GenAI PMBoris Cherny
A named individual quoted in connection with Mythos Preview and defensive cyber capabilities. He is presented as a source for a cybersecurity model update.
Key Highlights
- Boris Cherny is a recurring public source for Anthropic-related product updates spanning Claude Code, pricing, token usage, and model capability communication.
- His posts offer practical lessons for AI PMs on observability, self-serve debugging, subscription design, and handling quality incidents transparently.
- He was cited describing Mythos Preview as the first model to fully solve UK AISI cyber ranges end-to-end, signaling progress in defensive cyber AI deployment.
- Several mentions focus on operational product details such as custom token counting across cloud providers and a new Claude Code `/usage` command.
- His updates frequently reflect the PM-critical trade-offs between user experience, infrastructure capacity, enterprise demand, and trustworthy rollout messaging.
Overview
Boris Cherny is a frequently cited product and technical voice associated with Anthropic’s Claude ecosystem, especially Claude Code, usage controls, pricing communication, and model capability updates. Across newsletter mentions, he appears as a public-facing source for product changes, operational clarifications, and performance claims spanning developer tooling, token accounting, subscription design, and cybersecurity model progress.For AI Product Managers, Cherny matters because his updates sit at the intersection of product operations and model deployment: how usage is measured, how plans are packaged, how defaults are tuned in response to user behavior, how tooling quality incidents are handled, and how frontier capabilities like Mythos Preview are framed for real-world defensive cyber use. His comments offer useful signals on product instrumentation, customer communication, and launch discipline in fast-moving AI products.
Key Developments
- 2026-03-26: Said Claude Code Review automatically catches over 99% of bugs, positioning AI-assisted code review as a high-automation workflow with only light human sanity checking remaining.
- 2026-04-04: Explained that surging Claude usage from third-party tools had exceeded the design assumptions of existing subscriptions, and that capacity was being prioritized for core product and API customers.
- 2026-04-06: Shared that custom token-counting was built for Bedrock, Vertex, and Azure, while Anthropic’s native token-counting endpoint was used directly where available.
- 2026-04-13: Noted that Claude’s default dialogue style was changed to Medium after complaints about high token usage, while clarifying that response length was not simply reduced.
- 2026-04-14: Showed that although Claude can be configured to block `node_modules`, allowing access to dependency source code often improves debugging and code assistance quality.
- 2026-04-16: Clarified that new pricing plans had already been live for roughly six months and were introduced in response to enterprise customer demand.
- 2026-04-24: Published a post-mortem on recent Claude Code quality issues, outlining investigation steps, root causes, and lessons for improving LLM-powered developer tools.
- 2026-05-10: Reported that Claude Code’s move to a native installer meant npm-only statistics understated actual adoption, and said the product had reached its second-highest signup day ever with 15× growth since January 1.
- 2026-05-10: Also said Claude Code UX was being improved for faster responsiveness, with debug logs added to help users self-diagnose hangs.
- 2026-05-11: Announced a new `/usage` command in Claude Code that breaks down which actions consume tokens and where usage limits apply.
- 2026-05-14: Stated that Mythos Preview was the first model to fully solve UK AISI cyber ranges end-to-end, including the previously unsolved “Cooling Tower,” and said deployment to defenders was underway with more to come on Glasswing.
Relevance to AI PMs
1. Instrumentation and cost visibility: Cherny’s updates on token counting, `/usage`, and dialogue-style defaults show how important it is to expose actionable usage diagnostics to users. AI PMs should treat observability, token transparency, and self-serve debugging as core product features, not back-office tooling.2. Capacity, packaging, and pricing design: His comments on subscription strain from third-party demand and on enterprise-driven pricing changes illustrate that AI product packaging should reflect actual usage patterns, infrastructure limits, and strategic customer segments. PMs should regularly revisit plan design as model behavior and distribution channels evolve.
3. Trust through operational communication: The post-mortem on Claude Code quality issues and clarifications on product changes demonstrate strong launch and incident communication hygiene. AI PMs can apply this by publishing root-cause summaries, explaining trade-offs clearly, and setting expectations when defaults or access policies change.
Related
- Anthropic: Cherny is most closely associated with Anthropic-related product updates and public explanations.
- Claude / Claude Code / Claude Code Review: The majority of mentions involve developer tooling, usage controls, installer changes, debugging workflows, and code review automation.
- Bedrock, Vertex, Azure: Connected through custom token-counting implementations where native accounting APIs were unavailable.
- Pricing plans / enterprise customer demand: Related to his clarification that packaging changes were driven by enterprise needs and had been publicly available for months.
- Effort, Medium, Low, High: Connected to Claude dialogue style settings and token-usage trade-offs; Cherny specifically noted the default moved to Medium.
