← All essays
Han Feizi2026-09-23By Anderson#han feizi#ai discount#freelance pricing#ai and freelancers#legalism#solo business owner

Client Wants an 'AI Discount'? What Han Feizi Would Say

宋人有耕田者，田中有株，兔走觸株，折頸而死，因釋其耒而守株，冀復得兔 — A farmer of Song was tilling his field. A rabbit ran into a stump, broke its neck, and died. He put down his plow and stood watch over the stump, hoping for another rabbit. (《韓非子·五蠹》, literal translation)

Client Wants an 'AI Discount'? What Han Feizi Would Say

No — not automatically. A client asking for an "AI discount" is applying an old rule to a changed situation: value equals hours spent. The two largest freelance-platform studies on this say otherwise. AI work isn't getting uniformly cheaper. It's splitting hard into two prices.

Ask r/freelance. A recent thread — "AI has completely changed freelance work and honestly, I'm struggling" — opens with the exact question behind this article: "Are clients asking you to use AI instead of hiring you?" The same pressure has already reached the Big Four. It has a name now: the "AI discount."

Key Takeaways

  • An "AI discount" request treats value as hours × rate — a rule two labor-market studies show no longer holds evenly.
  • Generic AI output got 13% cheaper per contract in 2026 even as volume grew 90% (Upwork).
  • Judgment-added AI work paid 34% more per hour, same year, same platform.
  • Han Feizi's oldest chapter: what counts as valuable shifts by era, and clinging to the old rule is what makes you a laughingstock.

Why are clients suddenly asking for an "AI discount"?

Because the request sounds reasonable, and because a household name did it first. In June 2025, PwC's chief AI officer Dan Priest told Bloomberg that clients had started saying, in his words, "we want our fair share of those efficiencies." PwC obliged. It cut fees as AI sped up systems-integration work by roughly 30%.

So a client asking their freelance designer for the same thing isn't being unreasonable, not by their own lights. They're pattern-matching to a story they read in the business press. Here's the part they skip: those AI-driven discounts "plateaued," Priest said, once the same AI started improving quality, not just speed. Speed-based pricing collapsed into something else the moment the work stopped being pure time-saving.

Does AI actually make freelance work worth less?

Not evenly. For a large share of it, the opposite. Ozge Demirci, Jonas Hannane, and Xinrong Zhu tracked job postings on a major freelancing platform for eight months after ChatGPT launched, and published the result as a CESifo working paper later accepted at Management Science. Postings for automation-prone writing and coding work fell 21% relative to manual-intensive jobs. Image-related postings fell 17% after image-generating AI arrived.

Here's the sentence clients skip past: the jobs that didn't disappear "are of greater complexity and offer higher pay." Fewer automatable postings didn't mean cheaper automatable work. It meant the remaining work was the harder half — the half a generic AI discount can't touch, because it was never priced on raw hours to begin with.

Who actually loses money when AI enters a market — the weakest freelancers or the best ones?

Counterintuitively, the best ones — at least in the short run. Xiang Hui, Oren Reshef, and Luofeng Zhou studied the same window in Organization Science. Washington University's Olin Business School summarized the finding plainly: writing freelancers saw monthly jobs fall 2% and earnings fall 5.2%. Image freelancers fell 3.7% and 9.4%. But the group hit hardest wasn't the weakest performers. It was "more highly skilled workers... as defined by such attributes as the number and skill level of past jobs, hourly rate, and Upwork's 'top-rated' badges."

That's backwards from what an "AI discount" assumes. The tool didn't make everyone equally faster and therefore equally cheaper. It helped lower-skilled freelancers close the gap on output, which pushed the bottom of the market up and made it harder for top performers to stand out on volume alone. Your price problem, if you have one, was never really about your speed.

Is all AI-assisted work getting cheaper, then?

No — it's splitting. Upwork's Future Workforce Index 2026, released July 14, 2026 from a survey of 2,400 U.S. skilled workers plus platform data, found freelancers doing AI work earn 34% more per hour than those who don't. But in the same report: "generative AI and creative production work saw 90% year-over-year growth in contract starts while per-contract earnings declined 13%."

