Cursor Origin, Command Code, and the New Coding Agent Stack
Cursor is expanding from editor to forge while Command Code learns from developer choices. Here is what changed, including the verified Grok model and current pricing.
Two launches show coding agents moving in opposite but complementary directions.
Cursor Origin expands outward into repository infrastructure. Command Code moves inward by learning a developer's coding taste from accepted, rejected, and edited output.
Together they suggest that the next competition is not only about which model writes the best patch. It is about who understands the developer and who controls the workflow around the patch.
Cursor Origin is a forge for agent work
Origin is an early beta for paid Cursor plans. Cursor describes it as a git forge for the agentic era.
That framing makes sense. When several agents can work in parallel, the repository layer needs better answers for provenance, coordination, review, and merge safety. Conventional git hosting was designed around human-authored branches and human-paced review.
The test for Origin is whether it creates a better verification workflow, not whether it produces more activity.
Command Code learns from your decisions
Command Code is a terminal coding agent built around a preference model called taste-1. The product learns from what a developer accepts, rejects, and edits. It can also turn repeated behavior into generated skills.
This is a more defensible personalization strategy than a long static preference file. A developer's standards are visible in decisions: which abstraction was kept, which test was added, and which generated pattern was removed.
The risk is false confidence. Preference learning needs transparency and a clean way to correct a bad inference. Teams should be able to see what the agent believes and why.
About Grok 4.6 and Cursor pricing
The current Cursor pages do not list a model called Grok 4.6. Cursor's verified lower-cost plan references Grok 4.5, while Opus 4.6 is a separate Anthropic model. Those names are easy to combine accidentally.
Cursor pricing lists Pro at $20 per month and Teams at $40 per user per month. Cursor Start, announced for India, is ₹649 per month and includes Grok 4.5, Composer, cloud agents, mobile access, plugins, MCP, and skills.
Pricing and model availability change quickly. Check the live pricing page before making a purchasing decision.
What to watch
The coding agent stack is taking shape around four layers:
- A model that can reason about the task.
- A preference layer that learns how the team wants work done.
- An execution environment that can run and verify changes.
- A repository layer that can coordinate and review agent work.
The editor remains important. The larger opportunity is the system that surrounds it.
Sources
Bhaulik Patel
Forward deployed AI engineer and creator of Deployed Engineer.