Agents need governed work, not activity

GitLab’s Act 2 note is one of the clearer public signals from the software platform world: the next fight is not over more AI activity. It is over governed work.

The company announced a restructuring at the same time it laid out its agentic engineering thesis. GitLab plans to reduce its country footprint where it has small teams, remove management layers, reorganize R&D into roughly 60 smaller teams, and rewire internal reviews, approvals, and handoffs with agents. The human cost of that restructuring should not be glossed over. People are being asked to make career and life decisions inside a hard transition.

Do not copy the layoff. Read the harder signal: AI has moved from feature work into company architecture.

In GitLab Act 2, Bill Staples names the deeper platform shift. Software will be built by machines and directed by people. Agents will plan, code, review, deploy, and repair. Machine-directed delivery creates more demand for software, not less. It also changes what the platform has to provide.

The strongest sentence in the post is not about code generation. It is about outcomes: “Enterprises don’t need agent activity. They need running software that moves the business forward.”

That line should travel outside enterprise software.

SMB owners do not need AI activity either. They need the work done, proven, governed, and tied to the business result. A plumbing company does not need an agent that chats about dispatch. It needs the right job assigned, the exception surfaced, the customer promise preserved, the photo evidence attached, the invoice checked, and the callback risk reduced. A clinic does not need more automated notes. It needs the eligibility rule, denial pattern, handoff owner, documentation burden, and patient promise made visible.

GitLab’s five architectural bets point at the same requirement from a larger scale: machine-rate infrastructure, lifecycle orchestration, connected context, governance in the core, and one platform across human-owned, agent-assisted, and agent-autonomous work. Strip away the enterprise packaging and the operating rule is simple. Agents need a work system around them.

Many SMB AI deployments will fail at this boundary.

A model can draft faster than the office can decide. An agent can classify faster than the company can define the category. A workflow can run faster than the owner can see whether the result was good. Speed without an inspectable standard creates noise with a progress bar.

The answer is not to slow the agents down. Build the governing surface around the work.

Hadto’s version of that surface is owner-operator infrastructure: named workflows, visible rules, evidence trails, scoreboards, exception queues, apprenticeship paths, and authority to improve the method. The goal is not to replace the domain expert with a cheaper tool. It is to convert the domain expert into someone who can govern the work, train the next person, and make the system better.

This points away from the public-company restructuring playbook.

A large software company can respond to agentic AI with compression: fewer layers, fewer roles in some functions, smaller teams, consumption pricing, and more capital behind the new platform. An SMB usually cannot compress its way into competence. It is already thin. The missing asset is not another person to remove. It is the operating substrate that lets good people own more of the business.

GitLab is right that context becomes the moat. The same is true in a local business. The useful context is not only data in a database. It is the remembered exception, the judgment behind the estimate, the real warranty boundary, the no-fit signal, the callback pattern, the payer nuance, the customer promise, and the reason an experienced person knows when the standard case is not standard.

If that context stays private, agents make the business more dependent on whoever remembers it. If that context becomes a governed work system, agents can expand who is capable of owning the work.

That is the Hadto reading of GitLab Act 2. Agentic AI is not just a new tool layer. It is pressure on the business design underneath the tools. The enterprise world is responding with restructuring and platform rebuilds. SMBs need the same seriousness, but a different end state: more owners, clearer proof, stronger governance, and work systems that make judgment teachable.

The next useful question is not “how many agents did we run?”

It is: did the work move, did the business improve, and can a responsible owner prove why?