Read both numbers together and the shape is unmistakable. Generic, do-it-for-me AI output — the commodity end — got more abundant and cheaper. That's exactly where a discount request makes sense. Judgment-added AI work, what Upwork calls the "AI Orchestrator" role, got scarcer in relative terms and paid a premium. It connects tools to domain expertise and turns execution into an actual outcome. A client's "AI discount" is only accurate for one of these two categories — and it's rarely the one they're asking about.

What would Han Feizi say about a client wanting a cut of your new tool?

He'd point at a farmer standing in a field, waiting. In 《韓非子·五蠹》 ("The Five Vermin"), Han Feizi opens with the story in the hero quote above. A rabbit runs into a tree stump and dies. The farmer, who did nothing but watch it happen, quits plowing and waits by the stump for the next free rabbit. It never comes. He becomes a laughingstock in his own state. Han Feizi's target was rulers still governing by the customs of ancient sage-kings. But the shape of the mistake is exactly a client assuming last year's pricing logic still describes this year's work.

His argument for why runs in the same chapter: 故曰：世異則事異...故曰：事異則備變 — "when the age differs, the affairs differ... when affairs differ, the preparations must change." Then this: 上古競於道德，中世逐於智謀，當今爭於氣力 — in high antiquity, people competed on virtue; in the middle ages, on cleverness; now, on raw strength. His point wasn't that any one basis of value is correct. It's that the basis keeps moving. The farmer clinging to the old harvest and a ruler clinging to the old rites make the identical error: mistaking a rule that once worked for a rule that still does.

Han Feizi even diagnoses why the basis moves. A few lines later, in the same chapter: population grows while goods stay fixed. "The people multiply while goods are scarce... so they compete, even when rewards are doubled and punishments stacked" (人民衆而貨財寡，事力勞而供養薄，故民爭). Supply and scarcity redraw the competition, not virtue or vice. That's precisely the mechanism in the Demirci study. Postings didn't vanish because writing got easy. They thinned because AI-abundant supply flooded the commodity end, leaving the scarcer half — judgment, orchestration, the stuff AI can't yet do alone — to compete for a shrinking pool of buyers who still needed it. That's why it got pricier, not cheaper.

So who's actually the one still waiting by the stump — you or the client?

Possibly either. A freelancer who bills every hour exactly as before, ignoring that some of what they sell is now commodity-priced, is 守株 too — defending a harvest that's already changed. But a client who assumes "faster tool, therefore lower price" is running the identical fallacy the other way. They're treating your value as if it still equals time-on-task, and the two studies above show that's the one dimension AI has scrambled the most. Neither side gets to keep the old rule just because it's comfortable.

The useful question isn't "did AI help you." It's which side of the split the actual task sits on.

Is the work...What the data saysWho has the leverage
Generic AI output — drafts, first-pass images, boilerplate+90% volume, −13% per contract (Upwork)Client — this genuinely commoditized
Judgment, orchestration, taste, accountability for the outcome+34% average premium; "of greater complexity, higher pay" postings (Demirci et al.)You — this is the scarcer half
Speed alone, no change in what's deliveredPwC's own discounts "plateaued" once quality entered the pictureNeither — speed-only pricing isn't stable either way

How is this different from raising your prices in general?

It changes a different trigger than the rest of this series. Each entry below moves one lever in a small business; this one is the only lever a client's argument, not your own decision, tries to move for you.

The questionWhat movesWho initiates it
Raising your pricesThe number on the invoiceYou
Scope creepWork delivered per projectThe client, gradually
Client concentrationWhere the income comes fromYou, structurally
Your first hireTotal capacityYou
Feast or famineWhen income arrivesNeither — timing
Niching downWhat you sell, and to whomYou
The AI discountYour price, argued from your toolsThe client, explicitly

That last column is the tell. Every other question in this series is a decision you make about your own business. This is the only one where someone else hands you their theory of your value and expects you to price against it.

What do you actually say when a client asks?

Three moves, in order, and the first one does most of the work.

