Get The Overnight Agent Playbook

The pattern: keep a premium frontier model (Claude Opus 4.7, GPT-5, Gemini 2.5 Pro) as the orchestrator that plans, decides, and reviews; drop an open-weight coder model (DeepSeek V3.1, Qwen3-Coder-480B, GLM-4.6) in as the cheap executor that runs the tool calls, edits files, and grinds through the mechanical work. Same Codex harness, same tools, same MCP servers, same file edits — but the per-token cost collapses by roughly two orders of magnitude. This is the field playbook: the exact orchestrator + executor pattern, the four provider-routing options (DeepInfra, Fireworks, Together, self-host vLLM), the drop-in config.toml snippet with model_provider + base_url + env_key, six guardrails for overnight runs (hard cost cap, turn limit, file-write allowlist, network egress allowlist, auto-snapshot before destructive ops, dead-man timer), the cost-tracking dashboard jsonl schema (turn_id, role, tokens_in/out, cached_tokens, tool_calls, wallclock, cost_usd_estimate) with the daily roll-up workflow, the six-line routing decision matrix (when to use orchestrator vs executor), six failure modes (context bloat, tool-call loops, runaway rewrites, retry storms, rate-limits, broken-code-passes-review) with fixes for each, and the real cost table comparing premium-only ($1,200–$1,800) vs hybrid ($3.12) vs all-executor ($0.50) for a 21-hour 165M-token run.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.