Essays2026-11-05
Who Gets Paid When Software Is Cheap
More people build software than ever, and the money is moving to a few.
- Published
- 2026-11-05
Summary
- AI has made code cheap enough that software now ships faster than anyone can buy it, and the money is going to a few makers while the median falls.
- The first developers to lose out are the ones paid to turn someone else’s spec into code, and people in other fields now decide what to build and often make the first version themselves.
- The developers who stay on top hold a specific customer’s problem and the number that measures it, and that is where a working developer should move.
Code is no longer expensive
Picture a mid-sized distributor a few years ago. It wanted one system for its stock, orders, delivery and customers. An agency quoted hundreds of thousands of dollars and a year of work. The alternative was an off-the-shelf ERP, which meant reshaping the company’s work to fit someone else’s software. It wasn’t cheap either. Most companies like it did neither and kept running on spreadsheets.
That price was a wall, and the people who wrote code lived behind it. Because software was expensive to make, the few who could make it were paid well, even for work that was mostly translating a clear request into code.
Coding agents changed that. One developer working with AI can now do work that used to be split across a team. That is what lowered the wall, not cheaper agencies. An agency sells people by the month. In Korea, public software is still priced that way, by how many people a vendor puts on a job and how experienced they are, and the industry itself says that as AI raises development productivity, the basis for that price collapses. Software agencies are caught in the same trap as the law firms and call-center outsourcers that bill by hours and headcount: the faster AI makes the work, the less they can bill. What has changed is who can take the job.
Go back to that distributor. Today one developer working with AI can plan, design and build the same system alone, for far less than the agency’s quote. The company gets a system made for the way it already works, at a price it can justify. The developer, who might once have drawn a salary at that agency, can earn far more from a few clients like this in a year than the payroll ever paid. The agency does not get cheaper. It just stops getting the work.
Where software is easiest to count, in consumer apps, the jump in volume is plain. In the first half of 2026, the App Store took in about twice as many new apps as a year earlier, while downloads barely moved.
When supply grows that fast and demand does not, the money spreads thin. Solo founding is at a record high on Stripe, yet the median solo founder earned less in their first months than a year before, while the top tenth earned more.
Software is starting to pay the way books, music and games always have: many makers, a thin middle, a few big winners. The ability to produce code is no longer what the market pays a premium for.
The developer paid to turn specs into code goes first
The same thing is happening inside companies, and it starts with one kind of developer: the one who receives a well-defined ticket from a large team and turns it into code.
Amazon’s CEO wrote that six engineers built a new engine for Bedrock in 76 days, work that might normally take a team of about 40 people a year, and that Amazon flattened its organization to decide faster. Coinbase said it would cap management at five layers below the CEO and reorganize into small “AI-native pods.” The developer who was one part in a large machine is the part these companies need least.
Hiring points the same way, and the entry door is closing fastest. In Q1 2026, entry-level roles were 4.5% of US software development postings on Indeed, and senior roles were 69.3%. In Korea, software workers with under three years’ experience fell while experienced ones grew.
Teams are thinning with it. In Stack Overflow’s 2026 survey, the share of developers working in an organization of one rose from 4% to 10% in a year. Work that used to be split across several people now sits more and more with one person and an AI. The era of the developer who sat at a desk debating algorithms, or turning a clean spec into clean code, is ending.
People in the field now decide what to build
While that role shrinks, the decision about what to build is leaving IT.
When Anthropic studied how people work with its coding agent, what raised the success rate was knowledge of the field, not coding skill. The fastest-growing users work in management, sales and law. These people decide what the software should do. Making it hold up in a running business is another matter.
Some companies are already acting on it. A 65-person water treatment company replaced Salesforce with its own CRM, built with a five-person startup for $15,000 to $20,000.
Developers enter this picture late, at the point where things break. On Kmong, Korea’s freelance marketplace, requests for help with vibe-coded software, mostly reviews and partial fixes, grew more than sixfold in half a year. People got stuck at deployment, at data and at keeping the thing stable. By the time a developer is called, the customer has already defined the problem, owns the data and has chosen what to build.
The usual advice is that developers should learn to talk to people in other fields. That is not enough. A developer who wants to be paid well has to be in the room before that point, while the problem is still being defined.
The top holds a customer’s number
What the people at the top of these markets have in common, as far as the data shows, is not coding skill. It is closeness to a paying customer’s problem and the number that measures it.
