The $5M pitch Command Code AI announced that they recently raised $5M to build a proprietary AI agent that learns your coding style; taste.
Meanwhile, session-commit does the same thing with a markdown file and some vibes.
How I discovered Command Code AI When I posted about session-commit on X, another developer replied that he uses @CommandCodeAI for taste learning. I was intrigued.
session-commit vs. Command Code AI Dimension session-commit Command Code AI What it learns Durable project knowledge: architecture decisions, conventions, debugging playbooks Personal coding style: naming, patterns, formatting preferences Where it stores AGENTS.md, a human-readable, version-controlled file in your repo Proprietary model weights / taste profile (opaque) How it learns Explicit: you run the skill, review proposals, approve changes Implicit: watches every accept/reject/edit continuously Granularity Project-level knowledge Developer-level preferences Portability Fully portable. It’s just a markdown file that any agent can read Locked to their CLI (npx taste push/pull) Tool lock-in None. Works with Claude Code, Codex, Gemini CLI, OpenCode Requires command-code CLI Transparency 100%. You see exactly what’s stored and can edit it Black box. You don’t see the taste model internals Team sharing Share via git (it’s in the repo) Share via npx taste push/pull Cost Free / open source $10 credits, then paid The philosophical difference Command Code AI says: “I’ll silently watch how you code and adapt to your style.” It’s implicit, continuous, and developer-scoped. session-commit says: “Let’s explicitly capture what we learned and write it down for future sessions.” It’s explicit, intentional, and project-scoped. In reality, they’re complementary more than competitive. Command Code optimizes how code is written (style/taste), session-commit optimizes what the agent knows about your project (architecture, conventions, gotchas).
...
