OpenClaw

Several recent articles are raising alarms about OpenClaw: an open-source AI agent with full access to the user's computer, credential leaks, Cisco calling it a nightmare. All of that is serious. But it is not the most interesting part of the story.

OpenClaw is not succeeding just because it is technically impressive. It is succeeding because it promises something many people have been quietly wanting: to delegate everything. Not just tasks, but responsibility. Not just execution, but decision-making.

It is not a copilot. It is an operator. It uses the computer like a human does — navigating tools, accessing files, connecting systems, and simply solving. Or at least appearing to solve. The mental model behind it is simple and dangerously seductive: "just handle it." No clear brief. No explicit criteria. No metrics. No plan.

From a Product Management perspective, this is not disruptive innovation. It is an uncomfortable mirror. It shows how willing we are to trade clarity for convenience and thinking for speed.

When an agent receives total autonomy, it is not just executing. It is interpreting goals, defining priorities, and making decisions in your place. And in most cases, without you being able to explain afterward why something went right or wrong. The classic outcome that seems like magic when it works and inexplicable when it fails.

There is a phrase commonly attributed to Peter Drucker: what cannot be measured cannot be managed. Regardless of who said it first, the logic is relentless. If you do not define criteria, do not set limits, do not measure, you are not automating. You are outsourcing judgment.

And then the problem stops being just about information security. It becomes about cognitive security.

A recent Anthropic study already points in this direction: the indiscriminate use of AI in decision-making weakens the human habit of thinking, evaluating alternatives, and anticipating consequences. Not because AI thinks too much. But because the human thinks less.

This is not a case against autonomous agents. They are inevitable and, well designed, extremely powerful. The problem starts when autonomy becomes a black box and convenience becomes an excuse to abdicate responsibility.

As a PM, the role is not to stop automation. It is to create guardrails. To define what can be delegated and what requires human judgment. To demand explainability. To create checkpoints. Autonomy without governance is not productivity. It is an illusion.

In the end, OpenClaw does not just expose a technical risk. It reveals something more uncomfortable: our urgency to stop deciding.

The question is not whether we are going to automate. That already happened.

The question is: who is still thinking?