Overview

• Codo (formerly Codium AI) has evolved from a unit testing tool to a comprehensive testing platform with 1 million installations, now monetizing through team and enterprise offerings with specialized AI agents for different testing functions.

• Bolt.new from StackBlitz launched in October using Claude Sonnet 3.5 and achieved $4 million ARR in its first month by reimagining software development for both experienced developers and non-technical users, with some startups already launching entire products on the platform.

• The core innovation behind StackBlitz is Web Containers technology—a custom, lightweight operating system (around 1MB) that runs inside browsers, mapping Linux syscalls to WebAssembly implementations and enabling backend functionality without additional infrastructure costs.

• Bolt's business model rapidly evolved from an initial $9 plan to tiered pricing ($20-$200) with usage-based billing, targeting both casual users and power users (agencies/freelancers), while shifting from enterprise to B2C focus with plans to create an all-in-one solution for website building and hosting.

Content

Podcast Introduction and Background

• Itamar Friedman (Codo CEO) returns as a guest/co-host alongside hosts Alessio (Decibel Partners) and Swix (Smalley)
• Eric Simons (StackBlitz CEO) joins as the new guest to discuss Bolt.new
• Latent Space is hosting a Live Industry event at NeurIPS 2024 with various tech categories and speakers
• Both Itamar and Eric are second-time founders
• Codo (formerly Codium AI) recently raised a $40 million Series A and underwent a rename

Codo's Evolution and Strategy

• Codo focuses on software development, particularly testing
• Started with one IDE extension for unit testing and expanded to support more testing types
• Grew from bottom-up approach to reaching 1 million installations
• Began monetization in 2023-2024 with team and enterprise offerings
• Developed multiple specialized AI agents: Peer Agent, Codo Merge, Cover Agent
• Believes solving testing is key to solving software development
• Targeting enterprise software development with specialized agents rather than general-purpose tools
• Market observation: general-purpose AI agents are losing relevance while specialized agents are more valuable

StackBlitz and Bolt.new Introduction

• Bolt.new launched on October 1st using Anthropic's Claude Sonnet 3.5
• Achieved $4 million ARR in first month, repeated/exceeded in second month
• Initially attempted to build with earlier AI models in January/February but found them insufficient
• Created a new product interface specifically designed for AI code generation
• Bolt targets two distinct user groups:

• The tool has been compared to Cursor, but the founders argue Bolt aims to fundamentally reimagine software development rather than evolve traditional IDEs
• First startups have launched entire products using Bolt, with some users making their first money online using the platform

Eric's Background and StackBlitz Development

• Eric and his co-founder learned coding together at age 13 in mid-2000s
• Had to purchase O'Reilly programming books since online resources were scarce
• Identified a key challenge in coding: setting up local development environments
• Founded StackBlitz with a vision to create a web-based development environment
• Leveraged new browser APIs like WebAssembly and service workers
• Goal was to enable building web apps directly in the browser
• Primary focus on non-developers who struggle with technical setup
• Aimed to simplify development like Figma simplified design and Google Docs simplified word processing
• Demonstrated success with examples like a 71-year-old creating her first website
• Integrated simple deployment with Netlify, highlighting a one-button deployment process

Web Containers and Technical Innovation

• Created Web Containers, a technology designed to run an operating system inside a browser
• Developed a custom, purpose-built operating system for browser environments
• Key technical achievements include:

• Long-term goal is to enable "agentic loops" where code can run, detect errors, and self-correct
• The technology allows running backends like Git, Node.js, and potentially Python in a browser environment
• Unique selling point is the ability to serve one or a million users without additional infrastructure costs

Browser Evolution and Web Container Development

• In the late 1990s, there were two competing visions for the web:

• The web has evolved to incorporate both perspectives with new technologies like WebAssembly and Web GPU
• In 2017, developers discovered the potential of service workers to run web servers within browsers
• By 2021, they developed the first Web Container in partnership with Google
• The team worked closely with Chromium developers, filing numerous bug reports
• Their work has helped improve Chrome DevTools and browser reliability

Error Handling and Technical Infrastructure

• Custom web container infrastructure allows for more precise instrumentation and error tracking across different operating systems
• Their system can capture errors at multiple levels (process, runtime) more reliably than local development
• Bolt has an error capture and resolution mechanism that includes:

• Development methodology focuses on breaking complex tasks into smaller, manageable subtasks
• The product was still in early stages (existed for only 90 days at time of recording)

