OWL (Optimized Workforce Learning) Pipeline

The /owl command activates Antigravity's decoupled multi-agent engineering architecture. Instead of single-pass guessing or fixing symptom B with patch A, OWL splits the task into two specialized decoupled pipelines: DeepInvestigator and DeepCoder.

Architecture Overview

(Note: DeepInvestigator and DeepCoder are built-in native subagents supported by the core runtime engine).

Execution Routine

When the user invokes /owl:

Step 1: Investigation & Hypothesis Falsification (DeepInvestigator)

1. For any bug, refactor, architectural choice, or failure investigation, spawn DeepInvestigator first before writing any production code:

   {
     "TypeName": "DeepInvestigator",
     "Role": "Deep Investigator",
     "Workspace": "inherit",
     "Prompt": "Perform exhaustive root-cause investigation. Map out full call chains, audit invariant constraints, test boundary edge cases, and ask: 'What happens if X, Y, Z, F, G occur?' Formulate verifiable hypotheses."
   }

2. Do not proceed to implementation until DeepInvestigator reports findings, identified pitfalls, and hard invariant requirements.

Step 2: Implementation & Refinement (DeepCoder)

1. Pass the verified requirements, invariants, and edge cases from Step 1 into DeepCoder:

   {
     "TypeName": "DeepCoder",
     "Role": "Deep Coder",
     "Workspace": "inherit",
     "Prompt": "Implement solution satisfying all invariants discovered by DeepInvestigator. Enforce compiler gates, verify downstream consumers in the entire call graph, and run test suites."
   }

2. DeepCoder executes iterative improvement cycles until all gates pass without regression.

Step 3: Synthesis & Verification

1. Verify downstream callers across the entire call-tree.
2. Confirm zero collateral breakage before presenting results to the user.