Podcast Summary: The MAD Podcast with Matt Turck

Episode: Anthropic's Surprise Hit: How Claude Code Became an AI Coding Powerhouse
Date: August 7, 2025
Guest: Boris Czerny, Creator of Claude Code at Anthropic

Episode Overview

In this packed and approachable episode, Matt Turck (Partner at FirstMark Capital) interviews Boris Czerny, the creator of Claude Code at Anthropic. Claude Code is described as an “agentic” coding AI that runs in the terminal, and has quickly become one of the fastest-growing software products ever, widely praised for its transformative impact on software engineering. The conversation explores the accidental origins of the product, its unique CLI-first approach, the technological philosophy behind it, concrete use cases, and broader implications for the future of coding and AI-powered engineering workflows.

Key Discussion Points & Insights

The Accidental Birth of Claude Code

• Origins: Initially a personal prototype by Boris to automate notes and music control via the Anthropic API; coding wasn’t even a feature at the start.

  "Quad code didn’t code. It was called Quad CLI at the time... I used it to automate my note-taking. So it kind of controlled my notes app and it controlled my music player to kind of play music for me..." — Boris (02:47)

• Breakthrough: Once given “Bash” (command-line) tools, the model started coding competently, surprising the team with its natural fit for terminal-based interaction.

  "Immediately the model just went. It just started coding and it was just the craziest thing... it kind of knew, okay, okay, I can write AppleScript and I can automate stuff..." — Boris (03:00)

• Widespread Adoption: Became a daily-active tool for nearly all engineers at Anthropic soon after launch.

Agentic Coding & Modern Software Engineering

• Definition and Evolution:
  • Traditional coding: direct text manipulation in IDEs—manual, like editing in Microsoft Word, for 50+ years.
  • Agentic coding: engineer describes desired code changes; the model plans and executes them.

  "The way that it works is a human describes to the model the change that they want and then the model is the one that manipulates the text... more and more of coding is just going to be the model doing it and a human will have to intervene less and less." — Boris (06:14)

• CLI vs. IDE:
  • CLI (terminal) is universally compatible and simple.
  • Chosen as the initial interface because it minimized dependencies on platform or user preferences, facilitating rapid prototyping and broad adoption.

  "Start with a simple thing first. So this kind of made sense... the big one is that it kind of works anywhere. So it doesn't matter what kind of system you're on..." — Boris (10:35)

• Philosophical Approach: Build the minimal viable product that lets you “feel” model advancements as they happen; interface simplicity outweighs elaborate UI design in the current paradigm.

Why Claude Code Works — Model Advantages

• Model Choices: Supports Sonnet 4, Opus 4, and Haiku. Users can choose which model to use (13:31).
• Why So Good at Coding?
  • Deep integration between product and Anthropic’s researcher mindset: “the model should be able to do what programmers do.”
  • Coding is seen as a natural “interface” for model intelligence to affect the world, tying into AGI/ASI aspirations.

  "There's also something about it where maybe coding is the way that we get to the next level of intelligence... for a model, the natural way is code." — Boris (14:13)

• User Focus: Targeted at professional engineers but rapidly attracting new-use cases—even people outside coding, like designers and PMs, are using it for automation and workflow tasks.

Agentic Features and Subagents

• Definition of ‘Agentic’: Model proactively uses tools (like file reading, web searching) to accomplish goals, chaining actions “like a coworker.”

  "There's a newer kind of application... you send them a message and then they might do a little bit more. So we call this tool use... this is kind of the essence of being an agent..." — Boris (18:12)

• Actions: Reads/writes files, runs system commands, even performs internet searches—with human-in-the-loop safeguards.
• MCP Tools: “Model Context Protocol” enables seamless integration with organizational tools (Jira, Slack, etc.), making Claude Code extensible and customizable (23:10).
• Subagents: New feature allowing users to define multiple specialized “mini-Claudes” with different prompts and tool access, essentially creating a multi-role agent workforce within a session.

  "Subagents are just other quads, and they're prompted a little bit differently... you can customize what prompts they have, you can customize what tools they have..." — Boris (24:00)

Productivity, Learning, and Memory

• Early Impact: Drastically reduced onboarding times—engineers productive in days, not weeks.

  "Technical onboarding used to take a few weeks, but now engineers are usually productive within the first few days." — Boris (45:36)

• Built-in Team Knowledge: Uses simple, shared text files (“CLAUDE.md”) as editable, team-wide memory banks for context and learnings, fostering communal intelligence (29:35).
• Manual vs. Automatic Memory: Automatic memory is in internal testing; currently, memory entries are declare explicitly to avoid “bad remembrance” or forgetting critical details (33:39).

Safety, Autonomy & Interface Philosophy

• Human-in-the-loop: All non-trivial actions require human approval; policies can be defined for “allowlist/blocklist” to streamline repeated approvals (35:48).