- Mythos Preview: Cherny is cited as a source describing major defensive cyber capability progress for this model.
- UK AISI: Referenced in connection with cyber evaluation ranges that Mythos Preview reportedly solved end-to-end.
- Glasswing: Mentioned as an upcoming related initiative or follow-on disclosure tied to Mythos Preview deployment.
- RAG and agentic-search: Not directly attributed to his announcements here, but relevant adjacent concepts for AI PMs thinking about tool use, retrieval, and product architecture in model-powered systems.
- Opus, Opus-46, Sonnet-46, GPT-4-Turbo, Spotify: Related entities in the broader newsletter graph and competitive/product context, though not central to Cherny’s mentions in this dataset.
Newsletter Mentions (22)
“#10 𝕏 Boris Cherny : Mythos Preview is the first model to fully solve UK AISI’s cyber ranges end-to-end—including the once-unsolved “Cooling Tower”—and is being deployed to defenders as fast as responsibly possible, with more on Glasswing coming soon.”
#10 𝕏 Boris Cherny : Mythos Preview is the first model to fully solve UK AISI’s cyber ranges end-to-end—including the once-unsolved “Cooling Tower”—and is being deployed to defenders as fast as responsibly possible, with more on Glasswing coming soon. #11 in Peter Yang shows that Claude Code and Codex can effortlessly merge, edit, and crop scanned PDFs with simple prompts. He notes it far outperforms the clunky workflows in Preview or Adobe Acrobat.
“Boris Cherny launched a new `/usage` command in Claude Code that provides a deep dive into exactly which actions are consuming your tokens and their limits, so you can self-serve debug.”
#8 𝕏 Boris Cherny launched a new `/usage` command in Claude Code that provides a deep dive into exactly which actions are consuming your tokens and their limits, so you can self-serve debug. #10 𝕏 Boris Cherny doubled the 5-hour run limits on every Claude Code plan this week to streamline planning and execution.
“#5 𝕏 Boris Cherny says Claude Code’s switch to a native installer means npm-only stats undercount its real usage. On Thursday it hit its second-highest signup day ever with 15× growth since Jan 1—now you can ask Claude to debug your SQL. #6 𝕏 Boris Cherny is enhancing Claude Code’s UX for snappier performance and adding debug logs so users can self-serve hang diagnostics.”
GenAI PM Daily May 10, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 11 insights for PM Builders, ranked by relevance from X, Blogs, and LinkedIn. PromptLayer’s multi-step agent evaluation framework #1 𝕏 Jason Zhou launched `/goal` support in CodeX and Hermes agents for one-step autonomous coding, advising use of interview mode, clear stop conditions, and a goal-buddy to manage state and goal files. #2 📝 PromptLayer Blog What Is Agent Evaluation? A Practical Guide for AI Teams - Agent evaluation tests whether an AI agent reliably completes tasks across real inputs, edge cases, and new versions by scoring not just final outputs but multi-step behavior via black-box, trajectory, and component-level evaluations, using metrics like task completion rate, tool selection accuracy, unsupported-claim rate, latency/cost per step, and regression pass rate. PromptLayer offers tracing with span-level context, reusable datasets, batch evaluations, backtesting, regression testing, automated evaluation triggers on new prompt versions, and flexible pipelines including code execution, human input, conversation simulation, regex checks, and LLM assertions. #3 in Udi Menkes built his new product’s entire data flow in a single interactive HTML file—complete with diagrams, in-page navigation, and color-coded complexity—letting his team understand it in minutes instead of hours. #4 𝕏 Garry Tan suggests diagramming your AI agent codebases and architecture in plain ASCII, then relentlessly questioning each component to clarify design and accelerate product development. #5 𝕏 Boris Cherny says Claude Code’s switch to a native installer means npm-only stats undercount its real usage. On Thursday it hit its second-highest signup day ever with 15× growth since Jan 1—now you can ask Claude to debug your SQL. #6 𝕏 Boris Cherny is enhancing Claude Code’s UX for snappier performance and adding debug logs so users can self-serve hang diagnostics. #7 𝕏 Harrison Chase calls LangSmith an org-wide platform for building AI agents that speeds up cross-functional collaboration and tightens feedback loops. #8 𝕏 Santiago showcases a step-by-step guide for constructing Python-powered multi-agent systems from scratch, leveraging MCP and A2A patterns to incrementally add complexity and enable collaborative AI agents. #9 𝕏 Garry Tan spends $2K/mo on Openclaw AI tokens to turbocharge product development and startup insights. He’s “tokenmaxxing” now with a goal to make these capabilities affordable for everyone in 18 months. #10 𝕏 Harrison Chase argues that treating AI agents as systems to measure and iteratively improve isn’t just a technical challenge—it demands intentional human collaboration and team processes. #11 in Peter Yang warns that unedited AI-generated markdown can compound small errors over time—what starts as 5% “slop” quickly balloons into an overwhelming pile of confusing, unverified content. Found this valuable? Share it with another PM - they can subscribe at genaipm.com Unsubscribe • Switch to Weekly
“Boris Cherny published a post-mortem on recent Claude Code quality issues, detailing the investigation steps, root causes, and key takeaways for improving LLM tooling.”
