Paid coding assistants prove models can earn software revenue

GitHub now charges $10 a month for Copilot, which writes about 40% of code where enabled. Language models become a per-seat software business.
Ten dollars a month is now the price of a language model that writes code. GitHub made Copilot, its AI pair programmer built on OpenAI's Codex model, generally available to individual developers on 21 June for $10 a month or $100 a year, free for verified students and maintainers of popular open source projects. More than 1.2 million developers used the free technical preview over the past 12 months, and in files where Copilot is turned on it now writes nearly 40% of the code in popular languages such as Python.
The price tag matters more than the demo. A large language model is now sold per seat like ordinary software, to a user base counted in millions. Assistants that take over part of skilled work and charge per seat are the first proven business model for large language models.
Underneath the subscription runs a meter. The price per user is fixed while every suggestion costs compute, so gross margin depends on inference cost per active developer, a question every AI application will face while compute stays scarce and expensive. The other side of the ledger is generous. Against the cost of a developer's time, $100 a year is a small fraction of the value of the work the tool takes over, which leaves room for pricing power in the team and enterprise plans GitHub plans to offer companies later this year.
Distribution is picking the early winners. Copilot works inside Visual Studio Code, Visual Studio, Neovim and JetBrains editors, where developers already work. Two days after the launch Amazon opened a preview of CodeWhisperer, its own coding companion trained on open source code and Amazon's internal code, and tied it to its cloud. General autocomplete now belongs to the platform owners.
What remains for startups is deeper and narrower: code review, test generation, security scanning of machine-written code, migration of legacy systems, and tools for specific languages and frameworks. Enterprises will pay for controls over what code leaves the company and for proof of productivity, so measurement is a product in itself. Copilot also sets a template outside software, a per-seat price for a tool that does part of a skilled job, with free access for students and open source as distribution.
Conversion is the open risk to all of this. How many of the 1.2 million preview users turn into paid seats, and what the business plan costs when it arrives, will show whether developers pay for what they tried for free. Amazon could price CodeWhisperer at zero after the preview to pull developers toward its cloud. A legal debate over training on public code under open source licenses also hangs over the category, and large companies have yet to decide whether to allow these tools on their private repositories.
Fanaura, a family office investing in AI-first companies with a horizon of years, sees the value going to whoever owns the workflow and controls inference cost. Teams that bring the per-seat assistant to other professions whose output is text or code sit at the centre of the companies we invest in.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building AI tools for code review, testing or legacy migration? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.