The most interesting thing AI is making cheaper is not labor.
It is capability.
Those are related, but they are not the same.
The old startup constraint
When I started my first company, there were things we simply could not do yet.
Not because they were impossible. Because they required a person we did not have, a skill nobody on the team had learned, or enough money to buy access to someone who had.
Industrial design. Manufacturing knowledge. Regulatory work. Finance. Photography. PR. Performance marketing. Legal research. Analytics.
A startup was partly a sequence of waiting until you could afford the next capability.
That constraint shaped which ideas got attempted.
Productivity is the smaller framing
A lot of AI software is sold with a productivity argument: the same employee can do the same work faster.
That matters. It is measurable. It is easy to put in a business case.
But I think it undersells the actual change.
If a founder who cannot program can now prototype software, that is not just faster programming.
If a three-person company can analyze markets, draft contracts for attorney review, build financial models, create sales material, research competitors and launch a credible web presence without hiring a specialist for every first pass, that is not merely productivity.
The set of things the company can attempt has expanded.
Microsoft's 2026 survey found 58 percent of AI users saying they were producing work they could not have produced a year earlier; among the most advanced users, that rose to 80 percent.
MICROSOFT WORK TREND INDEX · 2026-05-05 · FROM EXPERTISE TO AGENCY
That is a capability story.
The threshold moved
Every business capability has historically had a minimum economic threshold.
You did not hire a data scientist to analyze fifty customers. You did not hire a full-time copywriter for one landing page. You did not commission a market research firm to answer a question worth $2,000.
Below the threshold, the work simply did not happen.
AI pushes the threshold downward.
That does not mean the AI does expert work perfectly. It means a much larger class of work becomes economically rational to attempt, inspect and improve.
That changes who can start
Stanford's 2026 AI Index estimates generative AI reached 53 percent adoption in roughly three years, faster than the personal computer or the internet.
STANFORD HAI · 2026 AI INDEX · ECONOMY
The important consequence for entrepreneurship is not that every founder suddenly becomes an expert.
It is that expertise becomes easier to approach.
A person with a real understanding of a customer problem can now cross more of the distance between insight and execution before needing capital.
That favors people who know something specific.
A nurse who understands a broken workflow. A contractor who understands estimates. A restaurant owner who understands an ugly supply problem. A teacher who understands where students get stuck.
The software gap between knowing the problem and building the first solution is getting smaller.
This is the part I care about with small business
A local business has historically been disadvantaged by the fixed cost of expertise.
A large company can have a marketing department, analyst, developer, designer and outside counsel. A ten-person company cannot.
I am building YG3 partly around the belief that marketing should be one of the capabilities whose minimum efficient scale collapses.
A small business should not need a department to have a serious website, understand how it is being discovered, publish useful material, run campaigns and maintain a coherent market presence.
That is not the same as saying software replaces every specialist.
It means the specialist gets called when the problem actually requires one.
Capital still matters. Judgment matters more.
AI does not erase the hard parts of building a company.
You can produce a thousand bad ideas faster. You can automate a broken process. You can create polished material for a product nobody wants.
The cheapening of capability raises the value of choosing what deserves to exist.
That is the part I think gets missed when people describe this only as a labor-saving technology.
When capability gets cheaper, execution stops filtering as many ideas out.
Judgment has to.
