The Problem: Infrastructure Built for Human-Speed Development Is Breaking
GitLab CEO Bill Staples opened the Transcend 2026 keynote with a direct diagnosis: "As engineering teams scale agent activity, the infrastructure, governance, and commercial models built for human-speed delivery are showing strain."
The four capabilities announced at Transcend directly target the four bottlenecks that emerge at agent scale: task speed, context fragmentation, compliance exposure, and cost unpredictability. The event was broadcast live to over 15,000 registered participants worldwide, with Mercedes-Benz, Google Cloud, AWS, and Stanford's SWEPR research group as featured partners.
Innovation 1: Next-Generation Source Code Management (Private Beta)
Today's Git servers require agents to clone entire repositories, generating massive network traffic and token waste on large codebases.
GitLab's Next-Generation SCM lets agents query the repository server-side for exactly what each task requires. Each agent is scoped to the minimum visibility its task needs. Internal testing shows:
- Agent task completion up to 50x faster
- Token consumption up to 2x lower
- Network traffic up to 1,000x lower
Next-Gen SCM is in private beta. GitLab Ultimate customers can register interest at gitlab.com through their account team.
Innovation 2: GitLab Orbit — Context Graph (Public Beta)
Agents working across a software lifecycle see code, work items, pipelines, deployments, and production signals as disconnected fragments. That fragmentation is a primary driver of hallucinations.
GitLab Orbit maps all of these into a unified context graph for the entire software lifecycle — the same source of truth for both agents and engineers. Based on internal testing, agents using Orbit:
- Responded up to 11x faster
- Used up to 4.5x fewer tokens
- Produced up to 45x fewer hallucinations
Orbit also runs as a standalone data product with open APIs, making the same context layer available to third-party agents and external tools.
Innovation 3: Governance for Agents (Private Beta)
When agents commit code, remediate vulnerabilities, and trigger deployments — compliance requires auditability of every action.
Governance for Agents wraps identity, policy, audit, and approval around every agent action. It provides real-time visibility into inputs, reasoning, tool calls, and high-risk or anomalous activity across the organization.
| Control Layer | What It Does |
|---|---|
| Identity | Unique agent ID and permission scope per task |
| Policy | Automated enforcement of cost budgets and permission boundaries |
| Audit | Full log of inputs, reasoning, and tool calls |
| Approval | Human review gate before high-risk agent actions |
It extends GitLab Ultimate's existing security agents (which automate vulnerability triage and remediation) with a full compliance and governance layer.
A single annual commitment that covers platform seats, GitLab Credits, and new eligible capabilities. Organizations adjust their monthly reservations across all three without contract amendments — designed specifically to handle the unpredictable cost profile of agent-driven development.
Stanford Research: Measuring AI's Impact on Engineering Productivity
Stanford's Software Engineering Productivity Research (SWEPR) group presented findings from a study spanning 600+ organizations on the quantified impact of AI on engineering productivity — one of the first large-scale empirical studies of its kind.
GitLab surpassed $1 billion in annual revenue last quarter, now serves more than 50 million users, hundreds of thousands of organizations, and more than 50% of the Fortune 100.
- Next-Gen SCM (private beta): server-side queries deliver 50x faster agent tasks, 2x fewer tokens, 1,000x less network traffic.
- GitLab Orbit (public beta): unified context graph cuts hallucinations 45x and improves agent response time 11x.
- Governance for Agents (private beta): identity, policy, audit, and approval on every agent action for enterprise compliance.
- GitLab Flex: one annual commitment with monthly adjustments across seats, credits, and features — no contract amendments.
- GitLab crossed $1B annual revenue last quarter; Stanford SWEPR research validates AI-driven productivity gains at scale.