Before Your Architecture Firm Deploys AI Agents, Understand the Cost

“A workflow that costs $10 but saves a principal two hours may produce an excellent return. A recurring agent that costs hundreds of dollars but creates little measurable value should be redesigned or discontinued.”

AI adoption in architecture firms usually begins quietly.

A few employees open personal ChatGPT accounts. Someone uses Copilot to summarize meeting notes. A project manager experiments with Claude to draft a proposal. Before long, firm leaders begin asking a bigger question:

Should we deploy AI across the entire company?

For a 20-75 person architecture firm, that can be an important step forward. Properly deployed, AI can reduce administrative work, help teams find information faster, improve consistency, and give architects more time to focus on clients and design.

But as firms move beyond basic chat tools and begin using AI agents, another issue quickly emerges:

Cost.

We recently experienced this firsthand at ArchIT. Only two of us were actively testing agentic AI tools, yet our monthly spending increased from approximately $1,500 to $3,300 – even after we introduced spending limits.

That does not mean the technology is not worth using. In many cases, the return is significant. It does mean architecture firms need to understand how AI costs develop and how to control them before usage expands across the organization.

From AI Experimentation to AI Agents

Most firms begin with what we call the experimentation stage.

Employees ask AI to:

• Draft emails
• Summarize meetings
• Review documents
• Brainstorm design or business ideas
• Help prepare proposals
• Research products or materials

These activities are typically performed manually, one conversation at a time.

The next stage is more powerful. The firm begins creating AI agents that can complete recurring tasks, access company information, follow established instructions, and sometimes run automatically.

An agent might:

• Review emails, Teams messages, and meeting notes to identify decisions
• Prepare a first draft of a client proposal
• Organize project information for a business review
• Analyze a spreadsheet and prepare a payroll or bonus report
• Create a draft project page for the firm’s website
• Search past proposals to help a team respond to a new opportunity

This is where AI begins to move from a useful individual tool to a real operational capability.

It is also where costs can rise quickly.

Why AI Agent Costs Add Up

AI platforms generally charge based on the amount of information they process and generate.

Every time an AI agent reviews a long document, analyzes a spreadsheet, searches a large folder, or creates an output, it consumes computing resources. The larger the input and the more complex the task, the more expensive the process can become.

Imagine asking an AI tool to review seven years of proposals every time your firm creates a new response.

The agent may have to reread thousands of pages, interpret the information, compare it with the new opportunity, and generate a recommendation. Even when the output is relatively short, processing the source material can be costly.

The same issue can occur when an employee uploads a large manufacturer spreadsheet containing tens of thousands of products and asks AI to recommend three options. The employee may only need a small answer, but the AI still has to process the entire file.

For a single experiment, that may be acceptable.

Repeated across 25, 50, or 75 employees, however, those costs can grow rapidly.

The Goal Is Not to Minimize AI Use

Architecture firms should not approach AI cost management by preventing people from using the technology.

That would likely limit innovation and slow down adoption.

The better goal is to distinguish between three types of work:

1. Experimentation
2. Repeatable AI workflows
3. Production systems

Experimentation is naturally less efficient. Teams need room to test ideas, refine instructions, and determine whether a workflow creates real value.

Once the firm identifies a successful use case, however, it should not continue rebuilding the process from scratch.

That workflow should be standardized.

Turn Successful Workflows Into AI Skills

One of the most practical ways to reduce AI costs is to turn proven workflows into reusable skills.

A skill is a saved set of instructions, resources, and processes that an AI agent can reuse. Instead of asking the system to rediscover the correct approach every time, the skill gives it a repeatable method.

For example, suppose your firm uses an AI agent to create project pages for its website.

The first attempt may involve substantial experimentation:

• Reviewing existing project pages
• Understanding the website structure
• Developing the page layout
• Determining what information to include
• Refining the writing
• Adjusting calls to action
• Connecting the workflow to the website

That first version may be relatively expensive because the AI is helping build the process.

Once the firm is satisfied with the result, the process can be saved as a skill.

The next time the firm completes a project, the user can provide the new project information and invoke the existing skill. The AI no longer has to recreate the strategy and structure from the beginning.

It simply follows the established process.

This makes the workflow faster, more consistent, easier to share, and often less expensive.

AI Skills May Become the New SOPs

Architecture firms already rely on standard operating procedures to create consistency.

A good SOP explains:

• Who is responsible
• What information is required
• Which steps should be followed
• What the final output should look like
• How the work should be reviewed

AI skills serve a similar purpose.

As firms adopt agentic AI, shared skill libraries may become an important part of their operational infrastructure. Instead of every employee inventing a different approach, the firm can create approved, repeatable ways to use AI.

