not to slow teams down. It's to make AI output easier to trust, revise and approve. Read about AI governance and data security for architecture firms.

How your team learns it

Adoption depends on people knowing how to work with the tools, and how to check them.

• A self-paced AI course for architects. Six hands-on modules built around a fictional practice project: your first questions, your first file, your office standards, cleaning up a messy folder, building your own skill, and a closing drill in verifying AI output. Module 2 covers where your data actually goes. Module 6 teaches the habit that matters most: asking “show me where in the document it says that.”
• Rolling AI skills out to your team. How to install and distribute skills and the Architecture Studio plugin across a team, through Claude Desktop, Claude Code, project and managed settings, and team marketplaces.

Built and tested, not just advised

We build and ship the tools we recommend, and we publish how they perform.

• Architecture Studio is an open-source (MIT) agent harness and institutional-memory system for architecture practice, built on four layers: Studio, Knowledge, Tooling, Memory. It includes 61 skills, runs in Codex and Claude Code, and comes into Claude or ChatGPT through the hosted MCP.
• Norma is a materials library and agent that turns vendor emails, PDFs, quotes and schedules into a governed, cited library.
• Published evaluations. In 27 cold-start runs per arm, Architecture Studio 1.5 produced valid project records 55.6% of the time against vanilla Claude’s 14.8%. We also report where it made no difference. See how we evaluate what we build, along with our tests of five ways for agents to build a wall in Rhino.

When we recommend a layer of your stack, we’ve already built one.

Background

ALPA was founded by Federico Negro, an architect, designer and toolmaker based in Brooklyn, NY, who has spent the past two decades working at the intersection of architecture, technology and business.

In 2008 he co-founded CASE, a design-innovation and technology consultancy that grew to more than 70 staff working for organizations like Disney, Apple, and some of the world’s most well-known architects: Grimshaw, Snøhetta, SOM and many more. As Global Head of Design at WeWork, he led the scaling of the design and construction group from a few dozen to more than 1,000 architects, designers and engineers delivering work across 30+ countries. He went on to found Canoa, and has invested in AEC technology companies including Speckle, Monograph and Naya.

More about the studio →

FAQ

Why does an architecture firm need an AI strategy at all?

Your staff are already using AI. The real choice is between AI as a set of personal habits and AI as a firm capability. A strategy decides which work AI does, where the firm's knowledge lives, what you buy versus own, and who reviews output. Without one, those decisions get made one tool at a time, by vendors.

Should our firm train its own AI model?

Almost certainly not a foundation model. Pre-training takes frontier-lab budgets. Fine-tuning an open model on a narrow task is within reach, and for factual knowledge, retrieval from your own documents is usually the better option. Can your firm train its own model?

Which AI tools should we buy?

Buy tools that hold proprietary data or perform validated computation. Own the workflows whose value is prompts, templates and your firm's documents. How to evaluate AI tools.

Do we have to use Architecture Studio?

No. The strategy starts from how your firm works. Where it calls for an agent harness, Architecture Studio is one option. It runs in Codex and Claude Code and comes into Claude or ChatGPT through the hosted MCP.

How do we keep firm and client data under control?

Start with clear guidelines on what can be shared, then make sure your tools respect permission boundaries and keep review with your people. Read about AI governance and data security for architecture firms.

Can our team learn this on their own?

Yes. The /learn course is self-paced and runs on a fictional practice project, so a mistake costs nothing real. The team distribution guide covers putting skills in everyone's hands.