Yes, firms should charge clients for AI agent time — but the fee must reflect real costs and real value, not the hourly rate a human would have charged for the same task.
AI agents are not free. Every task an agent executes costs money: tokens, inference compute, orchestration overhead, tool calls, and human oversight. A 2025 analysis of enterprise AI deployments found that mid-tier agent operations cost between £2,500 and £10,000 per month for firms running 5,000 or more tasks. Those costs need to go somewhere. Absorbing them indefinitely is not sustainable. Passing them on without explanation is not ethical. The answer sits in the middle: charge fairly, disclose openly, and price on value rather than machine-hours.
Key Takeaway: Charge for AI agent work based on value delivered and real costs incurred — not human-equivalent hourly rates. Always disclose AI involvement.
Why Is This Question So Urgent Right Now?
AI agents have moved past the pilot stage. They handle legal research, financial modelling, content creation, code generation, and client reporting across professional services firms worldwide. A 2025 survey by a global management consultancy found that 53% of professional services firms plan to deploy AI agents by 2027. Many already have.
The problem is billing. Traditional hourly billing assumes a human did the work. When an AI agent completes a task in 4 minutes that would have taken a junior consultant 3 hours, the maths falls apart. Billing 3 hours for 4 minutes of work feels dishonest. Billing 4 minutes feels like giving away the firm’s investment in AI infrastructure.
Clients are asking questions. “Did a person write this?” “Why am I paying the same rate if a machine did it?” “What exactly am I being charged for?” Firms that cannot answer clearly are losing trust — and losing clients.
What Does AI Agent Work Actually Cost?
The perception that AI work is “basically free” is wrong. Here is what firms actually pay to run AI agents on client work:
Token and inference costs. Every prompt sent to a language model costs money. Input tokens, output tokens, and reasoning tokens all carry a price. Complex tasks that require multi-step reasoning can cost £1–5 per execution. Simple tasks cost pennies. Over thousands of tasks per month, the total adds up fast.
Orchestration and infrastructure. Running agents requires orchestration frameworks, vector databases for retrieval, logging systems, and monitoring dashboards. These fixed costs sit underneath every task the agent performs.
Human oversight and review. No responsible firm ships AI output to a client without human review. That review time is real labour. In most professional services workflows, human review accounts for 30–50% of the total time spent on an AI-assisted deliverable.
Quality assurance and error correction. AI agents make mistakes. Hallucinated citations in legal research. Incorrect calculations in financial models. Off-brand messaging in marketing copy. Catching and correcting these errors costs time and money.
Tool and API calls. Agents call external services — search engines, databases, code interpreters, design tools. Each call carries a cost.
A mid-size consultancy running AI agents across 20 client projects reported total AI infrastructure costs of £7,200 per month — before counting human oversight time. That is a real cost that must be recovered somehow.
What Is the Case for Charging Clients?
Five arguments support charging clients for AI agent work.
Real costs exist. The firm pays for compute, tokens, infrastructure, and oversight. These costs are directly attributable to client projects. Absorbing them permanently erodes margin.
The value delivered is the same — or better. A contract reviewed by an AI agent and verified by a senior lawyer produces the same output as one reviewed by a junior associate over three hours. The client receives identical value. In many cases, the AI-assisted version is more thorough because the agent checks every clause systematically.
Speed has economic value. A market analysis delivered in one day instead of two weeks has tangible value to the client. Faster delivery means faster decisions, earlier market entry, and reduced risk. That speed premium deserves compensation.
Clients already pay for technology. Professional services firms have always passed through technology costs. Software licences, database access, and research tool subscriptions appear on invoices routinely. AI agent costs are the same category.
Precedent exists across industries. Radiology practices charge for AI-assisted scan analysis. Engineering firms charge for computational modelling. Financial institutions charge for algorithmic trading infrastructure. Charging for AI agent work follows established norms.
What Is the Case Against Charging?
Four arguments push in the other direction.
Perception risk. Some clients feel that paying for “machine work” is unfair, regardless of the actual cost or value. This perception is strongest among clients who do not understand AI economics. It is weakest among technically sophisticated buyers.
Competitive pressure. If a rival agency absorbs AI costs and offers lower prices, the firm that charges separately may lose bids. In commoditised service categories, this pressure is real.
Trust erosion. If a client discovers that AI was used without disclosure, the billing conversation becomes toxic. The issue is not charging — it is charging without transparency. A 2025 survey of professional services buyers found that 68% were comfortable paying for AI-assisted work when disclosed upfront. That number dropped to 12% when AI usage was discovered after the fact.
Market expectation. Some clients expect AI to reduce their costs, not maintain them. They see AI as a reason for lower fees, not a new line item. Managing this expectation requires clear communication about where AI adds value versus where it reduces cost.
How Do Different Industries Handle AI Agent Billing?
| Industry | Common approach | Key consideration |
|---|---|---|
| Law firms | Reduced hourly rate for AI-assisted tasks, with disclosure | Bar association guidance varies by jurisdiction |
| Consulting | Blended into project fees, not itemised separately | Clients expect outcomes, not time breakdowns |
| Marketing agencies | Separate line items for AI-generated deliverables | Clients want to know what was AI vs human |
| Accounting firms | Bundled into service fees with AI efficiency passed as savings | Regulatory requirements for AI disclosure growing |
| IT consultancies | Cost-plus model with AI infrastructure as a pass-through | Technical clients understand AI cost structure |
Law firms face the most scrutiny. Several bar associations have issued guidance requiring disclosure of AI use in client work. Some explicitly state that firms should not charge human hourly rates for AI-generated work. Others leave it to the firm’s discretion, provided the client is informed.
