AI summary by TwinMind
Making Good Tech Decisions - MY FIRST CONFERENCE TALK
Theo - t3․gg
24m 39s
Speaker Introduction & Background
- Theo, CEO of startup ping.gg (content creator tools)
- Former Twitch employee (~5 years) + Amazon experience
- Y Combinator Winter 22 batch participant
- YouTuber/streamer doing technical content, live coding interviews (few thousand viewers weekly)
- Creator of T3 Stack (modularity + type safety focused)
Talk Scope & Limitations
- Will cover:
- Technical decision making process
- Meta framework comparisons
- Full stack solutions & API layers comparison
- Modular building impact on cost/decisions
- Will NOT cover:
- Performance metrics (response times, user speed)
- Technology history/evolution
- Focus only on current decision-making with present knowledge
Importance of Technology Selection
- Tech choices affect: hiring process, developer satisfaction, delivery times, product capabilities
- Tech = major investment like employees
- Must consider long-term impact + bet on maintainers
- Need ability to fork/maintain if necessary
- Challenge: more options than ever (Angular example - once viable, now not recommended for new projects)
Decision Measurement Framework
Cost Breakdown (3 components):
- Ramp up time: Initial learning/setup cost
- Maintenance overtime: Ongoing operational cost
- Ramp off time: Migration/exit cost
Value Measurement:
- Time savings from library/framework
- User experience wins
- Developer happiness impact
- Hiring process effects
Risk Assessment:
- Hardest to measure but critical
- Plan for potential future impacts
- Consider team's unique risk tolerance
- Acts as "delta" for cost range variation
- Can affect adoption/drop timing horizontally
tRPC vs GraphQL Comparison
tRPC Characteristics:
- Type-safe API between backend/frontend
- Pure TypeScript functions (no custom syntax/compiler)
- Requires full TypeScript stack (backend + frontend)
- Fastest way to make sweeping changes with full-stack type safety
- Risk: Primarily maintained by one developer
- Migration advantage: Code you'd write anyway, easy to port to GraphQL/Express
GraphQL Characteristics:
- Schema language for API description/calling
- Language agnostic (requires client/server selection)
- Wide ecosystem: Apollo, Relay (JS), GraphQL Go, etc.
- Main benefit: Centralizes APIs on single endpoint with unified schema
- Functions as communication method between teams
- Created by Meta/Facebook with strong community
- Supports multiple platforms: Swift, Kotlin, Python, Golang, etc.
Cost Visualization Framework
Chart Axes:
- Y-axis (Cost): Time, financial investment, risk impact on business/people
- X-axis (Employee Count): Represents business complexity, success stage, future planning needs
tRPC Cost Analysis:
- Ramp up: High initial cost (boilerplate understanding required)
- Maintenance: Flat cost across team growth (just write/call functions)
- Scaling limitation: TypeScript compiler slows/crashes with large codebases
- Off-boarding: Relatively cheap (copy-paste existing code + add validation/schema)
GraphQL Cost Analysis:
- Ramp up: Significantly higher (choosing ~5 technologies, not one)
- Maintenance: Ongoing cost with schema changes (no guarantee clients respect changes)
- Communication benefit: Standardizes backend/frontend team interaction
- Off-boarding: Extremely expensive (line goes off chart - "unimaginable" to migrate)
Tooling Solutions for Cost Reduction
Create T3 App:
- Reduces tRPC onboarding cost dramatically
- Pre-configured boilerplate with Next Auth, Prisma, etc.
- Slightly higher off-boarding cost (deferred learning debt)
- Maintains same maintenance cost as base tRPC
Redwood.js:
- Full-stack TypeScript solution with GraphQL layer
- Scales better than tRPC/T3 for larger teams
- Supports non-TypeScript clients (Swift, Kotlin mobile apps)
- Built expecting majority consumption within Redwood project
- Very expensive to migrate off once deeply adopted
- Can reach ~400 employee scale vs T3's earlier limits
Risk Considerations
- Risk = horizontal extension/shrinkage of technology lifespan
- tRPC: May need earlier migration with many endpoints
- GraphQL/Redwood: May extend further with non-engineer hires, contractor needs
- Every business has unique risk tolerance
Additional Decision Factors
Client Needs:
- Often forgotten in technical discussions
- Consider current user requirements
- Plan for future user needs
- Impact on user engagement/conversion
Current Team Considerations:
- Team's existing technology familiarity
- Impact on developer day-to-day experience
- Developer satisfaction = retention + quality output
- Sometimes worth risk if team confident in solution
Future Planning:
- Different projects have different trajectories
- Some projects: single developer forever
- Others: may scale to 10,000+ engineers
- Code replacement expectations should influence technology choices
- Modularity decisions based on expected future state
Create T3 App Features
- Modular tool selection during initialization
- Options: tRPC, Prisma, Tailwind, NextAuth
- Add/remove pieces anytime
- Swap components for alternatives
- Goal: Easy onboarding without sacrificing off-boarding benefits
- Maintains modularity advantages while reducing upfront costs
