Fixed-scope vs time-and-materials: which model actually delivers AI agents

When companies decide to hire an external team to build an AI agent, they typically choose between two contracting models: fixed-scope (fixed price, fixed deliverable) or time-and-materials (hourly or daily billing against a project). The choice feels tactical — a procurement decision about how to structure a contract. It isn't.

The contracting model is an incentive structure. It determines what the vendor is optimizing for. And for AI agent projects, the difference between the two models is the difference between an agent that ships and one that doesn't.

How time-and-materials works (and why it fails for AI agents)

In a T&M engagement, the client pays for time: an hourly or daily rate, usually against a not-to-exceed estimate. The vendor is incentivized to bill hours. The client gets a running meter.

For some projects, this works. Ongoing maintenance, exploratory research, or situations where requirements are genuinely unknowable up front can justify T&M. You're paying for expertise applied to a problem, not a specific output.

For AI agent implementation, T&M creates structural problems:

• No ship date. T&M contracts have budgets, not deadlines. "We'll work until we run out of budget" is not a delivery plan.
• Scope creep is profitable for the vendor. Every new feature request, every "while we're at it," every integration that wasn't in the original plan — all of them add hours. The vendor has no financial incentive to push back on scope expansion.
• The project "almost" never ends. On T&M, the agent is perpetually "almost done." There's always one more thing to refine. Shipping ends the billing relationship, so the incentive is to keep refining.
• Budget uncertainty is paralyzing. When you don't know whether the agent will cost $40K or $200K, it's hard to commit to the project or plan around it.

How fixed-scope works (and why it's different for AI)

In a fixed-scope engagement, the deliverable is defined before work begins: a specific agent, automating a specific workflow, delivered by a specific date, for a specific price. The vendor is accountable to the output, not the hours.

This creates alignment between client and vendor that T&M cannot:

• The vendor is incentivized to scope tightly. Every hour they spend on work outside the scope comes out of their margin. Clear scope is in their interest as much as the client's.
• There's a real ship date. The engagement ends when the agent is in production. That's the deliverable. Everything is organized around that date.
• The client knows the cost before they start. No meter running. No "depending on what we find." A price, agreed at scoping, that doesn't change unless the scope changes.
• Scope changes require explicit agreement. If something new comes up that wasn't in scope, both parties know it, price it, and decide whether to include it. No surprises.

The objection: "AI agent requirements are too uncertain for fixed scope"

The most common pushback against fixed-scope for AI agents is that requirements are inherently uncertain — you don't know what you'll discover until you start building, so how can you fix the scope?

This is partly true and often overstated. Most business workflow automation has well-understood requirements: the workflow exists today (humans are doing it manually), the inputs and outputs are defined, and the integrations are known. The uncertainty is in implementation detail, not in what the agent needs to do.

The scoping call exists to surface that implementation uncertainty before work begins. A good scoping process — one that defines the workflow in enough detail that a developer could start building on day one — removes 80-90% of the uncertainty that causes T&M projects to overrun.

The remaining uncertainty (edge cases discovered in build, unexpected integration constraints, performance tuning) is absorbed into the fixed price as a risk buffer. The vendor carries that risk, not the client.

When T&M might be appropriate

There are situations where T&M makes more sense than fixed scope:

• Exploratory research where you genuinely don't know what you're building yet
• Ongoing maintenance of an agent already in production
• Projects where requirements are legally or regulatorily constrained and can't be defined until compliance review completes

For first-time AI agent deployments — the situation most of our clients are in — none of these apply. The workflow exists. The requirements can be defined. The uncertainty is manageable with good scoping.

The decision framework

Before choosing a contracting model, answer these three questions:

1. Can I define what a successful outcome looks like? If you can describe the agent's trigger, workflow, and output clearly, you can scope it. If you can't, you're not ready to build.
2. Do I need a specific output by a specific date? If the answer is yes, T&M is the wrong model. Only fixed-scope creates accountability to a date.
3. What does the vendor's financial incentive optimize for? If they're paid by the hour, they're optimizing for hours. If they're paid for output, they're optimizing for output. Choose accordingly.

For most AI agent implementations, the answers point clearly toward fixed scope. The technology is mature enough to scope. The use cases are defined enough to specify. The only thing T&M adds is risk for the client and revenue optionality for the vendor.

Agent Implementations operates on fixed-scope only — one workflow, fixed price, 30 days to production. Book a scoping call to find out what your workflow would cost.