No PriorsNo Priors Ep. 92 | With StackBlitz CEO and Co-Founder Eric Simons
CHAPTERS
- 0:00 – 0:39
Sarah introduces Eric Simons and the Bolt.new hype cycle
Sarah welcomes listeners and tees up the central question: what does it mean when a browser-based AI builder grows explosively right after launch? She frames the episode around AI code generation, creative community dynamics, and whether traditional website builders are at risk.
- •Eric Simons is introduced as StackBlitz co-founder and Bolt.new creator
- •Positioning Bolt as enabling developers, designers, and non-technical users
- •Episode themes: AI code generation, community, and the future of building websites
- 0:39 – 2:02
What Bolt.new is: prompt-to-production full-stack apps in the browser
Eric explains Bolt as a ChatGPT-like interface, but purpose-built to generate and run real full-stack web applications. He describes the ambition to expand software creation from ~25M developers to far more “software composers.”
- •Prompt-based creation of landing pages through authenticated full-stack apps
- •Aims to replace agencies/shops for many common web builds
- •Usage scale: hundreds of thousands of weekly users early on
- •Vision: move from tens of millions of developers to ~100M creators
- 2:02 – 3:29
Why Bolt stands out: WebAssembly dev environments baked into the browser
Eric attributes Bolt’s differentiation to StackBlitz’s WebContainer tech—an “operating system” in WebAssembly running inside the browser. This removes the heavy lifting of provisioning dev environments while keeping iteration fast and low-latency.
- •WebContainers run toolchains and packages directly in-browser (WebAssembly)
- •Avoids server-side dev environment costs/latency
- •Supports real workflows: npm installs, frameworks like Next.js/Vite
- •Frontier models + in-browser runtime = end-to-end app creation
- 3:29 – 6:14
Beyond “GPT wrappers”: why open-sourcing prompts and product code helps
Sarah challenges the “wrapper” critique, and Eric explains why Bolt open-sourced system prompts and significant code. He argues defensibility comes from the complete product experience and speed of execution, not secrecy around prompts.
- •Wrapper risk: model improvements and lab integrations can erase shallow apps
- •StackBlitz’s multi-year WebContainer moat is hard to replicate
- •Open-sourcing prompts is pragmatic (they can be extracted anyway)
- •Open source accelerates innovation via forks/contributions
- •Winning is about end-to-end UX + rapid iteration, not hidden prompts
- 6:14 – 9:43
Community as a growth engine: teaching users how to use non-deterministic AI
Eric describes community not as marketing, but as an essential education layer for AI tools. Because prompting is a learned skill and model outputs vary, power users become the best teachers—and reduce churn.
- •AI tools require user education; prompting is not like deterministic software
- •Community knowledge-sharing improves retention and reduces churn
- •Power users often outpace the company in workflow sophistication
- •StackBlitz brings power users onto livestreams to share tactics
- •Parallel to Midjourney: community-driven best practices create advantage
- 9:43 – 13:16
Evals are broken for real products—Bolt Local becomes a practical benchmark
They discuss the gap between academic/standard coding evals and real-world app-building. Eric explains how the open-source “Bolt Local” is being used to test new code-generation models in a realistic product setting.
- •Existing coding eval suites aren’t representative of building real apps
- •Need evals that cover end-to-end tasks (auth, tooling, deployment patterns)
- •Bolt Local is used by communities (e.g., model testers) to compare models
- •“Can it run Bolt?” becomes a proxy benchmark, like “Can it run Crysis?”
- •Benchmarks emerge from real usage rather than synthetic tasks
- 13:16 – 16:57
Real-world outcomes: startups built on Bolt and dramatic cost/time compression
Eric shares concrete examples of people building and monetizing products quickly with Bolt. The stories highlight large reductions in cost and delivery time compared to traditional contracting and agency workflows.
