# Why Bolt.new Won and Most DevTools AI Pivots Failed

Victoria Melnikova, Evil Martians | AI Engineer World's Fair 2025 | 05:45

Source: https://www.youtube.com/watch?v=3YRrBFeQ1aw
Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks.
Page: https://aietalks.com/talks/why-bolt-new-won-and-most-devtools-ai-pivots-failed
Published: 2025-06-10
Tags: product-strategy, startups

## TL;DR
- Adding a chatbot to an existing developer tool often produces a feature that users try once and quietly abandon.
- Start with the product's unfair advantage, then ask what AI makes possible because of that advantage.
- StackBlitz succeeded with Bolt.new by changing the development workflow and creating a new category around browser-based app building.

## Summary
Victoria Melnikova explains why many developer tools companies struggle when they add AI. A chatbot layered onto an existing interface can feel underwhelming, while trying to AI-enable everything puts a small company against OpenAI and Anthropic. Waiting for perfect AI also delays learning, and failing to speak with users leads teams to build polished features nobody needs. She uses StackBlitz and Bolt.new as the successful example. StackBlitz had spent years building browser-based development infrastructure, which let it run a full development environment without containers, servers, or local setup. The company used that advantage to make app creation possible through natural-language descriptions, with the result visible in a browser and shareable by URL. Melnikova's framework is to find an unfair advantage, determine how AI can amplify it, and redesign the workflow into a category of its own. The argument is practical: AI should change what the product can do because of its existing strengths.

## Key ideas
### A chatbot added to an existing interface often becomes a feature users quietly abandon
[01:09](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=69s)
Melnikova says the most common mistake is to "sprinkle some AI into our UI." Customers are already tired of chatbot interfaces, and a new chatbot can be hard to justify when early results are underwhelming. Users may try the feature once, lose interest, and never return to it. In her description, the outcome is a quietly deprecated feature. The problem is not simply that the model might be imperfect. The product has given people no strong reason to change how they work. A chatbot attached to an old workflow does not automatically create a useful AI product.

### Small companies should avoid trying to AI-enable everything
[01:57](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=117s)
The second trap is "boiling the ocean," or trying to AI-power every part of the product. Melnikova says a small team cannot realistically compete across the board with companies such as OpenAI and Anthropic, which have much larger teams and investments. She recommends finding the company's unique domain, expertise, or technology instead. AI has a better chance of producing a useful product when it is connected to something the company already understands or controls. The choice is therefore about where a startup has an advantage, rather than how many AI features it can add.

### Waiting for perfect AI gives other companies time to learn first
[02:18](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=138s)
Some companies wait for AI to become reliable or complete before they build with it. Melnikova rejects that approach. She says teams have to build the tooling as they go, using scrappy and creative solutions while the technology is still changing. This does not mean pretending the technology is perfect. It means learning through real product work instead of waiting for a final version of AI to arrive. Her point is that the companies willing to experiment with the limits of current tools can move ahead while more cautious teams remain stuck in planning.

### User conversations prevent expensive products that nobody needs
[02:39](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=159s)
Melnikova calls the biggest killer a failure to talk to users. She has seen companies spend substantial money and resources building beautiful, intricate interfaces that do not solve a user problem. A polished interface cannot compensate for a missing market. Her rule is direct: talk to users first. This advice also changes how a team evaluates an AI pivot. Before investing in a new assistant or workflow, the company should find out what users struggle with and whether the proposed change matters to them. Otherwise, the team may spend its runway refining a product people do not want.

### StackBlitz built on browser-based development instead of attaching AI to its old workflow
[03:06](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=186s)
Melnikova presents StackBlitz as a different kind of AI transition. Its unfair advantage was the ability to run a full development environment in the browser without containers, servers, or local setup. That infrastructure had been built over seven years, even though the company initially had almost no revenue and flat growth. The board gave Eric Simons a deadline to show real progress by the end of 2024. Bolt.new then showed explosive growth soon afterward. Melnikova's point is that the later result depended on capabilities StackBlitz had developed long before the AI product appeared.

### AI became useful when it amplified StackBlitz's existing capability
[03:48](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=228s)
Melnikova asks companies to replace the question "How do I bring AI into my product?" with "What becomes possible when AI meets my unique capability?" For StackBlitz, the answer was a simpler way for an average user to create an application. Instead of working through the complicated process common to coding tools, a person could describe an app with words, see it in the browser, and share its URL publicly. StackBlitz's browser environment made this workflow possible. AI mattered because it worked with that infrastructure, rather than because StackBlitz had access to a model that other companies could not use.

### A successful AI pivot can create a new category by changing the workflow
[04:37](https://www.youtube.com/watch?v=3YRrBFeQ1aw&t=277s)
The third step is to create a category. Melnikova tells companies to stop thinking only in terms of existing workflows and ask how their capability could change the flow altogether. She describes StackBlitz as reinventing how people "vibe code." Its secret was not a unique AI model, since everyone could access Anthropic. The company had a distinct competitive advantage, used AI to amplify it, and then committed to the resulting user flow. Her framework asks whether a company's advantage can support a new category, rather than merely adding another feature to an established product.

## Notable quotes
- "The most common mistake I see is let's sprinkle some AI into our UI." (01:09)
- "The biggest killer, not talking to users." (02:39)
- "What becomes possible when AI meets my unique capability?" (03:48)
- "Everybody has access to Anthropic. It was their unique competitive advantage." (04:55)

## Tools & references mentioned
- Bolt.new
- StackBlitz
- Evil Martians
- Eric Simons
- OpenAI
- Anthropic

## Who should watch
- You are leading a developer tools startup and are deciding whether an AI feature belongs in the existing product.
- Your team is building an AI assistant without a clear user problem or a distinctive technical advantage.
- You have browser infrastructure, domain expertise, data, or distribution and want to find a product direction that uses it.

## Related talks

- [The AI Pivot](https://aietalks.com/talks/the-ai-pivot) (Brittany Walker, CRV & Chris White, Prefect & Bryan Bischof, Hex, 35:16)
- [Build Dynamic Products, and Stop the AI Sideshow](https://aietalks.com/talks/build-dynamic-products-and-stop-the-ai-sideshow) (Eliza Cabrera, Workday & Jeremy Silva, Freeplay, 18:10)
- [Bolt.new: How We Scaled $0-20M ARR in 60 Days with 15 People](https://aietalks.com/talks/bolt-new-how-we-scaled-0-20m-arr-in-60-days-with-15-people) (Eric Simons, Bolt, 17:33)
- [Building AI Products That Actually Work](https://aietalks.com/talks/building-ai-products-that-actually-work) (Ben Hylak, Raindrop & Sid Bendre, Oleve, 18:42)
- [Stop Ordering AI Takeout: A Cookbook for Winning When You Build In House](https://aietalks.com/talks/stop-ordering-ai-takeout-a-cookbook-for-winning-when-you-build-in-house) (Jan Siml, 10:45)