There is a gap forming inside most organisations right now. Developers are running AI coding agents on their laptops, experimenting enthusiastically. But almost none of that is happening at a company-wide level with proper oversight. Research from McKinsey shows that while nearly two-thirds of enterprises have experimented with AI agents, fewer than 10% have scaled them to deliver measurable value. The blockers are governance, compliance, and cost visibility, not the quality of the AI itself.
Warp's Oz platform is a direct answer to that gap. Oz acts as a management layer that sits above individual AI agents and gives teams a single place to launch, monitor, and control them. This week's update makes Oz the first platform of this kind to handle multiple AI agent systems at once: Anthropic's Claude Code, OpenAI's Codex, and Warp's own agent can all run inside Oz, each tracked separately, each with its own access permissions and cost limits.
Why does it matter which agent system you use? Because Claude Code and Codex are genuinely different tools with different strengths. Claude Code runs locally and keeps code on your own machine, which is meaningful if your organisation has strict data rules. Codex runs in a cloud sandbox and is better suited for parallel, asynchronous tasks. Most sophisticated engineering teams are already thinking about using both for different purposes. Oz makes that practical by giving you one place to manage and compare them.
The Agent Memory feature deserves attention even though it is still in early preview. The idea is that agents accumulate knowledge about how your specific organisation operates: your coding standards, your deployment topology, your data structures. That knowledge carries across sessions and across different agent systems. Today, every time you bring an AI agent into a task, someone has to re-explain the context. Agent Memory removes that repetition and, more importantly, means your organisation builds a compounding asset over time rather than starting from scratch each session.
For business operators outside the technology department, the governance features are the most immediately relevant part of this launch. Oz now supports per-team cost caps, granular access controls so that agents working on customer data have different permissions than agents working on internal code, and full audit logs for every action. These are the features that legal, compliance, and finance teams need before they can sign off on agents running autonomously inside production systems.
Warp is also expanding self-hosting options, meaning companies can run agents entirely within their own cloud infrastructure rather than routing data through Warp's servers. This matters significantly for companies in regulated industries, including insurance, financial services, and healthcare, where data residency requirements are non-negotiable.
The competitive context is worth noting. GitHub has also moved in this direction, adding Claude and Codex as options within its own agent platform for enterprise subscribers. The pattern is consistent: multi-agent management, shared governance, and the ability to switch between AI providers without rebuilding your processes each time. Companies that lock themselves into a single agent provider today may find themselves at a disadvantage as these tools continue to change rapidly.
Gartner predicts that over 40% of AI agent projects will be cancelled by 2027 due to cost overruns and inadequate risk controls. The companies that avoid that outcome will be the ones that treated governance as part of the deployment plan from the start, not something to retrofit later. A platform like Oz is one practical way to do that.