AI summary by TwinMind

Serverless kinda sucked without this...

Theo - t3․gg

12m 23s

Lambda Execution Limitations

  • Vercel solved major serverless pain point: Lambda dies after sending response
  • waitUntil() allows async code execution after response (like Go's defer)
  • Lambda has 3 invocation types: synchronous, asynchronous, polling
    • Synchronous: caller waits for completion, function returns value
    • Most web traffic (API Gateway, CloudFront) uses synchronous invocation
    • Synchronous has limitation: dies once response returned
  • Asynchronous invocation requires building event queue system
    • Must manually attach to web layer
    • Complex to implement successfully
    • Even experienced developers (including Dax) struggle with implementation

Technical Demo Implementation

  • Built simple web form with name input + submit button
  • Server action function with "use server" directive
  • Database: simple db.txt file for appending names
  • Two versions: regular append vs append with 5-second delay
  • Without await: immediate response but potential data loss on Lambda
  • With await: user waits full 5 seconds for response
  • Problem: removing await on Lambda = high chance append operation fails/never executes

waitUntil API Usage

  • Import: import { waitUntil } from '@vercel/functions'
  • Installation: bun add @vercel/functions
  • Usage: wrap async operations with waitUntil()
  • Ensures code execution completion even after response sent
  • Maintains Lambda instance alive for background operations

Vercel's Infrastructure Architecture

  • Vercel doesn't use AWS Gateway directly
  • Custom orchestration layer before Lambda execution
  • User → Vercel Function Edge Layer → Lambda
  • Traditional: Gateway kills Lambda immediately after response
  • Vercel's approach: keeps Lambda alive longer through custom protocol
  • Uses non-traditional web requests (possibly WebSocket) between edge layer and Lambda
  • Enables flexible Lambda management without standard AWS limitations

Real-World Use Cases

  • Logging operations
  • Database writes that don't block user response
  • Background image generation
  • Any async work not required for immediate response
  • Goal: fastest possible user response + background processing

Advanced Streaming Example (Guillermo's Demo)

  • Reimplemented React Server Components without React/Next.js
  • Used confetti library via CDN embed
  • Transform stream with readable/writable pattern
  • Immediate "Hello" response, then streamed "World" + confetti after 1.5 seconds
  • Clever hack: 1,024 nbsp characters to exceed minimum chunk size for gzip/brotli compression
  • Forces immediate chunk flush to user
  • Pure HTML streaming without framework dependencies

Business Value Proposition

  • Vercel pricing justified by avoiding complex infrastructure building
  • Alternative: build custom queuing/management system (years of engineering effort)
  • Fundamental shift: prioritize fast user response + background processing
  • Can still use traditional await when blocking behavior needed

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome