There is a specific, sinking feeling that a lot of skilled people have had in the last two years. You open a tool, type a paragraph of instructions, and watch it produce in ninety seconds something that used to be half a day of your billable time. The output is not perfect. But it is a genuinely good first draft, and your client could have generated it themselves. The uncomfortable question follows immediately: if the machine did most of the work, what exactly am I charging for now?
That question is not going away, and answering it badly is expensive. Price the way you always have and you will slowly be undercut by people willing to resell AI output at a discount. Panic and slash your rates and you train your best clients to see you as a commodity. The way out is not to work faster or defend the old rate card. It is to change the thing you are pricing.
TL;DR
- Research from OpenAI (the “GPTs are GPTs” study, published in Science) finds that around 80% of the U.S. workforce could have at least 10% of their tasks affected by large language models, and about 19% could see at least half their tasks affected. The disruption is broad but shallow for most, deep for some.
- McKinsey estimates that generative AI plus existing automation could handle activities absorbing 60 to 70% of employees’ time today, and add the equivalent of $2.6 to $4.4 trillion a year in value.
- Almost all of that pressure lands on execution, the “how,” not on judgment, the “what” and “whether.” Execution is being repriced toward zero. Judgment is not.
- Pricing your time or your effort ties your fee to the exact thing AI is deflating. Pricing your judgment, taste, accountability and outcomes ties it to the scarce 20%.
- The original matrix below shows where value migrates for eight common expertise services, and a CEO-style audit helps you reprice deliberately rather than reactively.
The 80% is real, but read it carefully
The headline number that gives this article its title comes from real, citable research, and it is worth stating precisely so you do not over-read it. In “GPTs are GPTs,” a team led by Tyna Eloundou at OpenAI evaluated hundreds of occupations against the tasks that make them up. They found that roughly 80% of the U.S. workforce could have at least 10% of their work tasks meaningfully affected by language models, while about 19% could see 50% or more of their tasks affected. When you add the software and tooling built on top of the raw models, the share of tasks that can be done substantially faster rises to somewhere between 47 and 56%.
McKinsey’s June 2023 analysis points the same direction from a different angle. It estimates that the activities absorbing 60 to 70% of the average employee’s time are now technically automatable with generative AI and adjacent tools, up from around 50% in earlier estimates, and that the productivity unlock is worth $2.6 to $4.4 trillion a year across the use cases it studied.
Here is the part most people miss. “Affected,” “exposed” and “automatable” describe tasks, not whole jobs, and they describe execution, not judgment. Both studies are measuring the same thing: how much of the doing can now be done by a machine. Neither is measuring how much of the deciding can. That gap is the whole basis of your new price.
Why hourly and effort-based pricing breaks first
Most expertise is still sold on one of two logics: time (an hourly or daily rate) or effort (a fixed fee scoped to how much work the deliverable takes). Both were always slightly dishonest, because clients never actually wanted your hours. They wanted the result your hours produced. For decades that distinction did not matter, because hours and results were tightly coupled. More hours, better deck. More effort, cleaner code.
AI severs that coupling. When the deck takes twenty minutes instead of six hours, an hourly fee punishes you for using the better tool, and a scoped-by-effort fee shrinks because the effort shrank. Every efficiency you adopt quietly lowers your own price. This is the trap: the more you let AI do the 80%, the less your time-and-effort pricing lets you charge, even though the value you deliver has not fallen at all. The client still gets a great campaign, a sound contract, a working feature. You have simply repriced yourself downward by billing for the input that collapsed instead of the output that did not.
A CEO would never price a product by how many labor-hours went into it once a machine took over the assembly line. They would price it by what it is worth to the buyer. That is the move. A student would notice that the tool changed and update the model rather than defend the old one. That is the mindset.
