AI

"Software Is Over": When AI Makes Great Software Cheap, Is the Result Inflation or Deflation?

AnyPS5, Artcraft, and the strange economics of an economy with more to offer than anyone can consume.

By Kalemba Oct 20, 2026 Updated Oct 28, 2026 13 min read
#AI#software#economy

"Software Is Over": When AI Makes Great Software Cheap, Is the Result Inflation or Deflation?

AnyPS5, Artcraft, and the strange economics of an economy with more to offer than anyone can consume.

Two things happened in the last few weeks that, taken together, sketch out a future that many economists have been circling for a year but haven't quite seen land:

AnyPS5 — an open-source project — hit 100% coverage of the PS5's GPU shader instruction set, meaning it can now take a PS5 game's compiled code and rewrite it into a native Windows or Linux program that runs directly on your PC. Not emulation. Not a simulation of a console. The game, re-linked and re-pointed at new system libraries, running natively on hardware it was never shipped for.

Artcraft (not "aircraft" — it's the name of the project, from the studio behind an AI image tool) — built by one developer, Brandon Thomas, over essentially a weekend and a few weeks of follow-up — produced seven open-source clean-room reimplementations of Adobe's core apps: PhotoCraft (Photoshop), VectorCraft (Illustrator), FilmCraft (Premiere), LightCraft (Lightroom), EffectCraft (After Effects), DesignCraft (InDesign), and PdfCraft (Acrobat Pro). Written in Rust, generated with Anthropic's Claude Opus, MIT/Apache licensed, running in a browser via WebAssembly. Adobe charges $69.99/month for that suite. The clones cost nothing. Thomas, who said he'd been "bit by the dark pattern 'cancellation fee' one too many times," told Ars Technica the headline plainly: "Software is over."

Eric S. Raymond looked at the repos and wrote: "Adobe just got nuked. And closed source is dead, dead, dead." (The project's own framing is cleaner: clean-room reimplementation from observed behavior and public specs — no Adobe code involved — which is also, conveniently, a better legal position than Raymond's guess.)

Both projects are early. AnyPS5 still doesn't run most games reliably; the Artcraft apps won't hit 100% parity with Adobe any time soon, and "99% parity" is a different animal. But the direction is the story, not the current fidelity. The story is: the cost of producing a piece of professional-grade software is collapsing toward zero, while the quality floor rises. And that raises a question that is much bigger than whether Adobe's stock is in trouble:

When an economy can produce more goods and services than its members can use, what does that do to prices? Is the future inflationary or deflationary?

First, define the battlefield: good deflation vs. bad deflation

Let's be precise, because the word "deflation" has been ruined by a decade of central-bank fear.

  • Inflation is, in the standard CPI sense, a sustained rise in the price of a fixed basket of goods and services — usually driven by money/chasing or aggregate demand growing faster than supply.
  • Deflation is the mirror image: the supply of a category growing faster than demand for it, so the price of the basket falls.

But deflation comes in two varieties that feel opposite in the wallet:

  1. Good deflation (supply-driven). Productivity and output grow faster than demand; prices fall; real incomes rise. This is what happened with computers, TVs, and — until recently — software, for decades. The late 19th century's "Great Deflation" actually saw living standards climb as prices fell. A few years of -1% CPI in an expanding economy is benign, arguably welcome: more bang per buck.
  2. Bad deflation (demand-driven). Demand collapses; prices fall because people can't or won't buy; incomes fall too, and because most debts are nominal and fixed, deflation makes every debt heavier in real terms. This is the 1930s problem, and it's why central banks target +2%.

The AI question is which variety we're sliding toward — and the honest answer from the serious macro literature is: it depends on a race.

The race inside the machine

A 2026 macro-financial stress test paper ("Abundant Intelligence and Deficient Demand") formalizes the paradox neatly. Its core claim:

AI promises the largest productivity gain since electrification — but the channel through which that gain normally reaches aggregate demand is labor income, which is precisely what the technology displaces. They call the resulting phenomenon "Ghost GDP": output grows on the books, but the money doesn't circulate, because the people who used to get paid for the work are no longer getting paid.

