The Shift to Agentic Computing Has Begun, But Security Is Still DIY

In the early days of the Internet, people were terrified of buying stuff online. Not because they didn't want stuff. We all want more stuff, but people were terrified of typing their credit card number into a browser.

"What if someone steals it?" "What if the site isn't real?" "What if the internet is just a fad?"

Remember those days?

My current obsession is Agentic AI, specifically OpenClaw.

This week, five new vulnerabilities were dropped for OpenClaw. That's five ways your AI agent can be turned against you. In seven days.

My view is that technology doesn't achieve mainstream adoption until it's secure enough for ordinary people to trust it.

Credit cards on the internet didn't explode because people suddenly loved typing in 16-digit numbers. They exploded because SSL, fraud protection, and chargeback mechanisms made it safe enough.

The technology worked and security made it adoptable.

I think the same security-adoption challenges plague the autonomous vehicle and humanoid robotics industries currently. OpenClaw and agentic AI is at that same inflection point.

The technology works. 250K+ GitHub stars, making it one of the most successful open source projects ever. The shift to agentic computing has clearly begun.

But the security? That's still DIY. If your DIY skills are anything like mine, you'll know that DIY is not usually professional enough. What makes this even more challenging is that OpenClaw's attack surface is vast: third-party skills with unknown provenance, credentials floating in environment variables, network services listening where they shouldn't, sandboxes that aren't as isolated as we think.

The uncomfortable truth though is that we're in the early adopter phase of agentic computing, and early adopters have always had to roll their own security and take on greater risks.

The credit card companies didn't build fraud detection overnight. Early web merchants wrote their own encryption. The infrastructure we take for granted today was built by people who couldn't wait for it to exist.

What this means for leaders:

If you're deploying OpenClaw (or any AI agent framework) today, security isn't a feature you can defer. It's a prerequisite.

The shift to agentic computing has begun. That part isn't a prediction anymore — it's a headline.

What comes next depends on what we do in the next 12 months. Audit your agent deployments. Treat third-party skills like untrusted code. Isolate your instances. Build governance before you need it.

Credit cards didn't survive the early chaos by accident. Fraud detection, SSL, chargebacks. Someone built all of that. Deliberately. Before the breach that would have ended it.

Be that person for your organization. The agents are already running. The question is whether you're running them, or they're running you.