  1. Separate the task from the tool. Ask what specifically changed for them: turnaround time, revision count, or the actual deliverable. If nothing about the deliverable changed and only your speed did, you're being asked to discount your margin, not your output. That's a different, harder-to-justify request.
  2. Name which half of the split you're in. If the task really is commodity execution — a first-pass draft AI could produce with light editing — say so, and price it there. Upwork's own data shows that segment is genuinely getting cheaper; holding firm just loses you the client to someone who'll concede it. If the task requires you to judge, decide, and stand behind the outcome, that's the 34%-premium half. The tool you used to get there is irrelevant to what you're charging for.
  3. Don't let "AI discount" become the default response to being fast. Han Feizi's diagnosis, stripped of its politics, warns against solving a new problem with an old rule just because the old rule used to work. Concede every efficiency gain as a discount, and you've adopted a policy that scales against you forever. Every future tool becomes another pay cut.

Where the analogy stops

Han Feizi wrote for an absolute ruler, one with the power to unilaterally set terms and enforce them. He wasn't writing for two parties negotiating in a market, each with the standing option to walk away. His actual prescription for rulers — matching a subordinate's claim strictly against their delivered result, 循名責實 — cuts against you as much as for you. Claim the judgment-and-orchestration premium, and you need to show the outcome that earned it, not just assert that you used a fancier process. The chapter doesn't license freelancers to declare their own worth by fiat any more than it licenses clients to declare a discount by fiat. It's a standard for both sides, not a shield for one.

One more limit, honestly stated. The 21%, 17%, 2%, 5.2%, 3.7%, and 9.4% figures above all come from the eight-month window right after ChatGPT and image-generating AI launched in 2022–2023 — a shock, not a steady state. The Upwork 34%/90%/13% figures are from 2026, three years into adjustment. Markets keep re-sorting. Treat this as the shape of the pattern, not a number to quote back verbatim in a client email.

Curious what he'd actually say about your specific client conversation? Ask Han Feizi directly, or read his full profile first. He's also weighed in on what happens when a manager takes credit for someone else's work and why incentive programs backfire — same instinct for matching claims to actual results, aimed at a different target.

FAQ

Should I give clients a discount because I use AI?

Not automatically. It depends on which half of the work the request touches. Generic, commodity AI output genuinely got 13% cheaper per contract in 2026 even as its volume surged (Upwork Future Workforce Index). Judgment-added AI work, in the same report, paid 34% more per hour on average. Ask what specifically changed about the deliverable before conceding anything.

Is AI making freelance work cheaper overall?

No — it's splitting it. Two separate studies of freelance-platform data (Demirci, Hannane & Zhu on job postings; Hui, Reshef & Zhou on jobs and earnings) found the same shape. Automation-prone categories shrank in volume, but the surviving postings paid more. What was left was the harder, less-automatable half of the work.

Why did top-rated freelancers lose more income than average after ChatGPT launched, not less?

Because generative AI raised the floor more than the ceiling in the short run. It helped lower-skilled freelancers close the output gap on the top performers, compressing the market from below. Hui, Reshef, and Zhou's Organization Science study found more highly skilled, top-rated freelancers saw larger drops in jobs and earnings than less-skilled ones did, in the eight months after generative AI tools launched.

What does the "守株待兔" (guarding the stump) story from Han Feizi mean?

A farmer watches a rabbit accidentally run into a stump and die, then quits farming to wait by the stump for another free rabbit — and becomes a laughingstock. Han Feizi used it to attack rulers who governed by outdated customs. The same shape of error applies to anyone, freelancer or client, who assumes a pricing rule from old conditions still holds once the conditions — in this case, what AI actually commoditizes — have changed.

What is an "AI Orchestrator" and why does it pay more?

Upwork's term for a freelancer who connects AI tools to real domain expertise, judgment, and accountability for the business outcome — as opposed to simply generating raw AI output. Their 2026 Future Workforce Index found this category earned a 34% average hourly premium over freelancers not incorporating AI, even as commodity generative-AI execution work got cheaper per contract.

Sources and a note on the text

On the text and the translation. The Chinese source for 《韓非子·五蠹》 is quoted from the Chinese Wikisource edition linked below. English renderings here are my own and deliberately literal; Burton Watson's Han Feizi: Basic Writings (Columbia University Press, 2003) renders the chapter title as "The Five Vermin" and remains the standard scholarly English translation, alongside W. K. Liao's 1939 complete edition. Reddit quotations are reproduced as written.

Client Wants an 'AI Discount'? What Han Feizi Would Say · Hundred Masters