Eric Simons, CEO of Bolt, on consumer vs. B2B vibe coding

We've been following the rapid evolution of AI coding from the days of autocomplete in GitHub Copilot to the AI-IDE wave in Cursor ($500M ARR in May 2025) and the rise of text-to-app vibe coding tools like Bolt, Lovable, and Replit.

To learn more about where vibe coding is heading as the market matures, we reached out to Eric Simons, CEO of Bolt / StackBlitz.

Key points from our conversation via Sacra AI:

Let's talk about StackBlitz and WebContainers. Why was this key infrastructure, and why is it an important piece of differentiation for Bolt?

Absolutely. WebContainers was the technology we founded the company to build. It seemed inevitable that being able to build web apps on the web was going to be an important capability.

We spent the first few years of the company building technology that lets you run a VM inside your browser tab. That's what powers Bolt. When you type in a prompt and hit enter, we spin up an environment in your browser that costs us nothing to run.

There are two key benefits: one, it's instant because it runs locally on your machine—no cloud VM required. These zero latency, instant booting environments provide our customers with a superior user experience. The second is that we don't have to spend money on cloud VMs to run your application.

As it pertains to gross margins, it’s a key point where this really shines. You can reach very large amounts of scale for effectively zero cost. That's why our gross margins have been the best amongst our peers. We're close to traditional SaaS margins—nearly 70% right now. As a company, we are profitable (although that will shortly change as we reinvest aggressively). That's unique relative to other companies in our space. We wouldn’t be close to that without WebContainers.

Let's talk about your user base today. When we speak to companies in this vibe coding space, they tell us they used to serve devs who were learning, but now they're all professional users. If we took a snapshot right now, how would Bolt break down between consumers, solopreneurs, prosumers, small teams, and enterprise?

When we launched Bolt, it exploded overnight. We had hundreds of thousands of customers from all walks of life. The first half of last year, we were experimenting: is this a consumer play? Are we going to take out Wix?

We didn't find the consumer segment compelling to build around, particularly because of the churn rates. When you have large top-of-funnel, you get those users, but there's a difference between orienting for growth of those users and what business you're actually in.

Where we saw classic B2B SaaS retention was with product teams and development teams. That's where our company comes from—we were doing B2B developer tooling. We pulled out that playbook starting Q3 of last year.

Today, PMs, developers, B2B customers, enterprise customers—they're getting close to at least half of our revenue this year, if not more.

Our forecast for this year marks the majority of growth in B2B revenue because that's what we deeply care about growing. We've oriented less toward gathering top-of-funnel at all costs. A good analogy is that we are taking the “Anthropic” approach (majority B2B revenue, focused use cases) while most of our peers are taking the “OpenAI” approach (majority consumer revenue, spraying many fronts).

Roughly speaking, you're close to or on the cusp of surpassing 50% of revenue from users at their companies?

Certainly within the year, we'll be over that mark.

How does that break down between creating apps for internal consumption versus external consumption?

We see both. The primary pull when companies bring us in for pilots is them wanting to ship things externally. That's where competitiveness comes from—how fast they can ship to end customers.

Once that capability is in-house, internal tools become, "Oh, we can use this to spin up internal agents or tools." It's a mixture of what makes us land versus expand.

We're primarily landing with customers to let them use their own custom components and design systems to build their apps and vibe code with those. When you go to Replit or Lovable, they don't let you use what your company has already made.

When you prototype something on Replit, that prototype may prove the concept is good, but engineers have to completely start over because it's not using code that can go into production.

The classic example: if you're MasterCard, you want your PMs prototyping stuff. You certainly don't want the outputs to look like Visa.

That's where we've found a great wedge—companies bring their own coded version of the design system so what comes out of Bolt is the product. It's real code that can go into production.

Is that related to your tweet about "chat with Gemini, code with Claude, build with Bolt"? Can you explain that idea of people working with code and elements in Bolt?

You can swap out things at the top layers. Personally, I've stopped using ChatGPT and switched to Gemini for general inquiries and chat. I use Claude Code when I'm directly punching in something at a very low level.

But I go to Bolt for high-level scaffolding around prototyping ideas to the degree where I can have those landing in my production codebase. Then my engineers might use Claude Code to add certain functionality.