- •Early Bolt-built startups are already charging via Stripe and launching publicly
- •Example: viralhooks.ai built by a PM in Thailand (prompt-to-product)
- •Upwork quote vs Bolt: ~$5,000 and 2–3 months vs $50 plan and ~2 weeks
- •Example: ChilledCRM built by an agency veteran; $30k quote vs $200/month plan
- •Bolt enables “arbitrage” for agencies: faster delivery while charging similarly
- 16:57 – 21:09
Why engineers are rethinking no-code: the Sonnet 3.5 tipping point
Sarah raises skepticism about legacy no-code platforms and ecosystem lock-in. Eric argues a new technical threshold—especially with Sonnet 3.5—makes prompt-first creation viable while still producing standard, developer-friendly codebases.
- •Legacy no-code required WYSIWYG GUIs, causing lock-in and limited extensibility
- •Earlier 2024 models weren’t reliable enough; Bolt idea was shelved temporarily
- •Sonnet 3.5 marked a quality inflection for production-grade app generation
- •Prompting beats complex site builders in simplicity (a textbox vs configuration)
- •Outputs are mainstream frameworks (Next.js/Remix/Astro/Vite), avoiding dead ends
- 21:09 – 24:20
Hybrid workflows and new user segments: non-technical builders + pro dev handoff
Eric explains how Bolt creates a continuum: non-technical users can start, and professionals can later debug/extend the same code. He also notes how tools like Cursor attracted non-developers—and why comparisons between tools can be misleading.
- •Non-technical majority: entrepreneurs/PMs excel because “managing AI” resembles managing devs
- •Community enables hiring help by the hour for debugging and expansion
- •71-year-old mother successfully built and launched a first website
- •Developers can move projects between Bolt and tools like Cursor
- •To non-technical users, multiple AI coding tools blur into “it helps me build”
- 24:20 – 31:07
The long StackBlitz journey: childhood co-founders, WebContainers, and the Figma parallel
Sarah prompts the “overnight success after years” story. Eric recounts founding with childhood friend Albert Pai, the early pain of learning to code, and the insight that browsers became powerful enough to host full dev environments—similar to Figma’s origin story.
- •Eric and Albert Pai learned web dev together as teenagers via O’Reilly books
- •2016–2017 insight: browsers can run OS-like capabilities in a tab
- •Goal: “use the web to build the web” with shareable, web-native dev environments
- •Figma analogy: browsers became powerful enough to enable entirely new product classes
- •Multi-year build: WebContainers took ~4 years; StackBlitz reaches millions of devs/month
- 31:07 – 35:05
Founder grit and personal extremes: AOL-building survival to Ironman during launch
Sarah revisits Eric’s early Silicon Valley survival story and contrasts it with Bolt’s revenue success. Eric then explains why he chose to run an Ironman while navigating a newborn and an intense product scaling moment—using physical challenge to counterbalance mental stress.
- •Early SV days: Imagine K12/Y Combinator path, living off office access and leftovers
- •Company faced uncertainty months before Bolt; experimentation to find the growth product
- •Newborn in April plus market headwinds created a high-stress period
- •Eric commits to marathon/Ironman as a coping and intensity strategy
- •Ironman completed shortly after Bolt launch while the product scaled rapidly
- 35:05 – 37:35
Near-term predictions: from tab-completion to agentic “software composers”
Eric forecasts a shift from line-level autocomplete toward higher-level instruction and agentic workflows. He also argues code generation will keep improving rapidly because software offers clearer feedback loops for training and evaluation than many other domains.
- •Move from tab-complete to higher-level delegation (“go do X, Y, Z”)
- •“Software composer” as a new developer archetype enabled by agents
- •Frontier labs will keep pushing code-gen via scalable data + deterministic checks
- •Software is evaluable: code runs or fails; UIs can be captured/analyzed
- •Eric expects rapid improvements over the next 6–12 months
- 37:35 – 38:04
Closing and where to follow No Priors
Sarah wraps the conversation and thanks Eric. The episode ends with subscription and discovery information for the podcast and its transcripts.
- •Episode wrap-up and thanks
- •Where to find No Priors on Twitter and YouTube
- •Podcast subscription options (Apple, Spotify, etc.)
- •Transcripts and email signup at no-priors.com
