Bolt expected developers to dominate its user base, but Simons says roughly two-thirds of users were not developers. The most successful nondevelopers often resemble strong product managers: they understand enough technology to define scope, communicate executable requirements, judge quality, and help an agent debug. As code generation lowers implementation cost, deciding what to build and recognizing whether it is good become more valuable.

About 67% of Bolt users were not developers
Strong PM skills overlap with directing an AI developer agent
Scope definition and clear communication reduce agent failure
Product taste matters more when producing a first implementation is cheap
Engineering and design knowledge still increase a PM's leverage

“the people we see they're most successful are are people that are amazing PMs, for example.”
Eric Simons

“It's basically that. You really want to be good at defining scope and uh helping it go and debug various things or whatever have you.”
Eric Simons

#product-management#ai-coding#product-taste#future-of-work

Hot Take

The Product Org Chart AI Coding Could Rewrite

Simons expects AI coding to move more user-interface work from engineers to product managers and designers. In his forecast, PMs and designers directly craft experiences, while a smaller number of engineers review code and focus on hard technical problems that language models do not handle well. The likely unit is a larger product-and-design group paired with one or two highly leveraged engineers.

PMs and designers may directly implement more interface changes
Engineers can review generated code and guide technical quality
Developers shift toward intellectually difficult work that models handle poorly
Frontend engineering leverage rises sharply
Successful software companies may reorganize into different product pods

“The winners at least, the org charts are going to completely change in how they approach building products and shipping products completely”
Eric Simons

“the leverage that the front end engineer is gonna have is is in it it it is now insane”
Eric Simons

#org-design#engineering#product-teams#future-of-work

Explainer· 2

Explainer

Why Bolt Runs Its Development Environment in Your Browser

Bolt's core advantage comes from WebContainer, an operating system StackBlitz built to run locally inside the browser. Competing cloud development tools need a virtual machine for every active user, adding boot time, cost, reliability problems, and abuse risk. Browser-local compute uses the customer's CPU, which supports fast startup and a more permissive free tier.

WebContainer runs a development operating system inside the browser
The environment can boot quickly without waiting for a cloud VM
Local compute shifts transient execution cost away from StackBlitz
A free tier is easier to sustain when users cannot abuse company-owned servers
Consumer devices are far more abundant than rentable cloud virtual machines

“there's not even a hundred million VMs to rent on the planet. But there are like a billion devices that you can run this stuff…”
Eric Simons

“there would not be a bolt without this technology called web container that allows us to run an operating system in the browser.”
Eric Simons

#webcontainer#browser#infrastructure#scalability

Explainer

Bolt Builders Flips the Copilot: AI First, Human When Stuck

StackBlitz created Bolt Builders because nondevelopers still encounter problems an AI agent cannot solve. The planned service connects users with certified experts who can diagnose the issue for an hourly fee, then return the project to the AI-led flow. Lenny describes the inversion: instead of a human programmer receiving occasional AI suggestions, the AI builds most of the application and calls a human for a difficult intervention.

Certified human experts help when the AI cannot resolve a problem
Users pay for targeted debugging rather than an entire development project
After the intervention, the user continues building through prompts
Engineers spend more time on difficult debugging and less on repetitive application code
The copilot relationship changes from AI-assisting-human to human-assisting-AI

“there's gonna be a button where you can just say, hey, connect me with you know, uh a certified like expert.”
Eric Simons

“But but I love that it's like now AI first and person second versus person building the thing and then AI.”
Lenny Rachitsky

#human-in-the-loop#debugging#ai-agents#services

Story· 5

Story

Bolt Went From Near-Death to Almost $40M ARR

StackBlitz expected Bolt might add $100,000 ARR over several months. Instead it reached $20 million ARR in its first two months and was approaching $40 million roughly four to five months after launch, while adding about three million registered users. The surge exposed missing basics: customers exhausted the only $9 plan, servers struggled, and even Anthropic could not supply more GPU capacity.

The company was close to shutting down before Bolt launched
Bolt reached $20 million ARR in its first two months
The team was approaching $40 million ARR roughly four to five months in
Customers forced a rapid redesign of pricing because they wanted to buy more inference
Infrastructure demand exceeded both StackBlitz's plans and Anthropic's available capacity

“what ended up happening is uh, yeah, in the first two months, we you went from zero to twenty million of ARR.”
Eric Simons

“Anthropic ran out of GPUs for us.”
Eric Simons

#hypergrowth#arr#launch#pricing

Story

The Early Figma Demo That Inspired StackBlitz's Seven-Year Bet

Before Figma had a design product, Dylan Field and Evan Wallace demonstrated a 3D ball falling into water inside a browser tab. Their thesis was that new browser capabilities made a high-performance graphics engine possible, even though it would take years to build. Simons and his co-founder Albert later recognized the same pattern in web development and bet that WebAssembly and related APIs could support a complete browser-based development environment.

Figma began as a browser deep-technology thesis, not a finished design tool
New browser APIs created capabilities that previously seemed impractical
StackBlitz applied the same pattern to development environments
The founders accepted that the enabling system would take years to build

“Figma was was also a browser-based deep technology play.”
Eric Simons

“Their first pitch was this uh 3D ball dropping into water inside of a browser tab.”
Eric Simons

#figma#deep-tech#webassembly#innovation

Story

A Nontechnical Founder Built a Selling CRM in Three Weeks

One of Bolt's first customers, a nontechnical entrepreneur named Paul, built a CRM with AI features, Stripe billing, and paying customers. An agency had quoted roughly $30,000 and six months for the work. He completed it in about three weeks and spent approximately $300 on Bolt, illustrating the scale of cost and time compression for greenfield software.