There are abstraction levels around almost the same underlying APIs, but the interfaces and built-in services lend themselves to different types of the product development process.

You're saying there's stuff you can do in Bolt that you can't do in Claude Code. Can you flesh that out? What does Bolt allow that Claude Code doesn't?

The value we provide is better interfaces and context to the end user and for the agent.

If you use Claude Code, you're installing a CLI tool. If you want to use your company's design system, you have to figure out where it lives and ingest the documentation. Claude Code doesn't have a visual conversation where you can point and click at things.

Bolt provides a higher-level abstraction layer. A PM can use it without cracking open a command line tool. The Bolt agent has all the context about the project you're building, how your company sets things up, how your design system works.

We have built-in services for databases, auth, AI, and more. All this is there without manual setup, whereas in Claude Code you'd have to do that yourself.

That's the value we—and frankly every startup outside of OpenAI and Anthropic—provide: better interfaces and better context.

Let's talk about Bolt Cloud. You're bundling hosting, database, auth, payments. Why have such a thick infrastructure layer instead of just focusing on code generation?

We're a classic product-led growth motion. People play with this in their free time, then bring it into their job.

Companies often have infrastructure already set up, so they wouldn't need these things. Sometimes they do, though. When you're an individual building something on the side, you just want to ship a website. You don't want to think about where to buy the domain or what database to use. You just want it to work.

You want to describe your idea to the agent, it makes it, and you collaborate with it as if it's a co-founder.

The most magical Gen AI experiences are when you tie together services and context so the agent can automagically get stuff built and working for you.

Many use cases, especially top-of-funnel, revolve around people building apps in their free time. Even on our B2B side, they're often spinning up prototypes that need to function.

Do you expect B2B enterprise customers will be attracted by Bolt Cloud, or is it supplemental? Why would they host with you versus the gazillion other options?

In many cases, they already have infrastructure set up and Bolt just plugs into that. In some cases they're doing a greenfield project: "We just want to get this up and running into customers' hands as fast as possible." That's the value of Bolt Cloud.

There's nothing to set up on every project. It's there, enabled, and ready to use. It's powered by the same infrastructure proven to scale to public company size.

It's a question of what their specific product needs are. Are they using existing infrastructure? Are they optimizing for rapid iteration without infra? In that case, they'd rather use what's built in because it just works without adding context to the agent or looping in developers.

Do you feel a platform like yours needs adoption of something like Bolt Cloud to succeed commercially, or do you disagree?

Not in the current moment. It strikes me as table stakes today.

But the risk if we didn't have Bolt Cloud is that what people are coming to Bolt for will change a lot over the coming years. Companies are going to want to spin up lots of agents.

The more friction between the Bolt prompt box and getting results—the magic moment is when you hit enter on the prompt and it's there. If you have to then figure out how to connect all your existing stuff, it's not as compelling.

As an AI wrapper like all the others, the value we provide is the product experience, driven by interface, context, and services we give to the agent.

In the long run, there will be experiences companies want to build where their existing infrastructure is a pain to set up, especially for non-technical staff. They'll lean on external solutions, the same way they've adopted things like Retool for internal apps.

Over the long run, it's more of a strategic consideration. Today, it's really more to have a seat at the table in POCs.

The competitive landscape seems to be converging. Lovable, Vercel offer infrastructure and hosting. There's Base44. How do you think about your strengths and weaknesses versus these competitors? How do you find your wedge to grow and take market share?

Outside of Cursor and Claude Code, there aren't many folks who can really drop into an existing B2B environment, plug into existing services, and build real apps for workflows that matter to professional software teams.

Many use cases are throwaway prototypes, which is better than using Figma's original product. But it's not super sticky or retentive. The switching cost is zero between all these things.

Figma Make is adding pressure on the B2B side—if you're doing prototyping with throwaway code, they're shipping that with every Figma license. You have to contend with that if that's all the value you provide.

On the consumer side, Wix is dumping tremendous amounts of money into paid acquisition for Base44. Lovable, Replit, and Wix—regardless of what they say—are really gunning for top-of-funnel revenue from consumers and not going super deep on B2B and enterprise.

There's an OpenAI approach that others are doing: raising extremely large amounts of capital, very consumer-heavy revenue streams.

Then there's the Anthropic approach: doing more with less, better unit economics, really focusing on doing one thing or workflow really well for the B2B market, which means great retention and durability.