Consulting firms tend to embed AI costs in project fees without detailed breakdowns. Clients buy outcomes — a strategy deck, a market analysis, a transformation plan — and care less about whether AI or humans produced the first draft.
Marketing agencies increasingly separate AI-generated deliverables on invoices. Clients in creative industries want to know what was produced by humans and what was generated by AI, particularly for brand-sensitive work. For more on how agencies invoice AI deliverables, see our agency-specific guide.
How Do You Decide Whether to Charge?
Use this framework to make the decision for each client and service line.
Step 1: Calculate the true cost. Add up token costs, compute, orchestration, tool calls, and human oversight for the specific task or deliverable. This is your cost floor. You cannot charge less than this without losing money.
Step 2: Assess the value delivered. What is the outcome worth to the client? A contract review that protects against a £500,000 liability has a different value than a social media post. Price should reflect client value, not firm cost.
Step 3: Choose a billing model. Three options work:
- Cost-plus: AI costs + margin (typically 30–50%). Simple, transparent, low-risk.
- Value-based: Fee reflects the outcome’s value to the client. Higher margin, requires strong value articulation.
- Blended rate: Single rate that covers both human and AI time. Easy for clients to understand, harder to track internally.
For a deeper look at billing model options, see our full comparison guide.
Step 4: Disclose AI involvement. Show it on the invoice, mention it in the proposal, include it in the engagement letter. Transparency is not optional. Firms that disclose AI involvement and explain their pricing retain clients at higher rates than those that hide it.
Step 5: Review quarterly. Client expectations, AI costs, and competitive dynamics shift fast. Revisit your approach every quarter. What works today may not work in six months.
What Happens When Firms Hide AI Usage?
The downside risk is severe. When clients discover undisclosed AI usage — and they will — three things happen:
- Trust collapses. The client questions every past invoice. “What else was AI doing that I didn’t know about?”
- Fee disputes escalate. Clients demand retroactive discounts for AI-assisted work they were charged human rates for.
- The relationship ends. In professional services, trust is the product. Once it breaks, clients leave.
A 2025 case study from the legal industry illustrates this. A mid-size firm used AI agents for document review on a litigation matter without disclosing it. The opposing counsel identified AI-generated patterns in the work product. The client demanded a full audit of past invoices and terminated the relationship. The firm lost £2.1 million in annual revenue from that single client.
The lesson is straightforward. Firms that disclose AI usage and price fairly keep clients. Firms that hide it lose them.
How Should You Present AI Charges to Clients?
Three approaches, ranked from most to least transparent:
Full disclosure. The invoice shows: “AI agent — contract review — 12 minutes — £85 (includes AI cost + human review).” The client sees exactly what happened.
Category disclosure. The invoice shows: “AI-assisted legal research — £1,200.” The client knows AI was involved but does not see granular detail.
Bundled pricing. The invoice shows: “Contract review — £1,200.” AI involvement is disclosed in the engagement letter but not itemised on the invoice.
Most firms land on category disclosure as the right balance. Clients get transparency without information overload. For firms that want to go further, transparent AI billing practices build even stronger client relationships.
Frequently Asked Questions
Should you charge clients for AI agent time?
Yes. AI agent work has real costs — tokens, compute, orchestration, and human oversight. Firms should charge based on value delivered and actual costs incurred, not human-equivalent hourly rates. Always disclose AI involvement to clients.
Is it ethical to bill clients for work done by AI?
It is ethical when three conditions are met: the firm discloses AI involvement, the fee reflects real costs and value rather than human hourly rates, and the client receives the quality they expect. It is not ethical to charge human rates for undisclosed AI work.
How much does AI agent work actually cost?
Mid-tier AI agent operations cost between £2,500 and £10,000 per month for firms running 5,000+ tasks. Per-task costs range from £0.08 for simple operations to £4 for complex multi-step agent workflows. Human oversight adds 30–50% to total delivery cost.
Do clients expect to pay less when AI is involved?
Some do, some do not. A 2025 survey found 68% of professional services buyers accept AI-assisted pricing when disclosed upfront. Clients who understand AI economics focus on value delivered. Clients who do not understand it focus on perceived effort.
How should firms disclose AI usage to clients?
Disclose at three points: in the engagement letter (what AI will be used for), on the invoice (which tasks were AI-assisted), and in client reports (what the AI produced and what humans reviewed). Early disclosure builds trust; retroactive discovery destroys it.
What billing models work for AI-assisted professional services?
Three models work: cost-plus (AI costs + 30–50% margin), value-based (fee reflects outcome value to the client), and blended rate (single rate covering human and AI time). The right model depends on the service type, client sophistication, and competitive dynamics.
How do law firms handle billing for AI agent work?
Approaches vary by jurisdiction. Some bar associations require disclosure and prohibit billing human rates for AI work. Others leave it to firm discretion. Most law firms charge a reduced hourly rate for AI-assisted tasks or embed AI costs in flat-fee arrangements with full disclosure.
Keito tracks what your AI agents do, how long they take, and what they cost — so you can bill clients honestly. Start tracking AI agent time today.