How Agencies Should Quote AI-Assisted Software Projects
AI-assisted development changes agency quoting in two opposite ways. It can reduce delivery time for common software work, but it also introduces new uncertainty: tool usage, agent loops, model selection, prompt and retrieval quality, runtime API spend, and client expectations around "AI magic." If your quote only says "40 hours at our hourly rate," you are hiding the parts most likely to move.
Quote the outcome, then expose the assumptions
Clients usually do not care which coding agent you use. They care what gets delivered, when, and what it will cost to operate. Your quote should still expose the assumptions behind the number: team size, build duration, project complexity, AI coding intensity, runtime AI usage, and maintenance expectations. These assumptions make scope changes easier to price later.
Separate build cost from operating cost
An agency quote should not treat AI as one line item. Split it into at least three categories:
- One-time build AI tooling cost during delivery.
- Ongoing maintenance AI tooling cost after launch.
- Runtime AI API cost if the client's product uses AI for end users.
This helps clients understand that the launch invoice and the monthly operating budget are different conversations. It also prevents runtime API usage from being mistaken for your agency's internal tool cost.
Use ranges, not false precision
A fixed quote can still include a range internally. Low, expected, and high estimates give you a way to set contingency and protect margin. If the expected build AI cost assumes daily assistant usage but the team ends up running multiple autonomous agents against a messy legacy codebase, the quote needs room for that difference.
Make runtime AI a client-owned budget
If the product uses OpenAI, Anthropic, Google, or another provider at runtime, decide who owns that account before launch. For most client projects, the client should own the production provider account, billing, spend caps, and model keys. Your agency can manage implementation, but the client should understand that user activity drives this bill.
Put guardrails in the proposal
- State the assumed number of monthly active users for runtime AI estimates.
- State whether free-tier usage is included, capped, or excluded.
- Define what counts as a new AI workflow versus a tweak to an existing one.
- Include maintenance assumptions for model changes, prompt tuning, and monitoring.
- Require re-estimation when scope adds new integrations, document types, or agent tools.
AI can make agencies faster, but it does not remove estimation discipline. It makes discipline more important because the cost drivers are less visible to clients than ordinary development hours.