We're more in the latter camp. We're looking at Anthropic as a good example in this AI landscape—building great products that businesses actually want to adopt that are very sticky and not getting lost in the ever-revolving consumer landscape.

OpenAI has headwinds to contend with now on the consumer side, whereas B2B tends to be more durable and impervious to immediate compression.

Let's double-click on Replit. They say they're putting tremendous effort into enterprise and B2B—enterprise tier, specific features, security, identity management. You have that too. What else makes Bolt a better B2B enterprise product than Replit, beyond meshing with enterprise context?

It depends on the use cases. Anecdotally and in our sales conversations, we never hear about Lovable. We've heard about Replit a little bit. We've heard about Figma Make.

On Replit, I don't know exactly what their target end user is. From what we've seen, it tends to be more developer-team-oriented, whereas we sell to developers and product teams. We don't see much overlap on the product team side. Maybe there can be multiple companies doing well on the enterprise side. I don't think it's zero-sum.

When we engage with enterprise customers, it comes down to this: enterprises are having a very difficult time figuring out how to adopt this stuff because it's moving at such a rapid rate. It's been astounding talking to some of these folks—we live in AI and agents day to day.

When we drop in to help them get set up, they're often very confused about how to go about it. It's no surprise that some frontier labs are hiring deployed engineers to help make customers successful.

What's going to make or break the enterprise side, regardless of who's where today, is who's got a great organization actually dropping in and helping customers get set up.

There's a big difference between selling a POC or even a one-year license versus them not using it and churning next year. Not saying that's what anyone's current engagements look like, but that's what I'm concerned about.

Back in 2021, enterprises were buying damn near anything they could get their hands on. Then in 2022, they mercilessly cut everything they weren't using. In this cycle, the same could be true. You need to build something that's actually going to get integrated and used. That's what we care about.

Do you see go-to-market usage—creation of demos, internal apps? We've heard about that anecdotally.

A little bit. We're taking more of a focus on the teams building the products for the company. We'll land at product teams or engineering teams. Then sales and marketing get their hands on it and start building things. That's not the beachhead we're going after.

We have a lot of top-of-funnel usage that matches that. But we want to be precise about dominating with professionals whose job is to build professional digital products. That doesn't typically include GTM teams—not to say they're not important. There will be tools better geared to those folks, not that ours wouldn't be sufficient.

Returning to gross margins, you mentioned WebContainers gives you inherent leverage. What other types of margin expansion do you push on? Margins have been talked about a lot in this space.

WebContainers is huge for web-based tools. It removes a multimillion-dollar-a-month cost.

Beyond that, you're looking at inference and how to bring the cost down. There's a lot of engineering work to optimize efficiency around caching and such.

Then you get into taking open-source models and fine-tuning them with your own data, which we're in the game of now. Those are the main ways you can increase margins.

Are you developing your own internal models—something custom to you, whether open-source-derived or not?

Absolutely. The benefit for us is that because we're taking a more focused approach on the types of use cases we care about, it gives us a very targeted surface to train models against.

Why we're the best at taking corporate design systems and allowing them to use their own code is because we're spending a lot of time—everything from building systems that traverse codebases and build documentation to creating models that can best ingest that and then write out code more accurately and cheaply.

Across all ends of that spectrum, those are things we're doing and have done.

How does Bolt's ability to use corporate brand elements and fuse with enterprise context actually work? Is it interface-based where users guide an agent, or does it happen automatically as you onboard projects?

We started doing it manually—onboarding enterprise customers one by one, going to open-source equivalents to test our stuff. Classic "do things that don't scale until you scale them" approach. We did it manually, took insights back, wrote tooling that would traverse codebases and create documentation, started fine-tuning models that could best ingest that.

It's been a six—to nine-month process. But at this point, when we bring in an enterprise customer, they just plug in their codebase. Our agent kicks off its job of traversing everything, understanding how it works, ingesting, documenting, feeding that into our models or the frontiers. Generally speaking, it's ready to rock.

We've got it to almost full automation.

How would you quantify the size of the enterprise business at this point?

In regards to composition of our revenue or market size?

Composition of revenue or growth.

B2B revenue specifically is around 25% of our revenue today, growing rapidly.

