Anchor your career on a moat, not a role.
Core Identity
Commit to “Compound Systems Builder” as your lane:
- Systems
- Correctness
- Orchestration
- Observability
- Automation
- Infra-as-capital
Don’t hedge into sales as a primary move. If you ever add non-technical skills, add them later as multipliers, not as identity.
The Compound Engineering Vision
Build the core primitives of a compound engineering system that runs on top of Claude. Not more prompts, but a proper supervisor layer that treats LLMs like policy functions inside a real environment.
The Idea
- Formalise task-planning invariants (“every change must pass tests + observability gates”)
- Build a small gym for Claude (state snapshot, patch apply, test runner)
- Codify two master workflows:
- Forward-planning for new features
- Investigative planning for discovering regressions, perf gaps, and invariant drift
Multi-Agent Patterns
Layer on top:
- Many-perspectives
- Spec drift detection
- Invariant extraction
- Adversarial test generation
All orchestrated by a thin deterministic controller.
The Goal
Create a repeatable meta-factory where plans, execution, verification, and refactoring can compound over time.
This isn’t about making Claude write code; it’s about engineering a system where quality, velocity, correctness, and architecture all reinforce each other automatically.
Boundary Contracts
Vercel (UI Layer)
- Next.js frontend
- Thin BFF for SSR/ISR
- No business logic
- No direct DB writes
GCP/Cloud Run + Fastify (Domain Gate)
- All meaningful writes
- Invariant enforcement
- Auth verification
- Agent-safe operations
Supabase (Service Layer)
- Database
- Auth service
- Storage
- NOT the “backend brain”
Temporal (Orchestration)
- Multi-step workflows
- Retryable operations
- Long-running processes
- State machines
Terraform (IaC)
- Reproducible environments
- IAM configuration
- Cloud Run setup
- Secrets wiring
Feature Contract Checklist
Every feature must satisfy:
Behaviour (Tests)
- Unit tests for pure logic
- Integration tests for boundaries
- E2E tests for golden flows
Physics (OTEL)
- Trace propagation configured
- Key metrics instrumented
- Structured logging in place
Deployment
- Terraform updated
- CI/CD passing
- Rollback plan documented
The Deep Understanding
You are converging on the idea that software should be treated as a measurable dynamical system, not as a collection of features.
You are thinking like a quant-dev or theoretical systems engineer, treating code as an executable model of reality. You reason in terms of:
- State
- Signals
- Noise
- Feedback loops
- Invariants
Features are surface manifestations; the real work happens in the dynamics underneath.
Related
- Systems Thinking & Observability
- Liquidation Cadence
- Control Theory – state, signals, feedback loops, dynamical systems
- Optimisation – infrastructure trade-offs as constraint satisfaction problems
- Histogram Metrics for Batch Workloads – Pattern for observing aggregate system behavior