Apr 25, 2025 · 16m · y-combinator
How To Get The Most Out Of Vibe Coding | Startup School · Y Combinator
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Y Combinator Startup School guide, General Partner Tom Blomfield and YC founders share actionable best practices for 'vibe coding'—using AI coding assistants to build software efficiently. The video details essential techniques including pre-coding planning, git version control, test-driven integration, context management, and selecting optimal AI tools and tech stacks.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The partners as informed peer 0.0 Guest teaching 0.0 Guest disagreement 0.0 The partners pushing back 0.0
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 11:10 Dismissing MCP server complexity
Blomfield mildly pushes against current developer trends by calling MCP servers for documentation access overkill compared to downloading local docs.
Hardest push from the partners ▶ 0:09 Solo presentation format baselineNo active host pushback occurs in this instructional video format.
Biggest teaching moment ▶ 13:20 Explaining training data advantages in mature stacksBlomfield educates the audience on why older, highly conventional frameworks like Ruby on Rails generate far better AI code than newer languages like Rust or Elixir.
The partners hold their own ▶ 0:09 Solo presentation format baselineNo host counter-expertise is demonstrated because the entire episode is a solo presentation.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| Opening Title Graphic: How To Vibe Code | 0 | 0 | 0 | 0 | This is an introductory montage monologue featuring YC partner Tom Blomfield introducing vibe coding followed by recorded clips of YC founders sharing their tips. Because there is no active host-guest dialogue or interview dynamic, all interactive scores are at zero. | |
| Selecting Tools and Planning Before Coding | 0 | 0 | 0 | 0 | Tom Blomfield delivers a direct educational presentation detailing tool selection (Replit, Lovable, Windsurf, Cursor) and the necessity of writing a markdown planning document before generating code. Since this is a solo lecture format, all combativeness and host scores remain zero. | |
| Version Control and Managing Clean Git States | 0 | 0 | 0 | 0 | Tom Blomfield explains why strict version control via Git and high-level integration tests are essential to prevent LLMs from introducing regressions and code cruft. The monologue contains no host intervention or confrontational dynamic. | |
| Systematic Bug Fixing and Model Switching | 0 | 0 | 0 | 0 | The speaker advises on feeding raw error logs into LLMs, resetting failed bugfix attempts, switching models when stuck, and establishing explicit rule files. As a structured instructional monologue, host and combativeness metrics remain zero. | |
| Managing Documentation and Learning from LLMs | 0 | 0 | 0 | 0 | Tom Blomfield covers local API documentation handling versus MCP servers, isolated reference implementations, and modular architecture. With no host present to challenge or receive schooling in real-time, scores remain zero. | |
| Choosing the Right Tech Stack for AI Coding | 0 | 0 | 0 | 0 | Blomfield wraps up by highlighting why Ruby on Rails excels due to mature conventions and training data, the utility of multimodal input and voice dictation via Aqua, and ongoing model experimentation. The monologue maintains zero across interactive scores. |
Statements from this episode (15)
Blomfield: Product managers and designers are bypassing Figma for direct coding
“Many product managers and designers are actually going straight to implementation of a new idea in code, rather than designing mockups in something like Figma, just because it's so quick.”
Blomfield: Lovable struggles when modifying backend logic rather than UI
“Tools like Lovable started to struggle when I wanted to more precisely modify backend logic. Rather than just pure UI changes. I'd change a button over here, and the backend logic would bizarrely change.”
Blomfield: Multi-prompting AI to fix bugs accumulates layers of bad code
“I found I had bad results if I'm, like, prompting the AI multiple times to try and get something working. It tends to accumulate layers and layers and layers of bad code rather than, like, really understanding The root cause.”
Blomfield: Git reset and re-feed final AI solutions onto clean codebases
“You might go and try four, five, six different prompts, and you finally get the solution. I'd actually just take that solution, git reset, and then feed that solution into the AI on a clean code base, so you can implement that clean solution without layers and…”
Blomfield: Claude Sonnet 3.7 configured DNS and Heroku, accelerating progress 10x
“For example, I had Claude Sonnet 3.7 configure my DNS servers, which is always a task I hated, and set up Heroku hosting via a command line tool. It was a DevOps engineer for me and accelerated my progress like 10 x.”
Blomfield: Major AI coding tools will soon auto-ingest error logs
“It's so powerful that pretty soon I actually expect all the major coding tools to be able to ingest these errors
Without humans having to copy paste.
If you think about it, our value being the copy paste machine is, is kind of weird, right?
We're like, we're l…”
Blomfield: Extensive instruction files make AI coding agents far more effective
“Each tool has a slightly different naming convention, but I know founders who've written hundreds of lines of instructions for their AI coding agent, and it makes them way, way, way more effective.”
Blomfield: Using MCP servers for API documentation is overkill
“Using an MCP server to access this documentation, which works for some people, seems like overkill to me.”
Blomfield: Storing API docs locally yields more accurate LLM code generation
“So I'll often just download all of the documentation for a given set of APIs and put them in a subdirectory of my working folder so the LLM can access them locally. And then in my instructions, I'll say, go and read the docs before you implement this thing. An…”
Blomfield: AI Coding Might Shift Software Architecture Toward Modular Services
“I think we might see a shift towards more modular or service-based architecture where the LLM has clear API boundaries that it can work within while maintaining a consistent external interface Rather than these huge monorepos with massive interdependencies.”
Blomfield: Rails excels in AI coding due to consistent online training data
“Rails is a 20 year old framework with a ton of well-established conventions. A lot of Rails code bases look very, very similar. And it's obvious to an experienced Ruby on Rails developer where a specific piece of functionality Should live, or the right Rails w…”
Blomfield: Voice tool Aqua doubles input speed to 140 words per minute
“With Aqua, I can effectively input instructions at a 140 words per minute, which is about double what I can type.”
Blomfield: AI coding assistants easily tolerate imperfect voice transcription errors
“And the AI is so tolerant of minor grammar and punctuation mistakes that it honestly doesn't matter if the transcription's not perfect.”
Blomfield: Gemini Leads Codebase Planning, Claude 3.7 Leads Code Implementation
“For example, at the moment, Gemini seems best for whole code-based indexing and coming up with implementation plan, while Sonnet 3.7, to me at least, seems like the leading contender to actually implement the code changes.”