Turning to the future: Is Wix acquiring Base44 a symbol of what's ahead? Will companies like yours build on code generation and become giant Wixes or Squarespaces? Does that align with where things are going?

For some, maybe. Coming back to how we're positioning: I look at Lovable as very much the Wix of this thing. Whether they think it or not, their primary revenue is consumers. In their shoes, I'd be purely focusing on Wix, because Wix is the only incumbent that could cause meaningful damage. There's probably an opportunity for them to be Wix 2.0.

But what I see on the B2B and professional tool side—Wix is a consumer prosumer product used by small, medium businesses to build websites. What's most exciting to me is how do you create the next generation of professional grade tools, like Figma in 2015.

That's where the majority of value here will get created. How do you enable people to use sophisticated tooling to build really great apps and help companies do that? The vast majority of the world's software is funded by businesses.

How do you make tools and workflows that supercharge their ability to build great products at a rate faster than ever previously possible?

There will be the Wixes, the Figmas, the Cursors. The Wix side is very much not what we're interested in. It's cool, but we're much more interested in the impact tools like Figma and VS Code have had on professionals over the past decade.

At what point do you need to get into the design end of things? Or does the new architecture mean the design piece happens differently? Where do you start to butt heads with Figma?

We're already there. This is their response with Make—obviously a direct response to Bolt Lovable, Replit.

Figma as a product has always been about allowing designers or PMs to make fake things to prove you should build the real thing. That's the traditional product development process—make wireframes and mockups to prove it's a good idea, then hand to engineers to make.

Our viewpoint with Bolt is that the prototype is the product. Don't make fake things to prove you should build it. Just build it. That's a pretty big paradigm shift.

Even Figma Make is still making fake things. It's not something your engineering team can use. From a DNA level, Figma has never shipped a product about building real things—something actually real going into customers' hands.

Not to say they couldn't make that jump, but it's a very different mindset and shift. It's a classic innovator's dilemma. We don't have the shackles of history tying us to a previous way of life.

For Figma, that's strategically in their blind spot. For us, it's our opportunity as a startup.

We saw people freak out about OpenClaw over the last week or so. As a company with an agentic product moving into enterprise, any takeaways from the adoption that project got?

No lessons we haven't already seen working on agents ourselves. One of the things we're probably going to write a post on: we're one of the only AI startups that's remote. We're not RTO.

It's an interesting paradox. Everyone's saying, "Agents are going to take over the workforce. Everyone will have their own team of agents." Yet you've got 50 people in a room in SF somewhere. In reality, the pitch is you can have a single person, in any room, anywhere, with a team of 50 agents working underneath them.

As a company, we're very much leaning into that. We've capped our headcount. We want more agents than humans at the company, and we want to be the number one company in this space powered by agents—not just engineering, but marketing and sales too.

The results we've seen over the past quarter have been nuts. We're automating crazy amounts of our company today. Over the next six months, it's going to be even greater.

We're taking the most extremist bet on what people are seeing with OpenClaw.

If everything goes right for Bolt over the next five years, what does Bolt become and how do you change the world?

The cutting edge of AI and agents is growing at a crazy pace. Most people employed at companies don't have the skills background or time to learn how this stuff works to leverage it in their jobs.

What we're up to with Bolt is creating more ubiquitous access to leverage this stuff—both in how companies are building their products and delivering to end customers, but more broadly across the company.

How do you deeply embed agents, even custom-built ones, to power your company? We want to help companies become an agentic workforce—bring an agentic workforce to their company.

Today, we're primarily focused on helping you ship better products and leverage agents to eliminate a lot of the fake work required to do your job. But our ambitions are grander. We see that opportunity across many parts of running companies.

That's what I'm most excited about—there's a big gap in usability of this stuff for non-technical people.

You're saying the product you've created for helping folks build software—an aspiration is to take that and apply it to other functions, other tasks?

We think everyone's going to have a team of 50 agents they're managing. We call it an "agentic team." How do we create a product that allows folks to maximally leverage that team of agents underneath them?

Different job functions will mean different things. For the sales team, it's whipping up demos. For the marketing team, it's whipping up landing pages.

But—looking at Anthropic as an example—you want to nail one profession first, one landing point that's very durable, then build on that and expand to everyone else. That's our view of how we'll get there.

This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.