Sooner or later, any discussion about AI agents touches upon guardrails, but SkipLabs founder Julien Verlaguet tells The New Stack’s Darryl Taft that most AI guardrails are just prompt engineering. “I don’t see anybody who is trying to build real guardrails and real tooling from scratch,” he says
SkipLabs’ Skipper agent pairs an incremental TypeScript implementation with a reactive runtime that reruns only affected tests and updates live services without downtime. On internal benchmarks, it passes more than 90% of backend service tests versus 20% for Claude Code (though Claude Code is a more generalist agent, of course).
One of the arguments the SkipLabs team is making as it extrapolates the current agentic AI trends into the future: if agents are writing the code, the tools shouldn’t be optimized for human readability — that only holds the agents back. Skipper’s TypeScript-compatible type checker is deliberately stricter than standard TypeScript. That’s harder for humans to work with, but more useful for AI reasoning about cause and effect in a codebase.
SkipLabs founder Julien Verlaguet says most AI guardrail claims are just better prompting. He's building the infrastructure layer to prove the difference.
Everyone says AI needs guardrails. Julien Verlaguet wants to know who is actually building them.
Verlaguet, founder of SkipLabs, has spent the last year asking that question, but he doesn't like the answers he keeps finding....
The next step for continuous delivery: An IDP-driven platform
Continuous Delivery solved deployment, but it hasn’t solved the ‘access’ problem. As AI speeds up coding, pipelines remain gated by tribal knowledge and manual handoffs. Join us on April 23 to learn how platform teams bridge the gap between catalog data, environments, and pipelines to eliminate friction and automate consistency.
In this episode of The New Stack podcast with Dilpreet Bindra, senior director, engineering, VMware by Broadcom, we look at how Broadcom is extending its support for Kubernetes and key projects for cloud native. The podcast was recorded live in Amsterdam during KubeCon + CloudNativeCon Europe 2026.
Break the cycle of spending $250bn on manual IT operations with the IT knowledge graph
Rules-based automation solves only 20% of IT problems, leaving a $250 billion manual burden. BigPanda’s IT Knowledge Graph fixes this by connecting fragmented data and institutional knowledge. Learn how this shift drives 430% ROI and reduces outages by replacing reactive, manual operations.
