Generative AI’s Impact on Software Development

SAN FRANCISCO, June 3 – Two years after the introduction of ChatGPT, the returns on investments in generative AI remain somewhat unclear, yet one field shines: software development.

Startups focusing on code generation, often referred to as “code-gen,” are seeing soaring valuations as companies aim to utilize AI to supplement, and sometimes replace, costly software engineers.

For instance, Cursor, a San Francisco-based startup specializing in code generation that can propose and complete code autonomously, secured $900 million in May, hitting a $10 billion valuation, with backing from prominent tech investors like Thrive Capital, Andreessen Horowitz, and Accel.

Meanwhile, Windsurf, known for its widely-used AI tool Codeium, has become a target for OpenAI, which is reportedly negotiating to acquire the company for $3 billion.

Windsurf’s tool allows users to turn simple English instructions into code—an approach dubbed “vibe coding”—which enables those lacking coding knowledge to create software. OpenAI and Windsurf have chosen not to comment on these acquisition discussions.

Scott Wu, the CEO of Cognition, a code gen startup, mentioned, “AI has automated all the repetitive, tedious work.” He noted that the role of a software engineer has evolved significantly, moving away from just memorizing specific syntax.

Founders and investors in the code-gen realm believe they’re in a race to capture users quickly and establish a foothold as the go-to AI coding tool.

However, since many of these startups rely on AI foundational models from companies like OpenAI and Anthropic, their costs are climbing, and none are yet profitable.

They also face potential competition from major players like Google and Microsoft, both of which have recently unveiled new code-gen products, while Anthropic is working on its version as well.

Despite these challenges, the rapid growth is noteworthy, especially as these startups compete against tech giants. For example, Microsoft’s GitHub Copilot, launched in 2021 and considered the leading player in code-gen, reported over $500 million in revenue last year.

Microsoft didn’t comment on GitHub Copilot’s revenue figures, but noted it has amassed over 15 million users as of its April earnings call.

Job Market Changes

As AI transforms the industry, many roles, especially entry-level coding jobs that are fairly basic, might vanish. Signalfire, a venture capital firm tracking tech employment, indicated that the hiring of individuals with less than a year of experience dropped by 24% in 2024, attributing the decline to AI taking over tasks typically assigned to junior software engineers.

Google’s CEO stated in April that a significant portion—over 30%—of its code is now AI-generated, while Amazon’s CEO reported that the company saved “the equivalent of 4,500 developer-years” through AI utilization. Both companies declined to provide further comments.

In the same month, Microsoft’s CEO mentioned that about 20% to 30% of their code comes from AI. That month, Microsoft also announced layoffs affecting 6,000 workers worldwide, with over 40% of those being software developers based in Washington.

A Microsoft representative commented, “We’re focused on creating AI that empowers developers to be more productive, creative, and save time.” They acknowledged that while some roles will certainly shift due to AI advancements, human intelligence will remain pivotal in software development.

Financial Struggles

Interestingly, several “vibe-coding” platforms have reported impressive annual revenues. Cursor has achieved a remarkable $100 million in recurring revenue within just two years of its launch, employing only 60 people. Windsurf, founded in 2021, launched its product in late 2024 and has already reached annualized revenues of $50 million, according to insiders.

However, both startups are running negative gross margins, spending more than they earn, as reported by multiple investor sources.

“People will likely see the costs for coding assistants rise,” suggested Quinn Slack, CEO of Sourcegraph.

Both Cursor and Windsurf are helmed by young MIT graduates, illustrating the intense competition in the AI startup environment. Martin Casado, a partner at Andreessen Horowitz—an investor in Cursor—noted, “I haven’t seen this level of effort since the first Internet boom.”

What remains uncertain is whether these newcomers can maintain their customer bases as larger companies enter the arena.

Scott Raney, managing director at Redpoint Ventures, remarked, “It’s often not just about who has superior technology—it’s about who will effectively utilize and market their technology.”

Creating Proprietary AI Models

Most AI coding startups rely on the Claude AI model from Anthropic, which recently surpassed $3 billion in annual revenue, benefiting from fees paid by these code-gen companies.

However, some startups are working on developing their own models. Windsurf, for instance, announced in May that it is creating internal AI models specifically designed for software engineering to enhance user experience. Cursor is also hiring researchers to pre-train its own advanced models, reducing reliance on foundation model providers.

Yet, the challenge remains steep for startups seeking to train their own AI models, facing significant costs for the required computing resources.

Replit previously abandoned its own model training plans, whereas Poolside has raised over $600 million for a coding-specific model and partnered with Amazon Web Services, though it hasn’t launched a public product yet.

Another company, Magic Dev, has raised nearly $500 million since 2023 and mentioned a frontier-level coding model debut was slated for summer 2024, but no product has yet been released.

Poolside declined to comment on its progress, and Magic Dev did not respond to requests for information.