Enterprise Deployment Challenges

• The company made their system prompts and code publicly available for transparency
• Key technical deployment challenges in enterprise environments include:

• Enterprises have highly varied and complex networking environments
• Some enterprises have hybrid infrastructures (e.g., AWS cloud + on-premises GitHub)
• These complexities significantly impact technology architecture and vendor choices

AI Models and Development Approach

• The company has developed four specialized AI models:

• Current AI models are not true "system two" thinkers
• Breaking complex tasks into smaller subtasks improves model performance
• Different models require different prompting strategies
• Need to support multiple AI models (O1, Sonnet, Gemini)
• Some enterprise clients require air-gapped or specialized deployments
• Problem decomposition is crucial for effective AI model utilization

Enterprise Software Development Challenges

• Managing large numbers of microservices (tens of thousands of repositories)
• Difficulty for developers in finding and navigating code repositories
• Need for sophisticated indexing beyond basic search
• Recommended allowing tech leads to influence repository indexing
• Marking repositories with quality indicators (high/low quality, deprecation status)
• GitHub Copilot Enterprise observations:

• Enable developer platforms to have controllability and influence over AI code suggestion systems
• Incorporate organizational best practices into AI model training and suggestions

Open-Source Strategy and Business Considerations

• Transparency is valuable in the AI space, which is often characterized by excessive hype
• Open-sourcing can motivate internal teams by exposing them to competition and community feedback
• Confident companies are more likely to open-source their work, believing they can still outperform competitors
• Reference to Geohot (of Comma AI) suggesting companies should only open-source if they're confident in their ability to "crush it"
• Practical benefits of open-sourcing observed:

• Bolt has an open-source strategy with a free version and a premium/pro version
• WebContainer technology is partially open-sourced to encourage AI browser experiences

Business Strategy and Growth

• Initially expected modest revenue (couple hundred K ARR) for Bolt.new
• Actual revenue has significantly exceeded expectations
• Currently, revenue is heavily skewed towards Bolt compared to WebContainer
• Strategic options being considered:

• Six weeks in, revenue continues to grow without signs of slowing
• Comparing themselves to OpenAI's strategy of having a consumer product (ChatGPT) alongside API services
• AI labs and startups showing interest in WebContainer technology
• Potential for sandbox environments in chat applications

Rapid Growth and Product Improvements

• Bolt has experienced rapid growth, increasing from around 100K-200K ARR per day to approximately half a million ARR per day
• Product improvements and updates to underlying models (like Sonnet) have contributed to increased conversion rates
• Significant word-of-mouth growth, particularly through platforms like TikTok and YouTube
• The team has been working intensively, including working through weekends to improve the product
• An "emergency council of advisors" was convened post-launch to help address challenges
• StackBlitz grew from zero to 20-30,000 customers in six weeks
• The company is committed to growing the team slowly and intentionally
• They recognize the need to staff up, particularly in customer support and engineering

Pricing and Business Model Evolution

• Initially offered a $9 plan that was unsustainable due to high inference costs
• Quickly adjusted pricing by:

• Usage-based billing accounts for 20-30% of company revenue
• Power users include web development agencies and freelancers
• Emerging model suggests users will likely be either:

• Case study example: A product manager in Thailand used Bolt to build a viral TikTok hook analysis app

Business Model Transition and Future Direction

• The company is shifting from an enterprise software model to a B2C approach
• B2C requires different operational strategies, particularly around analytics
• First priority is hiring a data scientist to understand customer segments, analyze churn, and perform data segmentation
• Currently selling primarily to non-developers, which is a significant shift for the company
• Product vision includes creating an all-in-one solution that handles domain registration, website building, deployment, and hosting
• Currently JavaScript/Node.js focused with minimal support for Python and C++
• Exploring white-label hosting solutions with Netlify
• Early user testing revealed the critical importance of backend functionality
• Considering "Bolt base" - a potential integrated backend solution

Founder Mindset and Personal Insights

• Eric completed an Ironman triathlon in October with only 3 months of preparation
• Total race time was 12 hours and 15 minutes
• Had never been an endurance athlete before
• Hired a coach who was initially skeptical about his ability to complete the race
• Approach mirrors startup mentality of pushing boundaries
• Views soliciting expert advice as part of the founder's process
• Willing to challenge conventional wisdom
• Listens to expert advice but also trusts internal conviction
• Draws parallels between preparing for intense athletic challenges and building products
• Emphasizes the importance of assessing base reality, making personal assessments, and dealing with uncertainty