The Pricing Basis Migration Matrix
The table below is an original editorial synthesis. The service list is standard; the “AI now commoditizes” and “the 20% you still price” columns are our interpretation of where value moves, informed by the Eloundou and McKinsey findings on which task types are most exposed. It is a decision aid, not a market quote. Read each row as: what part of this service just got cheap, and what part did not.
| Expertise service | Old pricing basis | What AI now commoditizes | The 20% clients still pay a premium for | New pricing basis |
|---|---|---|---|---|
| Copywriting and content | Per word or per hour | First drafts, variations, basic SEO copy | Positioning, brand voice, knowing what not to say | Per outcome (conversion, launch, authority) |
| Brand and graphic design | Per deliverable or per hour | Layout options, image generation, resizing | Concept, taste, brand coherence over time | Per engagement or brand retainer |
| Bookkeeping and accounting | Per hour or per return | Data entry, categorization, reconciliation | Interpretation, risk flags, tax judgment | Per advisory relationship |
| Legal drafting | Billable hour | Template contracts, first-pass clauses, research | Risk allocation, negotiation, sign-off liability | Per matter or fixed-fee with judgment premium |
| Strategy and consulting | Day rate | Slide production, data summaries, benchmarking | Framing the real question, the call under uncertainty | Per decision or outcome-linked fee |
| Software development | Per hour or per sprint | Boilerplate, tests, first-pass functions | Architecture, tradeoffs, owning what ships | Per shipped outcome or product ownership |
| Market and user research | Per project hour | Survey drafting, transcript summaries, desk research | Choosing what to study, reading weak signals | Per insight that changes a decision |
| Financial advising | Percent of assets or hourly | Portfolio modeling, report generation | Behavioral coaching, accountability, judgment in a crisis | Per relationship and outcome |
The pattern is consistent across every row. The commoditized column is execution: producing the artifact. The premium column is judgment: deciding which artifact is worth producing, whether it is right, and standing behind it when it matters. Your fee should sit on the premium column, because that is the column AI is making scarcer, not cheaper.
Pricing the 20%: what actually survives
If execution is deflating, what specifically are clients still paying for? It is not mysterious, and naming it precisely lets you price it.
- Judgment. Deciding what is worth doing at all, and what to ignore. When generation is free, selection becomes the bottleneck, and bottlenecks are where money accumulates.
- Taste. The trained sense of which of ten good options is actually the right one. AI produces plausible; a professional supplies discernment.
- Accountability. Someone has to own the result. A model cannot be fired, sued, or trusted with a reputation. A human who signs their name to the outcome carries risk the client is happy to pay to offload.
- Context and relationship. Knowing this client, this market, this history. The model starts from zero every session; you do not.
- Speed of good decisions. Not speed of production, which is now free, but speed of the right call under uncertainty, which is not.
Notice that these map almost exactly onto the skills the World Economic Forum’s Future of Jobs Report 2025 lists as rising, analytical thinking, leadership, and the judgment-heavy human skills, while the tasks it lists as declining are the standardized, execution-shaped ones. The market is already repricing in the direction this article describes. Value-pricing advocates such as Ron Baker have argued for years that professionals should charge for the value they create rather than the hours they log. AI simply turned that long-standing best practice into a survival requirement.
Four pricing architectures, ranked by AI resilience
Once you accept that you are pricing judgment rather than execution, the mechanism you use to set the fee matters. The comparison below is an editorial rating, not a guarantee; the point is directional, showing how tightly each model is coupled to the collapsing 80% versus the durable 20%.
| Pricing architecture | What the client pays for | AI resilience | Best when |
|---|---|---|---|
| Hourly or day rate | Your input time | Low | Almost never now; time is the thing AI deflates |
| Fixed fee, scoped by effort | The deliverable’s size | Low to medium | Simple, well-defined output with little judgment |
| Value or outcome-based | The result achieved | High | The outcome is measurable and you influence it |
| Judgment retainer or access | Ongoing decisions and accountability | High | The client needs a trusted call over time, not an artifact |
The migration is from the top of that table to the bottom. Every step down moves your fee off the input AI is cheapening and onto the outcome and judgment it cannot supply. You do not have to leap straight to outcome pricing on day one. But every proposal you write should move at least one notch down the table.
A CEO-style repricing audit
Treat this like a pricing committee reviewing a product line, not a freelancer nervously eyeing a rate card.
