How To Be A 100x Engineer Using Ai Notes
todos
- create a PR creating agents workflow from github issues
- create claude.md for general code quality and rules
1. architecture rules – how we structure code here
2. known mistakes – what the ai got wrong and how to fix it
3. constraints – security, performance, and cost limits
how will we verify this works?
how will we monitor this in production?
what metrics will tell us if this is failing?ai code review tools: codium pr-agent/qodo, github copilot workspace, what-the-diff, grit.
let background agents to run async tasks - can integrate this with github issues and let them solve and I can review those PRs later whenever I am free.
prompts should have who is the agent, what is needed, how to go about the conversation, constraints, input with existing architecture and output as planning document
for pr agents (mention in github issues)
rules
explicit constraints (don't touch x, must preserve y)
tasks
linter warnings, test cases, migration, refactoring, strategies
prompt for planning how to go about a feature
who: act as a product-minded engineer who understands both technical and business constraints
what: we need to add real-time notifications to our dashboard. current polling is killing our database and users complain about 30-second delays.
how: first, analyze our current architecture and list integration points. then suggest 2-3 approaches (websockets, sse, polling optimization) with explicit tradeoffs on cost, complexity, and reliability.
input:
<current_architecture>
[paste relevant code/diagrams]
</current_architecture>
<constraints>
- max 500ms p95 latency
- current db can't handle more load
- budget: $200/month for new infrastructure
</constraints>
output: markdown table with columns: approach, implementation complexity (1-10), monthly cost estimate, latency impact, reliability concerns
focus on infra budgets, SLAs, latency requirements, rate limits, memory and compute constraints. feature development prompts
use well tested apis - write comprehensive code - force the AI to run tests
testing prompts
"list all edge cases that could break this function, then write property-based tests for them".
sandbox branches with protection
background agents never work directly on main or production branches.
they work in:
- dedicated feature branches with clear naming
- ephemeral preview environments with locked-down permissions
- isolated worktrees for parallel work
branch protection rules and ci gates must pass before any merge.