Boris Cherny, creator and head of Claude Code, told Lenny’s Podcast that software development is moving toward agentic workflows. Anthropic’s own documentation supports a more measured takeaway: developers get better results when they manage context, verify changes, configure permissions, and use tools such as hooks...
Boris Cherny, creator and head of Claude Code, discussed how agentic coding is changing software development in an interview with Lenny’s Newsletter and Lenny’s Podcast.
In the interview page for Head of Claude Code: What happens after coding is solved, Lenny’s Newsletter identifies Cherny as the creator and head of Claude Code and says the discussion covers practical ways to get more out of the product, its growth, and Cherny’s view that coding is becoming increasingly agentic.
Anthropic’s own Claude Code overview describes the product as an agentic coding tool that can read a codebase, edit files, run commands, integrate with developer tools, use Model Context Protocol servers, support hooks, and spawn multiple agents for parallel work. That framing is important: Claude Code is not presented only as a chatbot for answering programming questions, but as a system designed to operate inside a developer workflow.
WIRED, in its article How Claude Code Is Reshaping Software—and Anthropic, reported that Cherny uses Claude Code for all of his own coding. WIRED also reported Cherny’s statement that roughly 95% of the Claude Code team’s code is written with Claude Code. That figure is notable, but it should be read as a claim about one team’s internal usage, not as proof that most software teams can immediately achieve the same results.
Some social posts around the interview have framed Claude Code performance as a matter of secret settings or hidden menus. The source-backed version is less sensational: Anthropic’s official Claude Code best-practices documentation emphasizes workflow design, context management, verification loops, and explicit permissions.
Anthropic recommends that users ask Claude Code to explore a codebase before making changes, have it make a plan, and then verify its work. The company’s guidance also points to more advanced capabilities including hooks, permissions, MCP servers, skills, subagents, and parallel sessions. These are not magic switches; they are ways to give the tool clearer context, safer operating boundaries, and more structured tasks.
That distinction matters for developers evaluating agentic coding tools. A vague prompt in a large unfamiliar repository is likely to produce weaker results than a task with clear constraints, relevant files, tests, and a review step. Anthropic’s documentation repeatedly reinforces that Claude Code works best when it is embedded in a disciplined engineering process rather than treated as an autonomous replacement for one.
The Lenny’s Podcast episode description says the conversation looks at what happens after coding is “solved,” but the supporting sources point to a more nuanced shift. Claude Code can perform code edits and run commands, yet the developer still has to define the objective, manage risk, inspect changes, and decide whether the output is correct.
Anthropic’s best-practices page makes this especially clear through its emphasis on verification loops. In practice, that means developers should use tests, code review, diffs, and incremental changes to keep the system grounded. The more authority a coding agent has in a repository, the more important it becomes to manage permissions and review outcomes.
Subagents and parallel sessions also suggest a different style of work. Instead of asking one assistant to do everything in a single linear chat, a developer may break work into exploration, implementation, testing, and review. Anthropic’s overview says Claude Code can spawn multiple agents for parallel work, while the best-practices documentation describes subagents and related workflow patterns as ways to improve results.
The most defensible takeaway from Cherny’s interview, WIRED’s reporting, and Anthropic’s documentation is that agentic coding rewards operational discipline. Teams that want better results from Claude Code should focus less on viral claims about hidden performance boosts and more on documented practices: provide useful context, plan before editing, constrain permissions, connect the right tools, run tests, and review the output.
Claude Code’s rapid adoption inside Anthropic, as reported by WIRED, shows that the company is deeply invested in using its own coding agent. But the broader lesson for software teams is not that coding has already been fully automated. It is that coding workflows are becoming more interactive, tool-driven, and agent-assisted—and that developers who understand the configuration and verification model are likely to get more reliable results.
Boris Cherny, creator and head of Claude Code, discussed how agentic coding is changing software development in an interview with Lenny’s Newsletter and Lenny’s Podcast.
That framing is important: Claude Code is not presented only as a chatbot for answering programming questions, but as a system designed to operate inside a developer workflow.
WIRED, in its article How Claude Code Is Reshaping Software—and Anthropic , reported that Cherny uses Claude Code for all of his own coding.
Continue reading