- Split your last five deliverables into execution and judgment. For each, honestly estimate what share the machine could now produce versus what share required your call. The execution share is what you must stop selling by the hour.
- Find the moment the client actually exhaled. In every engagement there is one decision, insight or reassurance that was the real value. That is your 20%. Name it.
- Rewrite one offer around that 20%. Move a single service from hourly or effort-scoped to outcome or judgment-based. Price the result, not the production.
- Let AI do the 80% openly. Use the tools aggressively to compress execution, and pass the speed to the client as responsiveness, not as a discount. Faster is a feature; cheaper is a trap.
- Set a falsifiable test. Decide in advance what would tell you the new price works: a proposal accepted at the higher number, a client who renews the judgment retainer, a referral that mentions the outcome rather than the deliverable. Review in 90 days.
The goal is not to charge more for the same thing. It is to charge for a different thing, the thing that did not get commoditized, and to stop discounting yourself every time you adopt a better tool.
Your next move
Pick one service you sell and one recent client. Write down, in a single sentence each, the execution you performed and the judgment you supplied. Then draft one new proposal that prices the second sentence and gives the first away as speed. That is the whole transition in miniature. A CEO reprices the moment the cost structure changes rather than waiting for the market to force it. A student assumes the model will need updating again next year and keeps the audit habit. Do both, and the 80% AI can do becomes the reason your 20% is worth more, not less.
Frequently asked questions
Can AI really do 80% of knowledge work?
Not of any single job, but of many tasks across most jobs. The OpenAI “GPTs are GPTs” study found about 80% of U.S. workers could have at least 10% of their tasks affected, and roughly 19% could see half their tasks affected. It is a statement about tasks and execution, not about entire roles or about judgment.
Should I lower my prices because AI made the work faster?
Usually no. If you lower prices in proportion to your own efficiency gains, you hand the entire value of the tool to the client and train them to see you as a commodity. Reprice onto the outcome instead, and treat your new speed as a service feature rather than a discount.
What does value-based pricing actually mean here?
It means charging for the result the client gets rather than the hours or effort you spend. If a piece of work reliably produces a measurable outcome, a launch, a conversion lift, a risk avoided, the fee attaches to that outcome. This idea predates AI; writers on professional-services pricing like Ron Baker have argued it for years. AI just made it urgent.
How do I justify a high fee when a client could use AI themselves?
By pricing the part they cannot get from the tool: the judgment about what to do, the taste to pick the right option, and the accountability of a professional who owns the result. The client can generate a draft; they are paying you to know which draft is right and to stand behind it.
Which pricing model is most future-proof?
Outcome-based fees and judgment retainers, because both are tied to results and decisions rather than to production time. Hourly and strictly effort-scoped fees are the most exposed, since they are coupled to exactly the execution AI is cheapening.
Does this apply to employees, not just freelancers?
Yes, in a translated form. An employee’s “price” is their compensation and their security, and the same logic holds: the parts of your role that are pure execution are exposed, and the parts that are judgment, ownership and coordination are appreciating. Reallocating your hours toward the second is the internal version of repricing.
Sources
- Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, “GPTs are GPTs: Labor market impact potential of large language models,” published in Science (2024); working paper 2023 (about 80% of the U.S. workforce could have at least 10% of tasks affected, roughly 19% could see 50% or more affected, and 47 to 56% of tasks completed faster with LLM-based tooling).
- McKinsey Global Institute, “The economic potential of generative AI: The next productivity frontier,” June 2023 (activities absorbing 60 to 70% of employees’ time now technically automatable, up from about 50%; estimated $2.6 to $4.4 trillion in annual value across 63 use cases).
- World Economic Forum, Future of Jobs Report 2025 (rising human-centric skills such as analytical and creative thinking, leadership and AI literacy; declining demand for standardized, execution-shaped tasks; 39% of core skills expected to change by 2030).
- Ron Baker, Implementing Value Pricing (the case for charging professionals’ fees on the value created rather than hours logged).
- Adrian Slywotzky, Value Migration (the framework for how economic value moves away from outdated business designs toward new ones).
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
