How much software is being built in the AI era?

GitHub: the Octoverse report on its own platform

Octoverse is GitHub's yearly report on its own platform: a company counting itself, not an official statistic. The latest edition is dated 28 October 2025 and was updated on 28 February 2026. Its year, which it calls 2025, runs from 1 September 2024 to 31 August 2025.

GitHub puts new accounts 23% above the year before, and says 63% of all repositories were public.

So a developer is an account, and a repository counts whether or not anyone else can see it. An LLM SDK is a model provider's own library for calling its language models, so that row counts projects that use AI, not projects written with AI's help, which the report does not count. On AI assisted coding it says nearly 80% of new developers used GitHub Copilot, GitHub's own product, in their first week.

Stack Overflow: a survey of people who chose to answer

Stack Overflow's 2026 Developer Survey used 30,903 responses from 169 countries, recorded from 23 June to 5 August 2026. Not everyone answered every question: 17,464 answered the one on AI tools, "Do you currently use AI tools or AI agents in your role at work?", and they could pick several answers.

Stack Overflow sums that up as 83% using at least one AI tool. Of those who use a coding assistant or agent, 73.0% said they use it daily.

So this is not a sample of all developers. And it measures who uses a tool, not what anyone built.

Lovable: one company's own statement

Lovable sells a tool for building apps by chatting with an AI. In its funding announcement of 18 December 2025 it said more than 25 million projects had been created in its first year, and more than 100,000 new ones every day. In its next funding announcement, on 12 August 2026, it said people had created more than 60 million projects since its launch in November 2024. These are a company's own figures, published to announce money raised. Neither post says what counts as a project or how the count was made.

United States: business applications at the Census Bureau

Business Formation Statistics is a US Census Bureau series that counts applications for an Employer Identification Number (EIN), the number a business uses for tax. Its monthly file, released on 11 September 2026, runs to August 2026. The yearly totals below are our own sums of its monthly figures, not seasonally adjusted.

From 2023 the last column is the Bureau's projection. The first eight months of 2026 hold 4,345,310 applications, against 3,826,844 in the same months of 2025.

A formation, in the same FAQs, is the birth of a new business that pays wages. Worked out from the table, applications in 2022 were 45% above 2019 and the number that became employers within four quarters was 6% above. That does not mean the rest failed: a founder with no payroll is never a formation, and one with no EIN is not in the file at all. The series covers nearly every industry, has no row for software, and by its own FAQs does not yet hold the official economic indicator designation.

Why these figures cannot be added together

One product can be counted by several of these or by none. And the two AI figures measure different things: GitHub counts repositories that call a language model, Stack Overflow counts people who use an AI tool. Neither is a count of products built with AI.

What the figures do not show

• Users or revenue. No source here counts how many of these repositories, projects or applications have one user or one sale. The nearest is a line in Lovable's August 2026 post: its own user survey shows nearly 8 in 10 are building a business or side project they hope to earn from, and more than a third of those already earn revenue. It gives no sample size, date or question, and it describes people who answered, not the 60 million projects.
• Products on sale. A repository, a project and a tax number are each a start, not a product on sale. App store numbers has what reaches the two app stores.
• What happens next. Business survival rates has the official counts of how many new businesses last. Why startups fail has what founders say afterwards, which is surveys.

Nor do they show cause. GitHub says the release of a free version of GitHub Copilot in late 2024 coincided with a jump in sign ups, and of its activity figures it writes:

The Census Bureau's pages give no cause either, and most of the rise in its table came between 2019 and 2021, before the year Octoverse covers.

What to do about it

As far as they go, the figures agree on one thing: more is being started and built than before. GitHub had its biggest year for new repositories, and US business applications in 2025 were 62% above 2019. The next step is our reading, not a finding of any source: when this much gets built, having built something no longer sets a product apart. Getting it in front of the people it is for does, and that is distribution.

• Say who it is for and why it is different: ideal customer and positioning.
• Pick one channel where those people already are. Distribution channels takes each from its own official rules, among them Reddit, Product Hunt and SEO.
• Be where people now ask. ChatGPT and AI search covers getting cited by AI search.
• Check the product before the channel. Will it sell lists what decides whether a stranger pays.