A shared library might include skills for:

• Proposal preparation
• Meeting summaries
• Project closeout
• Client onboarding
• Website project pages
• Business development research
• Quality-control checklists
• Internal reporting
• Specification review
• Staff training materials

Firms should also maintain a library of effective prompts for tasks that do not require a complete agentic workflow.

The goal is to convert individual experimentation into institutional knowledge.

Use AI Only Where AI Is Needed

Another cost-control strategy is to avoid asking AI to perform every step of a workflow.

Some tasks do not require artificial intelligence.

Traditional automation can:

• Move files
• Extract data from known fields
• Update a database
• Apply a standard template
• Create folders
• Transfer information between systems
• Generate a Word document from structured data
• Trigger notifications

AI is most valuable when interpretation, judgment, synthesis, or language generation is required.

A well-designed workflow might use conventional automation to collect and organize information, send only the relevant subset to AI for analysis, and then use automation again to format and distribute the result.

This hybrid approach can deliver the same outcome with less AI processing.

For architecture firms, this distinction is important. It is easy to use an AI agent for an entire workflow simply because the agent can do it. That does not mean it is the most secure, reliable, or cost-effective design.

Structure Your Firm’s Data

AI becomes more expensive when it repeatedly has to interpret large quantities of unstructured information.

Consider a firm that wants AI to help prepare proposals based on previous successful work.

The simplest approach is to give the agent access to a folder containing years of proposals and ask it to review everything each time.

A better long-term approach is to organize and index that information.

Relevant proposal data might include:

• Client type
• Project type
• Building size
• Delivery method
• Services provided
• Fee structure
• Project location
• Win or loss
• Differentiators
• Team members
• Final contract value

Once that information is structured, the AI can retrieve the most relevant examples rather than rereading every proposal.

This reduces processing, improves the quality of the response, and creates a more useful internal knowledge system.

Give Employees a Budget for Experimentation

Employees need space to explore AI, but firms should establish reasonable boundaries.

A practical AI rollout should include:

• Approved tools
• An AI use policy
• Data-security guidance
• Employee training
• Department or user budgets
• Usage monitoring
• A process for approving new agents
• A method for documenting successful workflows
• Regular reviews of value and cost

The goal is not to evaluate AI based only on the monthly bill.

A workflow that costs $10 but saves a principal two hours may produce an excellent return. A recurring agent that costs hundreds of dollars but creates little measurable value should be redesigned or discontinued.

Leaders should evaluate both sides of the equation:

What does the workflow cost, and what valuable work does it eliminate or improve?

Start With the Work Your Team Should Not Be Doing

The best AI opportunities are often not the most glamorous ones.

They are the repetitive tasks that consume time but do not require the highest-value expertise of your architects, project managers, and firm leaders.

Examples might include:

• Moving information between spreadsheets
• Preparing recurring reports
• Summarizing meetings
• Organizing project records
• Drafting routine communications
• Finding information across multiple systems
• Creating first drafts of proposals
• Comparing documents
• Preparing internal follow-up lists

The purpose of AI is not to replace architects.

It is to remove work that prevents architects from spending more time designing, solving problems, serving clients, mentoring employees, and leading the firm.

What 25–75 Person Architecture Firms Should Do Now

Most architecture firms in this size range do not need to build an advanced AI infrastructure immediately.

They do need to establish the right foundation.

Begin by identifying how employees are already using AI. Approve a secure set of tools. Create a practical policy. Train the team. Give employees room to experiment within defined budgets.

Then watch for repeatable use cases.

When a workflow proves valuable:

• Document it
• Refine it
• Convert it into a reusable skill
• Share it with the appropriate employees
• Measure its cost
• Measure the time or value it creates
• Redesign it when traditional automation can perform part of the work

AI adoption is not a one-time software purchase. It is an operational transformation.

The firms that benefit most will not simply give every employee an AI license. They will create a structured system for experimentation, governance, workflow design, knowledge sharing, and cost management.

The Bottom Line

AI agents can create meaningful advantages for 25–75 person architecture firms.

They can help small and mid-sized teams operate with capabilities that once required significantly more administrative support, technical resources, or outside consultants.

But agentic AI is not free, and its costs can become difficult to manage when every employee is experimenting independently and every workflow is rebuilt from scratch.

The answer is not to avoid AI.

The answer is to adopt it deliberately.

Give your team room to experiment. Turn successful experiments into reusable skills. Use traditional automation for predictable tasks. Structure your data. Monitor costs. And continually compare the expense of each workflow with the value it creates.

That is how an architecture firm moves from scattered AI usage to a secure, repeatable, and financially sustainable AI practice.

At ArchIT, we help architecture firms deploy AI securely, build practical workflows, manage adoption costs, and connect AI initiatives to the firm’s broader IT environment.

Learn more at getarchit.com.