There are two forces in perpetual competition inside that setup:

  • The displacement spiral: white-collar jobs — the highest-consumers in the top income quintile — get automated first. Nominal income compresses. Demand falls. Bad-deflation territory.
  • The deflation channel: AI crashes the cost of goods and services. Prices fall. Real wages and real consumption rise even with flat nominal pay. Good-deflation territory — and the paper concedes this is a "genuine competing mechanism" that can soften the crisis if product-market competition is vigorous and cost reductions pass through fast.

Software is where the deflation channel is winning, visibly, right now. And here's the kicker — it's not a hypothetical. It's in the earnings calls:

  • Goldman Sachs' Global Institute published "The Deflation of Software": if generative AI keeps making software cheaper to build, businesses will increasingly replace packaged software with AI-generated solutions, deflating the industry — and the authors flag it as "a precursor to the obviation of larger categories of knowledge work."
  • India's Big Four IT services firms — the purest business model of "selling the hours it takes to build and maintain software" — all reported it in spring 2026. HCL's CEO forecast revenue would dip 3–5% next year purely from what he called "AI deflation." TCS's CEO said the same force is already shrinking the business and gave it the most honest label of the decade: "degrowth." Wipro pointed to margin compression in fixed-price deals. The clients aren't canceling — they're getting the same work done for less, because the client (or a competitor of the vendor) is using AI to do some of the work themselves.

The services that used to be scarce are becoming abundant. Abundance, in every market that has ever worked, means falling prices. That's not a bug in the free-market engine. That's the engine.

So why are the Adobe- and Sony-clones a big deal, specifically?

Because software has a property most goods don't: near-zero marginal cost of reproduction, historically protected only by friction. For thirty years, the friction was the cost of building and maintaining professional software — which is why you could charge $69.99/month for something that costs you pennies to replicate on a disk. The moat was never the bits. The moat was the thousand engineers and the decade of R&D standing behind them.

AI collapses exactly that moat. AnyPS5 and Artcraft are the first visible dents, but note what they demonstrated about the structure of the threat:

  1. Clean-room reimplementation at scale. You don't need Adobe's source code. You need its behavior — observable in a public product — and a frontier model to write fresh code that reproduces it. The "identical copy" framing overstates fidelity and understates the mechanism: what matters is that the specification of professional software is now extractable, and the implementation is now cheap.
  2. One person, one weekend, seven products. The unit of software production has collapsed from "engineering org" toward "one skilled operator + model access." That is the same collapse the music industry went through when the unit of music distribution went from "record plant" to "a folder of MP3s" — except software's version, unlike Napster, improves the product's availability to everyone, including the professional users, and keeps improving, because the reimplementation is alive and iterates.
  3. The incumbent's countermove is the same AI. This is the part doom-takes underweight: Adobe, Microsoft, and Sony aren't passive victims; they have the same frontier models, plus distribution, brand, file-format gravity, enterprise trust, and ecosystem lock-in. Microsoft is in the odd position of being both the deflation agent (Copilot) and the biggest packaged-software monopoly on earth. The realistic outcome is less "collapse" and more margin compression, price wars, and consolidation — the software industry's version of what happened to airline fares, hotel rooms, and eventually to PCs themselves.

Historical caution in both directions: Java, PHP, and Linux "killed" a lot of expensive proprietary software and made nobody's life worse; Napster didn't kill the music industry, it just moved the revenue from 12-inch discs to 12-streaming-tennis-ball subscriptions. But the historical cases also understate the difference: music and hotel rooms are fixed-scarcity goods. Software is not. A PS5 port or a Photoshop clone doesn't displace another PS5 port — there's no natural limit on how much "good enough professional software" the world can hold. That's the condition the user's question describes: more services than can be consumed.

The end of scarcity in services — and what "too much stuff" actually means

This is where it stops being a tech story and becomes the macro story of the 2020s and 30s.

For a century, the Western economic baseline has been scarcity: things cost money because making them uses up something (labor, material, land). Inflation was the expected weather. Now a general-purpose technology is attacking the scarcest input — skilled cognitive labor — across an entire class of services: software, legal drafting, bookkeeping, design, copywriting, support, analysis. The marginal cost of a large share of professional output is trending toward the cost of electricity and a token.

When supply of a category can grow without bound but demand grows only with population and income, the category deflates, fast — and the software clones above are the canary. The same curve will bend under creative services, then professional services, then anything whose "product" is mostly arranged intelligence.

