AI Agents Need Managers Too

For the past two years, AI coding tools have largely lived inside the developer workflow. A developer opened an editor, asked for help, reviewed the output, and moved on. The model assisted with implementation, but the work remained tightly coupled to the person using it. The AI acted like a copilot sitting in the passenger seat.

That model is starting to change.

This week, OpenAI announced plans to acquire Ona (the company formerly known as Gitpod) as part of its effort to expand Codex into a more persistent engineering system. The goal is to give AI agents secure cloud environments where they can work on long-running software tasks such as testing, issue resolution, modernization projects, vulnerability remediation, and other engineering workflows that may take hours or days to complete.

On the surface, this looks like another AI infrastructure announcement. In practice, it signals a much larger shift. The AI coding agent is moving out of the IDE and into its own workspace.

Most AI coding products today still operate inside a relatively simple loop. A human asks for work, the model performs a task, and the human reviews the result. The workflow remains centered on the developer.

Persistent agent environments introduce a different model. Instead of waiting for instructions one task at a time, agents can operate continuously inside controlled infrastructure. They can monitor progress, execute workflows, run tests, investigate issues, and return results later.

The relationship starts to look less like a tool and more like delegation. That distinction matters because delegation creates a different set of management challenges than assistance. Once work continues independently, organizations need ways to monitor activity, review decisions, control permissions, and understand what happened while the agent was operating. The question shifts from what the model can do to how the work it performs should be managed.

The significance of the acquisition is not the cloud environment itself. Engineering teams already run workloads in cloud infrastructure every day. The important shift is that OpenAI appears to be building infrastructure specifically designed for autonomous work rather than human interaction.

For years, software development has been organized around people. Tasks are assigned to engineers, work is reviewed through pull requests, and activity is tracked through tickets, approvals, and project management systems.

Persistent coding agents introduce a new participant into that process. They need access to codebases, permissions to perform work, logs to record activity, review mechanisms to validate outputs, and escalation paths when something goes wrong. In many ways, they require the same operational systems organizations already build around human contributors.

Most companies currently think about AI adoption as a tooling decision. Should developers use coding assistants? Which model should be integrated into the workflow? How much productivity improvement can be achieved?

Those questions remain important, but as agents become capable of handling sustained engineering work, organizations will need to decide how that work is governed. Who approves changes? How are actions audited? What level of autonomy is acceptable? When should a human intervene?

These are operating questions as much as technical ones. The companies that move fastest may not be the ones with the most advanced agents. They may be the ones that build the best systems for managing them.

The teams preparing for this shift are not simply deploying more AI. They are designing workflows around it.

Instead of treating agents as standalone productivity tools, they are defining boundaries for what agents can access, what tasks they can perform, and how outputs are reviewed before reaching production. Many organizations are also investing in observability and governance earlier than they did during previous automation cycles. They recognize that once agents begin handling meaningful engineering work independently, visibility becomes just as important as capability.

The goal is not replacing engineering teams. It is allowing engineers to focus their attention where human judgment creates the most value.

This issue of Redeployed is brought to you by Tecla: As AI coding agents evolve from assistants into long-running contributors, the challenge is no longer simply generating code. Organizations increasingly need engineers who can design workflows, manage infrastructure, govern automation, and integrate AI systems into real development environments. The teams moving fastest are building operating models around AI, bringing in talent that can work across software engineering, cloud platforms, and AI-enabled workflows. Tecla helps companies hire senior tech talent in the U.S. and nearshore who already operate in these environments, so teams can scale AI adoption without losing control of how software gets built.

Giving agents their own workspace creates new opportunities, but it also creates new responsibilities.

Security becomes more complex when autonomous systems have persistent access to repositories, testing environments, internal services, and deployment pipelines. Attribution becomes harder when work is produced collaboratively between humans and agents. Vendor dependence increases when critical engineering workflows become tied to a specific platform.

There is also the risk of overdelegation. Organizations may become comfortable assigning increasingly important tasks to agents before governance systems mature enough to support them. Productivity gains can arrive faster than operational discipline, and that gap creates risk.

This shift is changing what engineering organizations need from their people.

The demand is no longer limited to engineers who can write software. Companies increasingly need people who can manage systems where humans and AI agents work together. That includes engineers who understand infrastructure, security, workflow design, observability, and governance, as well as leaders who can decide which work should be delegated, which work should remain human-owned, and how the two should interact.

As AI becomes a participant in software development rather than simply a tool used during development, the value of coordination starts to increase. Managing the work becomes just as important as performing it.

The companies that benefit most from this shift will not focus exclusively on agent capability. They will focus on agent operations.

They will build review systems before incidents occur, define permissions before agents gain broad access, and establish accountability before autonomous workflows become business-critical.

Because once AI agents have their own workspace, they stop looking like software features.

They start looking like members of the team.

And the organizations that learn how to manage that transition will have a significant advantage over those that simply deploy the technology and hope for the best.

Connect With Other Technology Leaders

If you want to connect with other technology leaders having real conversations about AI and how it is changing business, check out GILD Curated Circuit.

More to come…

✔️ Ona is joining OpenAI — Ona

✔️ OpenAI Buys Ona To Run Codex Agents Inside Enterprise Clouds — Forbes

✔️ OpenAI's latest acquisition could see big changes on the way for its Codex coding assistant — TechRadar