Vision
For decades, software development moved at a human pace: write some code, commit it, ask a teammate to review it, then ship. Source control recorded the work, but it did not actively coordinate it.
That model is no longer enough. Engineers now direct agents that can explore, implement, test, and operate software in parallel. The bottleneck is no longer only writing code; it is keeping every actor working from the same trustworthy state.
VEX makes source control the shared control plane. Humans, agents, CI, and runtime systems work through one live graph of code, policy, workspaces, checks, and outcomes — so every change can be understood, verified, and carried forward.
We believe the next software platform must be agent-native and accountable by default: fast enough for machines, clear enough for people, and grounded in the source of truth both depend on.
This is what VEX is building.
Source control where people and agents can build, verify, deploy, and learn together.
One system of record for the work itself and the signals that tell us what should happen next.
The first place we are applying that idea is review. Agents can now produce large, dependent changes faster than a reviewer can reconstruct their context from a diff. Making people read more code, more quickly, is not a solution.
VEX keeps related work as a coherent stack instead of collapsing it into one enormous pull request. A reviewer can follow the changes in order, understand the goal behind each step, and see how the work fits together before deciding what should land.
Review the outcome, not just the diff.
Every agent-made change should explain what it set out to do, why it took that path, and what proves it worked.
VEX will surface the intent, the decisions, and the evidence alongside the code: the request the agent was working from, the stack it produced, the checks it ran, and the result it was trying to achieve. Reviewers can start with the outcome and drill into code only where judgment is needed.
That makes review faster without making it blind. People still decide what is safe to land. They simply do not have to piece together the whole story from commits, chats, terminals, and scattered CI output before they can make that decision.