Stripe’s top solo founders leaned toward selling to businesses, sold abroad from the start, and saw about 30% of their customers come back the following month, against 8% in the middle of the pack. They are people who watch who returns, who pays and why, and build toward that. AI gives such people wings, because reading data and running experiments now costs them little.
Small businesses show where that skill is needed. Korean SMEs say the hardest parts of bringing AI into their work are fitting a system to how they actually work and connecting it to the data they already have. Those are problems of knowing a business, not of writing code. Most of them would rather turn an existing employee into the person in charge than hire a specialist, so the developer who helps them is working with that employee, not replacing them. People are wary, too. In a Korean survey of workers in manufacturing, distribution, IT and finance, staff-level employees were more anxious about AI than executives, and the most common worry was losing jobs. A developer who wants to rebuild how such a company runs has to stay long enough to understand its stock, its customers and its people, and to earn their trust. That understanding is what the business pays for.
Where working developers are going
Work that turns someone else’s requirements into code is shrinking and unlikely to grow back, which is why developers who stay away from the customer’s problem are finding less and less ground to stand on. What sits above it is often something an experienced developer already has. Years on the job are spent inside some industry, with some set of customers, learning how a payment fails, why a shipment is late, or which screen loses a user.
The developers doing well have stopped selling the ability to turn a spec into code. They have tied their work to a field and to a number a customer watches: revenue, retention, cost per order. Some go inside one small business and rebuild how it runs. Some own one product’s growth inside a company. Others build a narrow product for one industry and watch every customer who leaves. AI-native companies go one step further and become the business itself. Crosby, founded by a former engineering lead at Ramp and a lawyer from Cooley, is a law firm that does only commercial contract review, and it is run on total turnaround time rather than billable hours. Harper’s founders first planned to sell AI software to insurance brokers, then became a licensed brokerage themselves. Some do this alone, as a one-person studio or agency, and others inside a company. The work is the same either way.
Software has become cheap. What has not become cheap is knowing which software a particular customer will pay for and keep using. That knowledge is where developers will be paid from now on.
References
- Digital Today, “AI arrives but rules stuck in SI era: SW industry urges network separation, pay system overhaul” (2026-04-13) https://www.digitaltoday.co.kr/en/view/47557/ai-arrives-but-rules-stuck-in-si-era-sw-industry-urges-network-separation-pay-system-overhaul
- 9to5Mac, “Report: App Store added nearly as many new apps in H1 2026 as in all of 2025” (2026-07-20, Sensor Tower data via The New York Times) https://9to5mac.com/2026/07/20/report-app-store-added-nearly-as-many-new-apps-in-h1-2026-as-in-all-of-2025/
- Stripe, “Solo founding is at an all-time high: Top performers have these traits in common” (2026-05-28) https://stripe.com/blog/top-solo-founder-traits
- Amazon, “Amazon CEO Andy Jassy’s 2025 Letter to Shareholders” (2026) https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-2025-letter-to-shareholders
- DL News, “Coinbase shares pop as CEO Brian Armstrong cuts 700 jobs and blames AI” (2026-05-05) https://www.dlnews.com/articles/markets/coinbase-cuts-jobs-and-blames-artificial-intelligence/
- 서울신문, “AI 시대, 소프트웨어 개발자 신입 일자리 소멸 중” (2026-01-06) https://www.seoul.co.kr/news/society/2026/01/06/20260106500238
- Stack Overflow, “The results of the 2026 Developer Survey are here!” (2026-10-06) https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here/
- Indeed Hiring Lab, “The Labor Market Is Tilting Toward Seniority” (2026-07-23) https://hiringlab.indeed.com/2026/07/23/the-labor-market-is-tilting-toward-seniority/
- Anthropic, “Agentic coding and persistent returns to expertise” (2026-06-16) https://www.anthropic.com/research/claude-code-expertise
- The Wall Street Journal (Dow Jones copy), “Meet the Companies Vibe Coding Their Own CRMs” (2026-03) https://www.itiger.com/news/2617554134
- 벤처스퀘어, “크몽, AI로 코드는 만들었는데 출시는 막혔다…’바이브코딩 후속 개발’ 반년 새 550% 증가” (2026-10-02) https://www.venturesquare.net/1117623
- 아시아투데이, “중소기업 AI 도입 필요성 71%…실행은 17% 그쳐” (2026-11-02) https://www.asiatoday.co.kr/kn/view.php?key=20261102010000537
- 스타트업엔, “실무자 3명 중 1명 “AI 도입이 더 불안”…가장 큰 이유는 ‘일자리 축소’” (2026-02-10) https://www.startupn.kr/news/articleView.html?idxno=56317