The work went to the integrators. The money went to the factories.

The job closest to writing code to someone else's specification is shrinking. In the Bureau of Labor Statistics' employer survey, US computer programmers fell from 132,740 in May 2022 to 92,230 in May 2025, down 30.5%, and the bureau expects a further 7.3% decline by 2035 because companies will use AI to automate repetitive programming. Software developers, the broader occupation, grew 10% over the same three years.

What changed is who gets hired. In the first quarter of 2026, 69.3% of US software development postings on Indeed were for senior roles and 4.5% for entry level, the lowest entry share of any sector Indeed tracks. Stanford's Digital Economy Lab finds 22 to 25 year olds in the most AI-exposed occupations 19% below where they would otherwise be, through fewer hires rather than layoffs, and its authors call that a pattern, not a measured effect of AI. At the large tech firms SignalFire tracks, the share of hiring going to forward-deployed engineers, who embed with a customer to make a system work in place, is up 30% since ChatGPT launched. In May, Anthropic set up an enterprise AI services firm with Blackstone, Hellman & Friedman and Goldman Sachs to supply exactly those engineers.

So the work is going to judgment and integration. The money is not following it there. Gartner expects IT services spending to grow 5.3% in 2026 against 15.5% for software, and Accenture, the largest integrator, grew revenue 5% in its 2026 fiscal year.

The money AI is spending goes to buildings, chips, memory and power. Microsoft, Alphabet, Amazon and Meta paid about USD 295 billion for property and equipment in the first half of 2026, up about 84% on a year earlier, by this research's sum of their filings. When Amazon raised its 2026 plan to USD 220 billion in July, its chief executive put the increase down mainly to the higher cost of memory.

That cost is someone's margin. Micron's quarter to 3 September brought revenue of USD 54.23 billion, almost five times the USD 11.32 billion of a year earlier, at a gross margin of 86.8%. GE Vernova holds 116 gigawatts of gas turbine orders and reserved slots, against output of 20 gigawatts a year. Wood Mackenzie put US lead times for power transformers at 128 weeks in mid-2025. The most valuable software business of the decade is waiting on memory fabs, turbine lines and transformer shops.

The economist Charles Jones gives the reason in one number: software spending is about 2% of US GDP, so even infinite amounts of what software does today would raise GDP by only about 2%. In his model, automation raises the price of what stays slow, and making goods is on his list of tasks that still need people.

This does not make hardware in general the next frontier. AI took 82.7% of US venture dollars in 2026 to the end of September. Figure AI is valued at USD 39 billion without disclosing revenue. US factories ran at 75.7% of capacity in August, below their long-run average, so machines as such are not scarce. The 2026 margins went to companies whose capacity took decades to build. A newcomer does not build a memory fab to join them.

What a small or mid-size product company lacks is access: qualified, scheduled, trustworthy capacity at small and changing volumes. Protolabs counted 6.1% fewer customers in 2025, partly because it chose to chase larger orders. Xometry, which routes work to about 5,000 suppliers, grew marketplace revenue 45% in the second quarter. Today's other field note shows the cost side of the same coin: a distributed network does not remove the fixed cost of a factory, it decides who carries it.

For Moduloa that cuts both ways. Its own factory OS and routing are software, and if code is cheap they are not a moat either. What cheap code cannot copy is knowledge that has to be gathered. Manufacturing know-how is scattered across thousands of small shops, and Moduloa's plan is to gather it into one shared framework, built in the open with contributors and with AI, where every production run adds to a record anyone can audit. Today that is one founder, one contributing company and the research published here. It becomes an advantage only if many more come.

P-14 in the register predicts a new profession, the robot-era production technologist, working across machines, robots, fixtures, software, quality and data. The software labor data are the nearest analogue: integration roles gain while entry-level hiring falls to 4.5%, and an industry that stops hiring juniors runs out of seniors. P-18, that the durable moat is the framework rather than the robots, gets a sharper test from the same evidence. If code is cheap, the framework's value has to sit in records, certification, training and the people who contribute to it, not in its software.

P-12, that a factory OS becomes central to flexible manufacturing, and P-15, that manufacturing capacity comes to work like cloud capacity, both get a caution: a layer that is mostly software is exactly the layer being repriced. Nothing here is scored and the register is untouched.

What would change the reading: entry-level software hiring climbing back well above 4.5%, Micron's gross margin falling back toward its 2025 level of about 40% as new fabs arrive, or the share of software tasks agents complete reliably stalling through 2027.

Google: Sundar Pichai shares news from Google Cloud Next 2026 ·Semafor: Google CEO says 75% of company's new code is AI-generated ·Anthropic Institute: When AI builds itself ·METR: time horizon benchmark results, version 1.1 (data file) ·BLS: Occupational Employment and Wage Statistics, national estimates 2022 to 2025 ·BLS Occupational Outlook Handbook: Computer Programmers ·Indeed Hiring Lab: The Labor Market Is Tilting Toward Seniority ·Stanford Digital Economy Lab: No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19% ·SignalFire: State of Talent Report 2026 ·Blackstone: Anthropic partners with Blackstone, Hellman & Friedman and Goldman Sachs to launch enterprise AI services firm ·Gartner: Worldwide IT spending to grow 14.2% in 2026, totaling USD 6.37 trillion ·Accenture: fourth quarter and full year fiscal 2026 results, Form 8-K ·SEC EDGAR: Alphabet, payments to acquire property, plant and equipment (one of the four series summed) ·Meta: second quarter 2026 results, Form 8-K ·Fortune: Andy Jassy said Amazon will spend USD 220 billion this year, and still won't have enough capacity to meet demand ·Micron: fourth quarter and full year fiscal 2026 results, Form 8-K ·GE Vernova: second quarter 2026 results, Form 8-K ·Turbomachinery Magazine: GE Vernova gas turbine backlog hits 116 GW as power orders more than double ·Wood Mackenzie and American Clean Power: Making the connection, executive summary ·Charles I. Jones, NBER w34779: A.I. and Our Economic Future ·PitchBook and NVCA: Q3 2026 Venture Monitor ·Figure: Series C ·Federal Reserve G.17: Industrial Production and Capacity Utilization ·Proto Labs: Form 10-K for 2025 ·Xometry via CIMdata: Xometry Reports Record Second Quarter 2026 Results ·Moduloa: Software is done. Hardware is next. ·Moduloa: Someone has to own the hall ·Moduloa: The robot was never the expensive part

The shares of AI-written code are the companies' own counts of lines, from companies that sell AI tools, and are not measures of productivity. METR's doubling time is from its published data file, at 50% reliability; at 80% reliability the best task length is about 3 hours, and METR no longer updates the tracker. Occupational counts are BLS employer-survey May estimates, which lag. The Stanford, Indeed and SignalFire findings are descriptive, and the Stanford authors caution against reading them as caused by AI. Equipment spending is this research's sum of cash paid for property and equipment in the four companies' filings, excluding finance leases. Micron's and GE Vernova's figures are their own filings; memory is a cyclical industry and Micron's margin a year earlier was 44.7%. Figure's valuation is from its own announcement and Bloomberg; no revenue has been disclosed. The claim that capacity in general is not scarce rests on the Federal Reserve's aggregate utilization rate, which can hide shortages in particular processes and regions.