#18 𝕏 Boris Cherny published a post-mortem on recent Claude Code quality issues, detailing the investigation steps, root causes, and key takeaways for improving LLM tooling. #19 𝕏 Jeff Dean provided advice on the Decoupled DiLoCo training system, enabling graceful failure handling in large-scale jobs by letting (N-1)/N units proceed when one fails.
“Boris Cherny clarifies that the new pricing plans went live six months ago (in November) in response to enterprise customer demand and have been publicly available on their site since then.”
#20 𝕏 Boris Cherny clarifies that the new pricing plans went live six months ago (in November) in response to enterprise customer demand and have been publicly available on their site since then.
“Boris Cherny shows that while you can configure Claude to block node_modules, in practice allowing it to read your dependencies’ source code often yields more helpful insights.”
#13 𝕏 Boris Cherny shows that while you can configure Claude to block node_modules, in practice allowing it to read your dependencies’ source code often yields more helpful insights.
“#11 𝕏 Boris Cherny says Claude’s default dialogue style was switched to Medium after users complained about high token usage.”
GenAI PM Daily April 13, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 14 insights for PM Builders, ranked by relevance from X, Blogs, and YouTube. #11 𝕏 Boris Cherny says Claude’s default dialogue style was switched to Medium after users complained about high token usage. He clarifies that response lengths haven’t been cut and future tweaks will follow user feedback.
“Boris Cherny built custom token-counting for Bedrock, Vertex, and Azure (which lack a native API) while using Anthropic’s built-in token-counting endpoint directly.”
#4 𝕏 Boris Cherny built custom token-counting for Bedrock, Vertex, and Azure (which lack a native API) while using Anthropic’s built-in token-counting endpoint directly.
“#4 𝕏 Boris Cherny explains that a surge in Claude usage from third-party tools has outstripped the design of existing subscriptions, so capacity is being managed carefully and prioritized for product and API customers.”
GenAI PM Daily April 04, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 17 insights for PM Builders, ranked by relevance from X, Blogs, and LinkedIn. Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. #4 𝕏 Boris Cherny explains that a surge in Claude usage from third-party tools has outstripped the design of existing subscriptions, so capacity is being managed carefully and prioritized for product and API customers. #6 𝕏 Boris Cherny argues that subscription plans are optimized for specific usage patterns rather than raw token counts. He notes that building at scale involves trade-offs that prioritize certain use cases over others.
“#23 𝕏 Boris Cherny says Claude Code Review automatically catches over 99% of bugs, leaving engineers to only perform a quick sanity check.”
#23 𝕏 Boris Cherny says Claude Code Review automatically catches over 99% of bugs, leaving engineers to only perform a quick sanity check. #24 𝕏 Anthropic introduced Claude Code auto mode, a safer middle ground that uses trained classifiers to automatically approve or reject code-generation requests instead of relying on manual permission prompts.
Related
A coding environment for Claude mentioned for its keyboard shortcut that opens a full-featured editor for prompt writing. It is highlighted as making long prompts far easier to manage.
The company behind Claude, mentioned as working with Peter Yang and Alex Albert on Claude's next iteration. It is referenced in the context of model design, harness design, and feedback evaluation.
Anthropic's AI assistant/model used here in multiple contexts: as the product being built next, as a system used to cluster feedback into synthetic evals, and as a tool that non-technical staff use.
Anthropic’s latest Opus-class model release with a 1 million-token context window. It is positioned for long-context planning, coding, and agentic task execution.
A pattern for answering questions by retrieving relevant context and generating responses from it. The newsletter highlights multimodal RAG for searching across audio, image, and video data.
A large language model used here to generate a corpus for retrieval evaluation. In AI PM contexts, it is relevant as a model choice for content generation and analysis tasks.
A Claude model version referenced for more intelligent outputs with higher token usage. It is discussed alongside Opus 4.6 and effort settings for economical runs.
An AI-powered code review feature from Claude Code designed to provide deep PR feedback, catch bugs, and improve development workflows. It is presented as a research-preview beta for Team and Enterprise.
Stay updated on Boris Cherny
Get curated AI PM insights delivered daily — covering this and 1,000+ other sources.
Subscribe Free