AI summary by TwinMind
Build first, plan second.
Theo - t3․gg
29m 15s
Two Types of Engineers at Big Tech
- Engineers who write detailed software design docs first vs those who build working proofs of concept first
- Speaker strongly advocates for prototype-first approach over design-doc-first approach
- Compares design-doc-first to "trying to paint a house you haven't seen yet"
Sponsor: Prisma Postgres
- Prisma was core to T3 stack experience
- Prisma Postgres solves serverless scaling issues, connection limits
- Features edge connectivity, caching layer, live updates/streaming
- One-click deployment, scales automatically
Twitch Dashboard Layout System Case Study
- Complex customizable dashboard using react-mosaic library
- Layout defined as binary tree with nodes, split directions (horizontal/vertical), percentages
- Data stored in local storage, worked well for single devices
- User request (5 votes) for syncing layouts across multiple devices
The Sync Layout Project Disaster
- PM spent months on 7-20 page design doc for layout synchronization
- Proposed 5 engineers, 6 months development time
- Fundamental flaw: different devices have different aspect ratios
- Desktop 1080p vs iPad different ratios
- Percentage-based splits break on different screen sizes
- Would have made user experience worse, not better
Technical Architecture Problems
- Twitch uses GraphQL - can't query binary trees of unknown depth
- Team's solution: flatten binary tree into array with parent/child relationships
- Reconstruct on client, deconstruct for server storage
- Overcomplicated solution to non-existent problem
The Simple Solution Discovery
- Speaker contacted GraphQL team directly
- Tony (GraphQL team lead): "Why wouldn't you just store that as JSON?"
- 15-minute call eliminated 2 months of design work
- GraphQL team would have approved JSON storage immediately
- Design doc team never consulted actual GraphQL experts
User Research Failure
- PM never talked to actual creators/users
- Users wanted same layout on vertical monitor, laptop, iPad (different feature entirely)
- Speaker forced to collect data proving different devices = different aspect ratios
- <5% of users with multiple devices had same aspect ratio
- Proved feature would be regression, not improvement
Parallel Success: Twitch Studio Layout
- Speaker assigned different project while sync disaster unfolded
- Built responsive customizable layout for Twitch Studio in under 1 week
- Team calls it "The Theo Method" - build complex thing first to understand problems
- No design doc needed, just working prototype
Quick Actions Bug Example
- Buttons have 4 states: loading, ready, action, done
- Clicking button changes state to "action", useEffect triggers actual action
- Layout saves include button states
- Resizing during ad causes button to save in "action" state
- Dashboard becomes "bomb" - triggers ad on every refresh
- Caused multiple major incidents including ads during League of Legends tournaments
- Engineer who designed it insisted design doc proved this impossible
- Event logs showed ad triggered 2ms after page load (impossible for user click)
Problems with Design Doc Culture
- Discourages talking to people who know solutions
- Persists bad decisions once written down
- Creates imaginary complexity
- Promotes people who write docs over those who build working software
- Results in 6+ months work on regressions instead of features
The Promotion Problem
- PM who wrote bad sync spec became Twitch exec
- Engineer who designed broken quick actions got promoted 2 levels
- Engineer who built working Studio prototype still fighting for promotion after 5 years
- Design docs used for promotions in toxic environments, not actual engineering
Recommended Approach
- Start with prototype, not design doc
- Try building it to understand actual complexity
- Talk to users and domain experts
- Write docs AFTER understanding the problem
- If prototype works, analyze complexity
- If low complexity, ship the prototype
- If high complexity, rebuild with new knowledge
- Design docs only valuable once you understand the complexity
Trust and Environment Issues
- Some engineers use design docs from fear of maintaining forced-to-production POCs
- Solution: build prototype privately, don't let management know until ready
- "Ask forgiveness not permission" approach
- If necessary, sabotage prototype to prevent premature shipping
