Key Concepts
Vibe coding, Gen AI, Bolt.new, Lovable.dev, Cursor, Windsurf, Cloud Code, CLA code, Pair programming, Product-led service, Multimodal AI, Stable Diffusion, Notebook LM, mCP servers, Firecrawl, Superwhisper, Context, AI Engineer Summit, Latent Space.
Vibe Coding and its Applications
Patrick Ellis, CTO at Snapbar, discusses how they leverage GenAI throughout their company, particularly focusing on "vibe coding." Vibe coding, in their workflow, involves using tools like Bolt.new and Lovable.dev for rapid prototyping and initial code base development. These are then often taken over by tools like Cursor, Windsurf, or Cloud Code for further refinement and integration.
Empowering Non-Technical Teams: A key benefit has been empowering non-technical staff (product, sales, CEO) to prototype and explore ideas. This allows them to define requirements more concretely and communicate them effectively to the engineering team.
Replacing Figma: For new apps and features, Bolt.new and Lovable.dev have, in many cases, replaced Figma for initial design and requirements gathering.
Development Speed: A slideshow project went from Bolt.new to Cursor and into production within a very short timeframe, demonstrating the speed gains. More complex projects, like a new dashboard, benefit from the exploratory nature of vibe coding to define requirements.
Augmenting Traditional Development with CLA Code
Snapbar augments traditional development workflows using CLA code, treating the models as a pair programmer or a "rubber ducky." This involves a disciplined, iterative approach where the model provides feedback and insights, leading to production-grade, scalable code bases.
Pair Programming Approach: The key is to intentionally think through the code, review it, and iterate, maintaining a traditional software engineering workflow while leveraging the productivity boost from AI.
Automating Solutions Engineering and Enterprise Customizations
A significant development has been the emergence of a "product-led service" business model, where Snapbar offers more customized solutions to enterprise clients.
Product-Led Service: By pushing into the service side, Snapbar captures more value and provides more value to customers.
Automation: Code generation tools like Cursor and Windsurf are used to aid or fully automate frontend customizations for clients. This allows for a more scalable and efficient approach to solutions engineering.
Strategy, Marketing, and Product Research
GenAI is also used for deep market research, company strategy, and product strategy.
Deep Research: Leveraging models like GPT-4 (via Deep Research) provides powerful capabilities for thinking through coding challenges and strategic decisions.
Meeting Summarization: Transcriptions from meetings are fed into Notebook LM (especially the podcast feature) for summarization. These summaries are then used in Claude to create PRDs or other documentation.
Tools and Workflows
Patrick provides a list of specific workflows and tools they use:
- AI Models:
- Claude 3 Opus: Good but ambitious for code generation and architecture.
- GPT-4 and GPT-3.5-turbo: High for thinking through architectural approaches.
- Deep Research: Quickly learning about a new domain.
- Code Generation Tools:
- Cursor, Windsurf, Claude Code: Development.
- Bolt.new, Lovable.dev: New prototypes.
- mCP Servers:
- mCP directory (official Anthropic): Resources.
- Browser tools: Context to Claude (network logs, console logs, screenshots).
- Firecrawl: Turning any website into markdown.
- Superwhisper: Transcribing speech in context.
Context Management: Patrick emphasizes the importance of context. He creates a "context" folder in his repo with a "docs" file containing relevant documentation (API docs, markdown files from Firecrawl, Google Docs exports).
Strategy and Helpful Patterns
Context is Everything: The most valuable insight is that context really matters. Think about what the model has access to (context, indexed code, mCPs) and provide concrete examples, precise documentation, and references to previous threads.
Define Your Stack Up Front: Anchoring the output to your codebase is crucial, especially with Bolt.new and Lovable.dev.
Choose Libraries with Ample Training Data: React, Next.js, and Node.js have many open-source examples.
UI Design and SEO: GenAI is particularly good at UI design and SEO suggestions due to the training data and prompting capabilities.
Educational Resources
Patrick recommends several resources for staying up-to-date in the rapidly evolving AI space:
- Latent Space: Podcast and website (latent.space).
- AI Engineer World's Fair.
- Fast.ai (Kathy): ML resources.
- AI Tinkerers (Joe): Conference.
- AI Explained: YouTube channel covering new white papers and advancements.
- AI Engineer Conference: YouTube channel.
Conclusion
Snapbar is aggressively leveraging GenAI across its operations, from empowering non-technical teams to automating solutions engineering. The key takeaways are the importance of context, the value of treating AI models as pair programmers, and the potential for new business models enabled by AI-driven automation. The speaker emphasizes the need to stay informed and adapt quickly in the rapidly evolving AI landscape.
AI summaries can miss context or contain errors. Check important details against the original video.