Human architect + Claude Code + Codex

Every line generated by AI.
Guided by one human.

CronControl was built in 2 days using AI agents. The product specification and architecture were designed through coordinated work between Claude Code and OpenAI Codex, with one human directing the process. All implementation code, SQL, documentation, and this website were AI-generated.

23,354
Lines of code
398
Tasks completed
8
Epics at 100%
6
Critical bugs caught

Build timeline

1
Session 1

Foundation + Scheduling

  • • 8 epics audited & all task files updated
  • • SSH strict host key verification + kill
  • • Worker execution + enrollment persistence
  • • SSM executor (aws-sdk-go-v2)
  • • K8s executor (client-go)
  • • 3 logging backends (DB, file, OpenSearch)
2
Session 2

Identity + Frontend

  • • Google OAuth full flow
  • • Email verification + password reset
  • • Workspace switcher + onboarding banner
  • • Batch enqueue + webhook test delivery
  • • Command palette (Cmd+K)
  • • 5 SDKs (Python, Node, Go, PHP, Laravel)
3
Session 3

Polish + Admin

  • • 6 critical bugs found by code review
  • • Concurrent session fix (SSH/SSM/K8s)
  • • DNS rebinding protection
  • • OAuth credential exposure fix
  • • Platform super admin system
  • • Worker connectivity documentation

Tech stack

Backend

  • Go 1.24
  • chi router
  • pgx/v5 + SQLC
  • aws-sdk-go-v2
  • client-go (K8s)
  • golang.org/x/crypto

Frontend

  • React 19
  • TypeScript
  • Vite
  • Tailwind CSS 4
  • TanStack Query v5
  • shadcn/ui + Lucide

Infrastructure

  • PostgreSQL 16
  • Docker Compose
  • GoReleaser (4 binaries)
  • Prometheus metrics
  • OpenSearch (optional)
  • go:embed SPA

Agentic

  • MCP server (15+ tools)
  • CLI (cronctl)
  • OpenAPI 3.0.3 spec
  • PHP heartbeat SDK
  • JSON-RPC 2.0
  • 4 documentation guides

Specification: two AI agents

The product specification, business model, competitive analysis, and architecture decisions were designed through coordinated work between Claude Code and OpenAI Codex. Each agent contributed different perspectives on the product design, with the human orchestrating the process and making final decisions.

C
Claude Code
  • Product specification & glossary
  • 8 epic documents & task lists
  • All Go, TypeScript, SQL implementation
  • Code review & bug detection
  • SDK, CLI, MCP server, documentation
  • This website (Astro + Tailwind)
X
OpenAI Codex
  • Business overview & positioning
  • Competitive analysis (36 products)
  • SaaS market research (21 services)
  • Architecture validation
  • Roadmap structure & sequencing
  • Product language & terminology

How it works

The human sets direction: coordinates between agents, reviews output, makes decisions. Codex helps shape the product spec. Claude Code implements everything. A code review agent catches bugs before they ship. The result is a production-grade system built at a pace no traditional team can match.

codex: *analyzes 36 competitors, identifies two-tier scheduling as key differentiator*
human: "implement the 6 remaining worker polish items"
claude: *reads 8 files, implements waiting_for_worker, ProcessResult(), reassignment, GoReleaser, worker guide*
human: "next"
claude: *implements batch enqueue, webhook test delivery, workspace health, secret masking, 4 more endpoints*