AI’s Impact on Engineering: Key Takeaways from Lenny Rachitsky’s Analysis

In a recent LinkedIn post, Lenny Rachitsky shares key takeaways from a conversation with Boris Cherny, Head of Claude Code at Anthropic, offering a glimpse into the transformative power of AI in software engineering and beyond.

Rachitsky highlights Cherny’s perspective that coding, for many use cases, is now a “solved” problem, largely thanks to advanced AI models like Claude Code. Cherny himself reportedly hasn’t written code manually since November, with all his work being AI-authored. Despite this, he remains a highly productive engineer, shipping a significant number of pull requests daily while leading his team.

“Back at Meta, with hundreds of engineers working on productivity, we’d see gains of a few percentage points in a year. Now we’re seeing hundreds of percentage points.”

Productivity Leaps and Evolving Roles

One of the most striking points Rachitsky relays from Cherny is the dramatic increase in engineer productivity observed at Anthropic since the adoption of Claude Code. This surge, described as a 200% increase, dwarfs previous productivity gains achieved through traditional means. As Rachitsky notes, Cherny contrasts this with his experience at Meta, where significant efforts yielded only minor percentage point improvements.

Rachitsky further elaborates on how AI’s role is expanding beyond mere code generation. According to Cherny’s insights, AI is beginning to contribute to the ideation process itself.

“Claude is starting to come up with ideas. It’s looking through feedback, bug reports, and telemetry, then suggesting features to ship.”

This evolution suggests a future where AI assists not just in execution but also in strategic planning and feature development. Rachitsky points out that roles adjacent to engineering, such as product managers, designers, and data scientists, are likely to be the next to undergo similar transformations as AI agents become more sophisticated.

Designing for the Future of AI

A core principle discussed by Rachitsky, attributed to Cherny, is the importance of building for future AI capabilities rather than current ones. This forward-looking approach, while potentially challenging in the short term, aims to position products for immediate success once more advanced AI models become available.

“It’s going to be uncomfortable because your product-market fit won’t be very good for the first six months. But when that model comes out, you’ll hit the ground running.”

Rachitsky also touches upon the concept of “latent demand,” where tools like Claude Code were developed by observing existing user behaviors and making them more efficient. The emergence of Cowork, an AI tool for non-coding tasks, exemplifies this, stemming from users repurposing Claude Code for analyzing medical images or recovering data.

Strategic Considerations for AI Adoption

The discussion, as relayed by Rachitsky, delves into practical advice for businesses integrating AI. Cherny advises against premature optimization of token costs, especially during experimentation phases. He argues that the cost of tokens is negligible compared to engineer salaries at small scales, and that focus should remain on validating ideas.

Furthermore, Rachitsky highlights Cherny’s strategy of intentionally underfunding headcount on projects. This constraint forces engineers to leverage AI more effectively, driving creative applications rather than simply increasing typing speed.

The Rise of the Generalist

Looking ahead, Rachitsky conveys Cherny’s prediction that generalists will be the most successful professionals in the AI era. The ability to cross disciplines and understand broader problem contexts is becoming increasingly valuable.

“Try to be a generalist more than you have in the past. Some of the most effective engineers cross over disciplines. The people who will be rewarded most won’t just be AI-native—they’ll be curious generalists who can think about the broader problem they’re solving.”

Finally, Rachitsky emphasizes Cherny’s recommendation to always use the most capable AI model available, even if it’s not the cheapest. The efficiency and accuracy of a superior model can often outweigh the cost, preventing the token expenditure associated with corrections from less capable models. Cherny reportedly uses the highest-tier model for all his work.