AI has made software almost free to build, and that has created the vibe coding paradox: more apps are being made than ever, and fewer people seem to need them. In the first half of 2026 the App Store took in about 560,000 new apps, but downloads grew only about 2%.
A developer on the Chinese programming forum V2EX summed up the mood in a post on 20 October 2026. Clones of every kind of app appear daily, they wrote, “but how many have you used?” Their argument is that if an AI agent can do the whole job, you no longer care about the app at all.
Key takeaways
- Sensor Tower counted about 235,800 new App Store submissions in Q1 2026, up 84% year on year.
- The first half of 2026 brought roughly 560,000 new apps, close to the 600,000 for all of 2025. Downloads rose only about 2%, to 17.6 billion.
- Investors made a similar bet in February 2026, when fears about AI agents knocked an estimated $2 trillion off software stocks.
- The evidence suggests demand is moving from apps you operate to outcomes you ask for. Software isn’t vanishing.
What is the vibe coding paradox?
“Vibe coding” means building software by describing what you want in plain language and letting an AI model such as Claude Code, Codex or Replit write the code. The paradox is that the cost of making software has collapsed, but people’s attention, their phone storage and their need for yet another to-do app have stayed about the same.
Think of a town where everyone suddenly gets a free bakery oven. You get a lot more bread. You don’t get a lot more hungry people.
What do the numbers actually show?
The figures come from Sensor Tower and were first reported by The New York Times on 20 July 2026. AI Weekly’s summary calls Q1’s 84% jump the tracker’s biggest growth rate in four years. Apple World Today covered the 2025 baseline.
| Period | New App Store apps | Downloads |
|---|---|---|
| Full year 2025 | About 600,000 (up 30%) | 35.4 billion (up 3%) |
| Q1 2026 | About 235,800 (up 84%) | Not reported |
| H1 2026 | About 560,000 | 17.6 billion (up about 2%) |
Supply roughly doubled while demand barely moved. Apple has been blocking updates to, or removing, vibe-coding apps such as Replit, Vibecode and Anything under its rule against apps that change behaviour after review. Apple says it is enforcing its existing guidelines rather than going after a whole category.
Is the V2EX poster right that software demand is gone?
They’re half right. Their core point is sound. GUIs, themes and polished menus all exist to help a human do a task faster. If an agent does the task for you, that layer matters less. You type a sentence and wait for the result.
The same logic hit Wall Street early in 2026. Anthropic released 11 Claude Cowork plugins on 30 January 2026, and investors began to fear that agents would replace paid software seats. Philipp Dubach tracked the selloff: roughly $2 trillion in software market value disappeared in early February, and the IGV software ETF fell about 32% from its September 2025 peak.
Dubach’s counterpoint is the part worth keeping. Software sector earnings were still on track for about 17% growth in 2026. Goldman Sachs projects application software reaching $780 billion by 2030, with agents making up much of it. The demand didn’t disappear. It moved into agents.
Which kinds of software are actually at risk?
Not all software is equally exposed. Here’s a rough split, based on Dubach’s “systems of record vs systems of engagement” framing:
- Most exposed: thin apps that wrap one task, such as converters, simple trackers, basic analytics dashboards and tier-1 support tools. An agent can do these directly or rebuild one in an afternoon.
- Changing shape: seat-priced SaaS. Vendors are moving to usage-based or outcome-based pricing. Our look at Stripe’s SaaS survey covers how fast that’s happening.
- Least exposed: systems of record, meaning the databases, ERP, security and observability tools that agents themselves depend on. An agent still needs somewhere trustworthy to read from and write to.
My opinion: the V2EX post describes the end of the app as a product you open, not the end of software. The code is still there. It has moved behind the chat box.
What does this mean for developers?
Optimists cite the Jevons paradox: when something gets cheaper, people tend to use more of it. CTO coach Stephan Schmidt pushes back. He argues the extra demand shows up as more ideas and more software, not necessarily more developers. His summary is “more software, fewer developers”.
The hiring data points the same way. Anthropic’s own junior hiring fell to 9% as its agents took on longer tasks. What still pays is the work a prompt can’t do: knowing which problem is worth solving, wiring agents into real systems safely, and owning the result when it breaks. Our guide to production-grade AI agents shows how much engineering that takes.
What happens next
- The app glut gets filtered. At its first-half pace, the App Store could pass one million new apps in 2026, above the 2016 record of about 890,000. Expect tougher review and more takedowns.
- Distribution beats building. When anyone can ship, having users and trust becomes the scarce part.
- Agents become the main way people use software. More products will expose APIs and tools for agents to call, instead of screens for people to tap.
- Reality checks are coming. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, so the agent-does-everything future will arrive unevenly.
FAQ
Is AI killing the demand for software?
No. It’s reducing demand for simple standalone apps. Demand is moving to agents and to the systems of record they rely on. Software sector earnings were still forecast to grow about 17% in 2026.
How many new apps were released because of vibe coding?
Sensor Tower estimates about 560,000 new App Store apps in the first half of 2026, nearly the full 2025 total. Vibe coding is widely seen as the main driver, but Apple hasn’t confirmed that.
Should developers still build apps?
Yes, but build for a real, specific problem and plan how you’ll reach users before you write code. Building is no longer the hard part. Getting noticed and staying reliable are.
Next step: before you prompt your next side project into existence, write one sentence on who will use it and why an agent can’t just do the job for them. If you can’t, skip the build.
2 comments