  "For actions that we know can't have any kind of dangerous repercussions... But for other actions, like editing a file or running a command or using the Internet, this always needs a human in the loop..." — Boris (35:49)

• Enterprise-Ready: Easily integrates with regulated and on-prem environments; only requires API access, no third-party integrations like code indexing databases (37:11).
• Terminal UX: Significant design attention paid to making the CLI interface delightful and clear—fun touches like randomized status verbs (“schlepping”, “herding”) and intuitive status indicators (39:54, 40:23).

  "At this point, most code at Anthropic is written using Quad code, and almost everyone at Anthropic is using it every day." — Boris (38:20)

Pricing Model

• Two main subscription tiers (as of August 2025):
  • Pro ($20/mo), Max ($100–$200/mo), with generous rate limits.
  • Additional API-based “pay-as-you-use” for limitless power users.

  "Some people have this army of quads, you know, like 5, 10, 20, that are just running in parallel all the time and just doing work..." — Boris (40:58)

Use Cases & Workflow Examples

• Breadth of Automation:
  • Planning, coding, debugging, testing, managing tasks, incident response, onboarding, and more.
  • Particularly excels at codebase research and new engineer onboarding; rated “10 out of 10” by Boris for that (45:23).
• Daily Usage Patterns:
  • Boris’ own usage: code research, automating small/medium feature work, prototyping complex features by running multiple quads in parallel and merging the best results, iterating on plans with Claude before implementation (46:43).
• Onboarding Example:

  "This is part of technical onboarding. We teach them, ‘Here's quad code, here's the code base. If you have any questions, don't bug engineers on your team. Just ask quad code and they can answer these questions probably better than they can...’” — Boris (45:36)

The Broader AI-Coding Ecosystem & Competition

• Market Dynamics:
  • Sees a huge space for multiple differentiated solutions.
  • Encourages building for model capabilities six–twelve months in the future, given the rapid pace of LLM advancement.

  “My advice to companies building is definitely build for what the model will be able to do six months from now, not for what the model can do today.” — Boris (51:39)

• Ecosystem Evolution:
  • Predicts many more applications will be built on top of AI APIs and SDKs than in-house vertical stacks, due to sheer diversity and scale of need (54:58).

The Future for Coders and Coding as a Profession

• Paradigm Shift:
  • Sees agentic programming as the next big transformation—akin to the leap from punch cards to high level languages.
  • Coding will feel more like orchestrating and reviewing agents’ work than low-level text editing, greatly expanding who can build software.

  "Programming is no longer direct text manipulation, but it's more working with agents to get the work done. And I think it's going to be hugely empowering where a lot of people that couldn't create before can now create..." — Boris (56:37)

• Advice for Future Developers:
  • Learn both: still understand languages/frameworks/system design, but also learn to leverage and direct AI tools.

  "You have to code so that you can check what the model does and you know how to direct it... you have to be using quad code, and you have to be using all these new agent encoding tools, because this is what the future is." — Boris (57:44)

Product Roadmap

• Upcoming Features:
  • Native Windows support, single-file distribution (no Node.js required), more platforms/interfaces, increased agent autonomy, more action flexibility, and continuous UX improvements.

  "...working on getting quad code into more places, so wherever you are, you can use quad code more easily, the same way that you can on GitHub today. And expect a lot more agents..." — Boris (58:48)

• Exploratory Approach:
  • Iterating quickly, experimenting and shipping what works best for users.

Notable Quotes & Memorable Moments

• On the surprise of the product’s appeal:

  “Are people going to like this? Is it going to be that useful? Can you actually use it for a lot of coding? We had no idea.” — Boris (02:19)

• Boris on the radical productivity boost:

  “Anthropic Technical onboarding used to take a few weeks, but now engineers are usually productive within the first few days.” (00:50, 45:36)

• On UI design in the terminal:

  “Terminals have been around for 50+ years at this point... we’re rediscovering how to design for a terminal... we spent probably 30 or 40 iterations [on the spinner].” (38:19)

• On the broader future:

  “It's going to change programming, where programming is no longer direct text manipulation, but it's more working with agents to get the work done...a lot of people that couldn’t create before can now create.” (56:37)

• On product strategy:

  “Build for what the model will be able to do six months from now, not for what the model can do today.” (51:39)

Timestamps for Key Segments

• Product Origins and Adoption: [02:00]
• Agentic Coding Defined: [06:14], [18:12], [21:48]
• Terminal vs. IDE Approach: [08:23], [10:35]
• Subagents Overview: [23:47]
• Memory and Team Knowledge: [29:35], [33:39]
• Human-in-the-loop Safety: [35:48]
• UI/UX Details: [38:19], [39:54]
• Pricing Structure: [40:40], [43:12]
• Use Cases and Workflows: [43:49], [45:23], [46:43]
• Future of Coding Profession: [55:37], [57:44]
• Product Roadmap: [58:48]

Concluding Thoughts

Boris and Matt finish on the note of active exploration: Anthropic’s team is testing rapidly and taking a “show, don’t tell” approach—shipping what delights users as AI’s capabilities accelerate unpredictably. The episode offers both a deep technical dive for software professionals and accessible analogies for general technologists, capturing a snapshot of a transformative moment in the future of software development.