How did Quad Code start and evolve over the first year?

Quad Code began as a small prototype and grew through user feedback, bugs, and telemetry. It took months before it clicked with users; its growth accelerated as the model improved and latent demand was discovered. Timestamp: 1680

Is Boris still coding personally, and how much of his work is AI-written?

Boris states that 100% of his code is written by Quad Code, with him reviewing the output but not hand-writing lines since November. He emphasizes that there is still a safety and review layer, so humans remain involved. Timestamp: 980

What is latent demand and how did Quad Code leverage it?

Latent demand refers to discovering new use cases when users adopt a technology in unexpected ways. Quad Code and Co-Work emerged from seeing people use the terminal for non-coding tasks, prompting productization. Timestamp: 2960

What is plan mode and how do you use it in Quad Code?

Plan mode is a simple prompt that tells the model not to write code yet; you plan and validate the approach before execution. Boris highlights it as a core productivity tip for getting reliable first-pass results. Timestamp: 4195

How does Anthropic think about AI safety in the context of Quad Code and Co-Work?

Safety is layered: alignment, mechanistic understanding of the model's neurons, and testing in controlled settings before real-world use. They release early as a research preview to learn from real-world interactions. Timestamp: 3330

What is co-work and how did it come about?

Co-work started as a way to extend Quad Code into desktop and non-terminal workflows, built within about 10 days and iterated quickly with safety guardrails. It demonstrates latent demand outside pure coding. Timestamp: 3190

What is Rich Sutton's Bitter Lesson, and how does it apply here?

The Bitter Lesson posits that more general models outperform highly specialized ones. Boris uses this to argue for relying on strong general models and minimal scaffolding, betting on next-gen models rather than over-engineering today. Timestamp: 3891

Should people be worried about AI taking jobs, and how should teams respond?

Boris suggests a broader societal shift may come with disruption, but emphasizes experimentation, token budgets for exploration, and becoming generalists who can cross disciplines to adapt. Timestamp: 1546

What does Boris plan to do after AGI, or what keeps him engaged today?

Boris describes a future where miso-making and other long-term crafts appeal to him, illustrating a balance between high-tech work and patient, hands-on skills. Timestamp: 4353

Head of Claude Code: What happens after coding is solved | Boris Cherny

Lenny's Podcast

People & Blogs3 min read88 min video

Feb 19, 2026|347,073 views|7,160|977

Save to Pod

Key Moments

TL;DR

Quad Code makes coding largely solved; engineers focus on tooling and strategy.

Key Insights

Coding is largely solved: Claude/Quad Code now writes the majority of code for many developers.

Productivity soars: per-engineer productivity around 200% increase; PR throughput climbs dramatically.

AI as a co-worker: tools like Quad Code and co-work automate coding and many adjacent tasks.

Next frontier is not just code: AI proposes fixes, handles non-coding work, and aids project management.

Token economy matters: generous token budgets accelerate experimentation and innovation; underfunding can spark creativity.

Democratization and disruption: a future where many can build software; society must address job impact and policy.

INTRODUCING CLAUDE CODE AND ITS IMPACT

Claude Code, the centerpiece of Anthropic's AI-assisted coding efforts, has catalyzed a dramatic shift in software development. In the year since its public unveiling, it has moved from a promising prototype to a core productivity tool: public metrics show quadruple growth in usage and a meaningful share of industry-wide code activity, with estimates suggesting millions of commits now informed or authored by the AI. The technology pairs code generation with tooling use, bug-fix suggestions, and telemetry-driven improvements, progressively acting as a robust co-worker. Despite its maturation, Boris emphasizes that safety and human oversight remain crucial: the AI writes code, but humans still review to ensure correctness and resilience in real-world systems.

