For nearly 50 years, software engineering operated under a simple, painful truth: throwing more people at a late project just makes it later. Fred Brooks coined that idea in 1975, and it became gospel. Now a new generation of AI tools is making the whole framework look quaint.

Companies like OpenAI, Anthropic, and the AI-powered code editor Cursor are demonstrating that smaller teams, armed with the right models, can produce output that once required headcounts several times larger. The traditional scaling bottleneck wasn’t computing power. It was communication overhead between humans. AI doesn’t have that problem.

Brooks’s Law meets its match

Here’s the quick version of Brooks’s Law: every new person you add to a software team creates new communication channels. Three developers means three pairwise connections. Ten developers means 45. The math gets ugly fast, and so does the coordination tax.

AI coding assistants sidestep this entirely. They function as non-human contributors that don’t need to attend standup meetings, don’t misunderstand Slack messages, and don’t take PTO. In English: the coordination overhead that made large teams inefficient simply doesn’t apply to an AI pair programmer.

OpenAI’s latest models are specifically designed to maximize output with fewer personnel, focusing on high-context coding tasks and automated deployment processes. Anthropic’s Claude 3.5 serves a similar function, handling the kind of boilerplate and repetitive work that used to eat up junior developer hours.

The result is a fundamental shift in how engineering organizations think about staffing. Instead of hiring 20 engineers and hoping they gel, a team of five equipped with advanced language models can cover similar ground.

Cursor’s breakout moment

If OpenAI and Anthropic are building the engines, Cursor is building the car people actually drive. The AI-native integrated development environment has become one of the fastest-growing tools in software, estimated to generate $500M in annual revenue.

That number comes with a significant asterisk, though. Cursor is currently operating at a loss, weighed down by the substantial costs of running frontier AI models and maintaining the infrastructure to support them. It’s a familiar playbook in tech: grow fast, figure out margins later. Whether that strategy holds depends on how quickly model inference costs come down.

An empirical study examining real-world Git-based projects found that adopting Cursor produced a significant increase in development velocity. The catch? The boost was transient. Initial productivity gains tapered over time as developers adjusted their workflows, suggesting that AI coding tools deliver the biggest impact during adoption rather than as a permanent multiplier.

That finding matters because it tempers some of the more breathless claims about AI replacing developers entirely. The more accurate picture is that these tools compress certain phases of work, particularly initial scaffolding and boilerplate generation, while leaving the harder problems of system design and code review squarely in human hands.

What this means beyond traditional software

The implications extend well past conventional tech companies. In crypto development specifically, AI tools are already enabling smaller teams to deliver smart contracts and on-chain infrastructure more efficiently. Building a DeFi protocol used to require a well-funded team with deep Solidity expertise. Now, AI assistants can handle much of the routine contract generation, letting a handful of experienced developers focus on architecture, security audits, and the novel logic that actually differentiates a protocol.

This dynamic is also contributing to the growth of decentralized AI infrastructure projects, where the intersection of machine learning and blockchain creates natural synergies. Smaller teams building AI-adjacent crypto tools benefit doubly: they use AI to build faster, and the products they’re building serve the AI ecosystem itself.

For investors watching the broader landscape, the companies that matter here aren’t necessarily the ones with the biggest engineering departments. They’re the ones that figured out how to pair a small, senior team with the right AI tooling stack. That’s a meaningful change in how to evaluate early-stage projects, particularly in crypto where lean teams have always been the norm rather than the exception.

The risk, naturally, is over-reliance. AI-generated code still needs human review, and the transient nature of productivity gains observed in studies suggests that organizations need thoughtful integration strategies rather than simply handing every task to a chatbot. Security-critical applications, the kind that dominate crypto infrastructure, demand especially careful oversight.

Look, the economics are moving in one direction. Model costs will likely decrease. Tool sophistication will increase. The teams that adapt early to this new scaling dynamic will have a structural advantage over those still hiring by the dozen. Brooks probably wouldn’t be upset about it. He just wanted the software to ship on time.