Your Client Knows You Use AI. Now They Want a Discount.

A client asked me directly, three months ago: "You use Cursor and Claude Code — why should I pay your full rate?"

Fair question. Honest one. I could've gotten defensive and started explaining myself. Instead I asked: "Does a surgeon using a robotic system charge less for the operation?"

The conversation went somewhere different after that.

Why clients ask — and why they're not wrong to

In 2025, "do you use AI?" became a standard question in early project conversations. Clients read about Cursor, Copilot, Claude Code. The logic they land on is simple: AI writes code faster, faster means fewer hours, fewer hours means lower bill.

That's not malice. That's arithmetic.

It's a pricing question to them: tools get cheaper, work should follow. The formula makes sense for commodities. Software development — the expensive parts of it, anyway — isn't one.

Speed of delivery isn't the same as cost of a decision

When a client says "AI speeds you up, give me a discount," they're asking about speed. What they're actually paying for is judgment. For someone who knows what they're doing and takes responsibility for it.

AI is genuinely good at clear tasks. Write a component from a spec, reformat a data structure, scaffold a test. That does speed things up.

It's bad at something else. Figuring out whether the task is even the right one. Looking at a legacy codebase with ten years of history and saying "this is where you'll get hit in three months." Taking responsibility for an architectural decision that has to live for five years. That work didn't get cheaper because of Cursor.

When we migrated an e-commerce store with 28k SKUs to a headless architecture, Claude Code handled the routine parts. Standard components, API response reshaping. Delivery got faster.

But deciding what moves to Next.js and what stays on Bitrix — that was 40 minutes of live conversation with the client, three hours of API surface analysis, and one contested call on authentication that we worked through separately. AI wasn't in that room. Because a mistake there doesn't cost hours. It costs a rewrite a year later.

Good tools reduce the cost of execution. They don't reduce the cost of judgment.

Three questions I ask instead of arguing

When a client raises the AI discount topic, I don't push back. I ask questions.

First: "What do you actually want to happen faster?" Half the time the client is thinking about something else entirely — they want tasks to move without waiting, not a lower invoice. That's a process question, not a pricing question.

Second: "What's the riskiest part of this project for you?" I've never heard "the stuff that generates quickly." What I hear is something specific — a past launch that killed conversion, or "I just need this to hold up under load during the busy season."

Third: "What are you actually paying for?" If the answer is "hours of work," we have a model problem. That's worth naming and fixing. If the answer is "the outcome, and knowing you won't screw it up" — there's no problem at all.

Those three questions tend to redirect things without anyone getting defensive. It works because the client figures it out themselves.

What changed in how we price

After that first conversation I rewrote how we explain pricing at the start of an engagement.

Before: list of deliverables, hour estimates, hourly rate. Client naturally multiplied the hours.

Now: risk-based pricing — a description of the project risks, what each one costs if it materializes, what we specifically do to prevent them. Price is not for time. Price is for risk reduction. We mention AI tools in context: "this lets us be more precise on the detail work, we don't double-bill for boilerplate."

The client stops counting hours. They start thinking about what could go wrong and why it won't.

I wrote about how I build these estimates: I stopped estimating by hours. Here's how I price projects by risk now.

This framing connects to something I worked out earlier around tech debt conversations — how to talk about technical cost in the client's language, not ours. That piece is here.

When a discount is the honest call

It happens. If AI genuinely changed the scope — not just the speed, but the actual volume of work — revisiting the price is fair.

Say a task used to take 40 hours, and with AI it's 15. If we charge the same rate, we're just making more money for less work. Sometimes that's fine. But a client who notices it isn't wrong to say something.

The line: scope changed — we revisit. Tool got faster — we don't. The surgeon with the robotic system is still doing the surgery. Just more precisely.

Being straight about this matters more than winning the negotiation. A client who sees you engage honestly with the question doesn't go shopping around — they come back when the next project starts.

The takeaway

When a client asks this, they're not attacking your rate. They're asking you to explain something they don't understand yet.

If you can't explain it, that's worth sitting with. The problem usually isn't the question.

And this conversation is easier if you've already done the filtering work before the project starts. Here's how I spot a difficult client before the project begins.