FROM HANDS-ON CODING TO AI-DRIVEN PRODUCTION

Boris Cherny details a personal journey that mirrors the broader industry shift. He briefly left Anthropic for a promising project, then returned because the mission—safety in AI—felt indispensable. His early prototypes began in a terminal-based workflow; initially dismissed as unorthodox, the approach proved surprisingly effective as the model rapidly improved. The lesson: start small, learn the model’s boundaries, and let user feedback drive iteration. The terminal interface proved resilient to rapid model changes, underscoring a key product lesson: build for speed and adaptability rather than perfection in the initial phase.

ACCELERATED GROWTH AND ADOPTION ACROSS INDUSTRIES

Over the past year, Quad Code’s impact has broadened from a niche tool to a global catalyst for engineering productivity. Boris cites striking numbers: a significant fraction of code contributions are now AI-assisted, and daily active users have doubled in recent months. The product’s design—combining coding with tool use, and eventually with broader automation—has unlocked latent demand. The adoption gains have been reinforced by feedback loops from diverse customers, from startups to large tech giants, as teams learn to integrate AI into their existing pipelines and workflows.

WHAT IT MEANS TO WRITE 100% WITH AI

A pivotal milestone is the claim that Boris now writes all his code with Quad Code, with humans still overseeing correctness and safety. This shift reframes what it means to be a software engineer: code becomes a product of collaboration with intelligent tooling rather than solitary crafts. Yet human judgment remains essential for design decisions, architectural patterns, and safety checks. A memorable example from their experience shows Quad Code diagnosing a memory leak by creating a just-in-time analysis tool, demonstrating how AI can accelerate problem-solving while still requiring human interpretation and governance.

BEYOND CODING: CO-WORK, AUTOMATION, AND NEXT FRONTIERS

The trajectory extends beyond writing code. Quad Code and the co-work paradigm begin to autonomize adjacent tasks: triaging bug reports, consolidating feedback, and orchestrating project management across Slack, email, and spreadsheets. The team is exploring how AI can generate ideas for bug fixes and features, not just implement them. A core takeaway is that coding is increasingly solved, while the next frontier lies in applying AI to planning, coordination, and non-coding work—transforming how teams decide what to build and how to execute it.

THE FUTURE OF WORK, SKILLS, AND SOCIETY

Cherny frames the broader implications as a potential democratization of programming, akin to the printing press's impact on literacy and knowledge. If AI lowers the barrier to software creation, a wider audience could build software, accelerating innovation but also injecting new social and policy challenges. He argues that agent-enabled tools—AI that can use apps and run tasks—will permeate roles beyond engineering, including product, design, and data science. The path forward involves experimentation, generalist skill cultivation, and thoughtful policy discussion to manage transitions and maximize societal benefit.

Mentioned in This Episode

●Tools & Products

●Books

●People Referenced

Quad Code: Practical dos and don'ts for AI-assisted coding

Practical takeaways from this episode

Do This

Use the most capable model (e.g., Opus 4.6) for tasks to minimize handholding.

Start tasks in plan mode to validate intent before execution.

Give engineers ample token budgets to explore and experiment.

Release features early (even if rough) to learn from real-user feedback.

Be a generalist and encourage cross-disciplinary skills on teams.

Avoid This

Don’t box the model with overly rigid step-by-step workflows; let it explore tools and goals.

Don’t rely solely on fine-tuning or tiny models; prefer leveraging a strong general model when possible.

Don’t underfund the initial exploration phase; early token scarcity can stifle innovation.

Quad/Claude Code impact metrics cited

Data extracted from this episode

Metric	Value	Notes
GitHub commits authored by Quad Code	4%	Public commits; private may be higher
Engineering productivity per engineer	200%	Reported productivity gain with Quad Code
Daily active users growth	Doubled in past month	Measured growth trajectory
Code written by Quad Code (milestone)	100% by November	Reached milestone in that period

Common Questions

Claude Code is Anthropic's coding assistant that can write code and use tools autonomously. Boris argues it has dramatically increased engineer productivity by taking over minutiae and enabling more high-level work, signaling a shift in how software is written. Timestamp: 56