Boris Cherny hasn’t written a single line of code himself this year. Not one. He ships dozens of pull requests a day anyway. Sometimes 150 in 24 hours. Most of that work happens on his phone through the Claude app. The models handle everything.
Cherny created and now leads Claude Code at Anthropic. He built the tool that lets him operate this way. In a recent talk at the AI Ascent 2026 conference, he laid out his reality without hesitation. “For me it’s 100%. The model writes 100% of my code,” he said, according to the event coverage. The implications stretch far beyond one engineer’s workflow.
Yet the phrase everyone uses to describe this shift irritates him. “Vibe coding” entered the lexicon after OpenAI’s Andrej Karpathy introduced it in early 2025. It stuck. Collins Online Dictionary even named it word of the year. Business Insider reported this week that Cherny has grown sick of it. He finds the term glib. It fails to capture the scale of what these systems deliver. Billions in revenue. Millions of lines of production code. Cherny has asked Claude itself for better names. One suggestion, “agentic engineering,” landed flat. He wants something sharper. He invited the public to offer ideas on X.
The dismissal feels characteristic. Cherny spent years as a principal engineer at Meta before joining Anthropic. He authored the O’Reilly book on Programming TypeScript. Technical precision matters to him. So does results. When he describes his output, the numbers stun. A few dozen PRs daily on average. One recent day hit 150. Zero manual edits on his own code since late 2025. He told listeners he simply hasn’t needed to touch the keyboard for writing code in 2026.
His setup relies on dozens of loops running in parallel. Agents spin up, handle tasks, report back. “I have dozens of loops that are running,” Cherny explained. “I sort of feel like loops are the future at this point.” The architecture departs from linear code execution. It favors persistent, self-correcting processes that operate across contexts. One loop might refine a feature. Another reviews it. A third deploys and monitors. Humans step in only when judgment or final approval is required.
This approach has spread inside Anthropic. Recent reports indicate no one at the company writes code manually anymore. Engineers, product teams, even researchers rely on Claude Code instances that communicate autonomously over Slack to resolve issues. The Times of India detailed Cherny’s internal comments from the past day. The title of software engineer, he has argued for months, will start to fade in 2026. Big Tech peers at Google, Meta, and Amazon show signs of moving in the same direction.
Productivity metrics back the transformation. One analysis drawn from Cherny’s appearances noted that he and similar practitioners achieve output levels once unimaginable. Teams report 10 to 30 times gains in some cases. Yet the gains come with caveats. Cherny himself distinguishes between different modes of work. For prototypes and throwaway code, loose natural language instructions suffice. The results arrive fast. Maintainability takes a back seat. For core systems, he insists on alignment first. Plans get reviewed. Specifications refined. The model iterates until standards are met. “If the code sucks, we’re not gonna merge it,” he said in a prior interview. “It’s the same exact bar, and you just ask the model to improve the code and make it better.”
That discipline separates serious adoption from experimentation. Many developers experiment with natural language prompts and accept first drafts. Cherny treats the AI as a collaborator that must meet production bars. The distinction explains why some organizations see transformative results while others accumulate technical debt. Context, iteration, and review remain human responsibilities even as generation shifts to machines.
Cherny pushes the boundary further. He believes the current form of Claude Code itself may soon shrink dramatically. Perhaps to just 100 lines of code. The intelligence lives in the models and the orchestration layers. The interface becomes trivial. This compression mirrors broader patterns in software history. Complex systems give way to simpler abstractions as capability grows.
The larger forecast feels sweeping. Software, in Cherny’s view, stands on the edge of full democratization. Anyone will build what they need. He draws the parallel to the printing press. Literacy rates in Europe surged after its invention. More literature appeared in 50 years than in the previous thousand. Similar forces now apply to code. “Software will be a thing that is fully democratized, that anyone can do,” he said. Non-programmers already use these tools for internal applications, data analysis, and automation. The barrier that once protected the profession dissolves.
Recent developments reinforce the momentum. Anthropic announced fresh compute capacity through a partnership with SpaceX during its Code with Claude conference this week. The added power will fuel larger models and more concurrent agents. Demand continues to climb. Both Claude Code and competing systems from OpenAI generate substantial income. Their output scales into millions of lines monthly.
Yet questions linger. What happens to the craft when generation becomes automatic? Cherny’s answer focuses on higher-order skills. Architecture. Judgment. Taste. The ability to define problems clearly. Review output ruthlessly. Integrate systems across domains. Those competencies grow more valuable. Pure implementation shrinks. Some titles may disappear. New ones will emerge around orchestration, verification, and strategic design.
Critics point to risks. One recent incident involved an agentic setup that deleted a production database in seconds after a vague instruction. Stories like that circulate on X and developer forums. They highlight the need for safeguards, sandboxing, and human oversight. Cherny acknowledges the gaps. His own practice combines multiple loops with explicit review stages. He optimizes for reliable change rather than token cost. Compute becomes cheaper. Attention remains scarce.
Inside Anthropic the shift appears nearly complete. Teams deploy subagents for testing, documentation, and deployment. They coordinate via chat channels. The codebase evolves through continuous AI-driven contributions. Cherny’s personal experience scales to the organization. No manual code. Just direction, verification, and iteration.
The term “vibe coding” may fade. Cherny’s search for a replacement signals a desire for language that matches the seriousness of the change. Whether “agentic engineering” or something fresher wins out matters less than the underlying reality. The activity has moved beyond casual prompting. It now resembles distributed systems of specialized agents guided by human intent.
Developers who master this new mode report dramatic output increases. They maintain quality by staying involved in planning and critique. Those who treat the tools as magic black boxes encounter bugs, inconsistencies, and mounting debt. The difference lies in process. Cherny’s loops, parallel reviews, and high standards offer one template. Others will invent their own.
What arrives next remains uncertain in detail. The printing press analogy suggests broad access and explosive creation. Software could proliferate the way books once did. New applications. New industries. New problems to solve. The engineers who once typed every character may find themselves directing symphonies of agents instead. Some will resist. Others will adapt quickly. The data from early adopters like Cherny shows the productivity curve bending sharply upward.
He continues to ship from his phone. Dozens of PRs. Loops humming in the background. The code arrives clean enough to merge without his direct edits. That outcome once sounded like science fiction. It now defines his daily work. And if his predictions hold, it will soon define much of the industry’s work too. The question is no longer whether the change is coming. It is how organizations and individuals position themselves inside it.
WebProNews is an iEntry Publication