The founder did not know how to code
The product included AI and Stripe billing
The agency estimate was about $30,000 and six months
The Bolt build took about three weeks and $300
Customers were paying to use the finished CRM

“He had gotten a quote from an agency for this. It was like gonna be 30 grand and takes six months. He had it done…”
Eric Simons

“he's and he's making money off of this. Like this is his startup, right?”
Eric Simons

#no-code#crm#entrepreneurship#cost-reduction

Story

The Model Upgrade That Took Bolt Off the Shelf

StackBlitz tried to build Bolt with the frontier models available about a year earlier, but the generated code was too unreliable and visually poor. After seeing an early version of Sonnet in May, the team revived the project and built Bolt in about 90 days. Simons sees Sonnet as the moment coding agents crossed from autocomplete-style assistance into reliably producing functioning, attractive applications.

An earlier Bolt prototype failed because model output was unreliable
A preview of Sonnet convinced StackBlitz to restart the project
The team built the launch version of Bolt in about 90 days
Simons links the wider surge in AI coding products to Sonnet's release
The AI now produces most of an application before a developer intervenes

“we should take that project back off the shelf and green line it because this this might be it.”
Eric Simons

“a very critical threshold has been passed with LLM's ability to write production grade code and apps that actually look beautiful and actually function well.”
Eric Simons

#sonnet#anthropic#ai-coding#model-capabilities

Story

Eric Simons Lived in AOL's Office on One Dollar a Day

At nineteen, Simons came to Silicon Valley for an education startup backed by Imagine K12. When the company's small funding allocation ran out, he noticed the AOL office had couches, ramen, showers, and laundry, so he quietly lived there for four or five months while coding. Security guards on different shifts each assumed he simply worked unusually long hours until a guard removed him at four in the morning.

Imagine K12 participants worked from office space inside AOL
The startup's roughly $20,000 allocation ran out within months
AOL's couches, food, gym, showers, and laundry made the office livable
Opposite security shifts mistook his continuous presence for extreme work ethic
His stated personal burn rate was one dollar per day

“My burn rate was a dollar a day.”
Eric Simons

“the guys at night just were like dang this guy works really hard you know and and then in the morning they'd be like wow…”
Eric Simons

#founder-story#silicon-valley#scrappiness#aol

Q&A· 1

Q&A

Where Bolt Works Today—and Where a Large Codebase Still Wins

Simons draws a clear boundary around current AI app builders. They can produce production-grade greenfield software, public websites, marketing pages, and high-fidelity prototypes, and can sometimes work with smaller repositories. A very large existing application remains a poor fit because current models struggle to reason reliably across thousands of files; those teams still need developer-oriented tools and experienced engineers.

Greenfield applications can be built to production quality
Existing companies can use Bolt for sites, admin tools, and prototypes
Repository support exists, but usefulness depends on application scale
Models become less reliable as a codebase grows to thousands of files
Large legacy applications still require developers and tools such as Cursor

“if you have a large existing code base, you're gonna need something like cursor.”
Eric Simons

“today with this current state of frontier models, you could absolutely build production grade software, right?”
Eric Simons

#codebases#greenfield#prototyping#limitations

Takeaway· 1

Takeaway

Learn Technical Foundations—Not for a Guaranteed Job

Asked what a child should study, Simons recommends understanding how to leverage AI tools and learning the foundations of programming and systems. He does not treat a computer-science degree as a guaranteed route to employment, especially when it can require six-figure debt. His advice is to follow intrinsic interest, go deep when the subject genuinely matters to you, and use the abundance of free online learning before making a large commitment.

Knowing how to leverage AI tools is a foundational skill
Basic programming and systems knowledge remains useful
Deep specialization still makes sense when driven by genuine interest
A degree should not be purchased on the assumption that a high-paying job is guaranteed
Free exploration can precede a major education commitment

“I understand how to leverage these AI tools um is key.”
Eric Simons

“it costs nothing to go and explore and learn for free online, you know.”
Eric Simons

#education#programming#ai-skills#careers

Frameworks in this episode

Chef's Menu Prioritization

Default-No Until Product Pull

Minimum-Context Executable Brief

More Context, Fewer Hands

The Determinism Test for AI Verticals

Insights & moments

◆Hot Take· 2

Why Product Managers May Be the Best AI App Builders

The Product Org Chart AI Coding Could Rewrite

✶Explainer· 2

Why Bolt Runs Its Development Environment in Your Browser

Bolt Builders Flips the Copilot: AI First, Human When Stuck

❝Story· 5

Bolt Went From Near-Death to Almost $40M ARR

The Early Figma Demo That Inspired StackBlitz's Seven-Year Bet

A Nontechnical Founder Built a Selling CRM in Three Weeks

The Model Upgrade That Took Bolt Off the Shelf

Eric Simons Lived in AOL's Office on One Dollar a Day

?Q&A· 1

Where Bolt Works Today—and Where a Large Codebase Still Wins

▲Takeaway· 1

Learn Technical Foundations—Not for a Guaranteed Job

Hot Take· 2

Explainer· 2

Story· 5

Q&A· 1

Takeaway· 1