But here's the twist that makes this genuinely new, and genuinely dangerous, and it's the one the stress-test literature keeps landing on:

  • Deflation is only good if the income side holds up. The people who buy all that free PhotoCraft are the same people who used to be the Adobe engineers. If the deflation channel (falling prices → higher real income) and the displacement channel (falling nominal income → falling demand) run at the same time, the winner is determined by speed of adjustment — by how fast prices fall versus how fast wages and employment compress.
  • And because modern economies are leveraged on nominal obligations — mortgages, corporate debt, sovereign bonds — sustained deflation is a tax on everyone who signed a fixed number. A world of $0 software and 3% falling prices sounds like a paradise; run it through a debt ledger and it looks like the 1930s with a better GUI.
  • There's also the old 1930s "maladjustment" problem, which Keynes actually wrote about in his "Economic Possibilities for Our Grandchildren": an economy that succeeds at its core mission — produce enough for everyone — discovers the bottleneck has moved from production to distribution of purchasing power. "We can make everything cheap. Who is going to buy it?" is not a rhetorical question anymore; it's a policy question.

So — is it inflation or deflation?

The most defensible answer, putting the pieces together:

Sectorally, it's deflation — clearly, measurably, and already. "AI deflation" is now a line item in IT services earnings. "Software is over" is a working description of what's happening to packaged software's price-to-value ratio. Goldman calls it the deflation of software and warns it's the first domino of knowledge work. The AnyPS5/Artcraft episode is the consumer-facing flashpoint: professional-grade digital goods are about to face a supply curve that is effectively vertical (infinite) at a price that is trending to zero.

Aggregately, the honest answer is: it depends on the labor channel, and that's the part we can't see yet. If the productivity gains distribute broadly (as with electricity, which raised wages as it spread), you get a good-deflation boom: falling prices, rising real income, "the end of scarcity" as Keynes hoped. If the gains concentrate in the owners of capital and models while displacing the high-spending middle, you get Ghost GDP: a statistics bureau reporting record output over an economy whose demand base is quietly hollowing out — deflation with a debt crisis bolted to the bottom.

The same technology produces both outcomes. The difference is institutional: wages and their floor, the form of capital gains (productivity dividend vs. asset-price dividend), debt levels, and how fast central banks re-anchor from a +2% inflation target to whatever the price system actually does when half of professional work costs nothing.

The practical bottom line

For the person inside this economy — the one who just watched their industry's flagship product get one-shotted over a weekend — three things follow:

  1. The incumbents won't die the way Napster-era predictions said the music labels would. They'll compress: lower prices, AI-heavy delivery, fewer headcount, tighter margins, eventual consolidation. Adobe, Microsoft, and Sony CEA all have the same AI their challengers use — the question is organizational velocity, not capability. But "compression" still means real job loss and a step change in what these companies are worth.
  2. Scarcity is migrating. As software and cognitive services deflate, value pools move to the things that stay scarce: the physical (chips, datacenters, energy, logistics — which is why "AI is changing how work gets done" coexists with record capital expenditure on atoms), trust and verification, taste and curation, and status. When everything is free, what you use, and that you can be trusted to have made it, becomes the premium.
  3. The macro question is real and unresolved. An economy that can produce more professional services than its members can consume has to solve the distribution problem or it will "solve" it through falling demand, falling wages, and the arithmetic of fixed debt. Central banks have never had a mandate that covers "abundance." That's why this isn't just a tech story — it's the first serious stress test of the post-scarcity hypothesis, and the stress test is running in production.

The PS5 games and the Photoshop clones are just the leading edge. The actual event is that the price of a whole class of human effort is about to find a new, much lower floor — and whether that lands as the greatest deflation in history or the quietest demand collapse in history depends on choices being made right now, in budget meetings and policy debates that have far less drama than a Reddit launch thread.

Software is over. The economy that had to be built around that is just getting started.

Sources

Note: AnyPS5 and Artcraft are early-stage, unofficial projects; neither currently runs all target software at full parity. The "identical copy" framing from press coverage overstates current fidelity — the deflationary mechanism is the collapsing cost of reimplementation, not bit-perfect copying.

Join the conversation

Comments

0 approved

No approved comments yet. Start the conversation.

Comments are reviewed before publication. Please keep the conversation useful and kind.