Roundtable: Test and Validation in the Age of Agentic AI

Something is breaking in software, and it is not the models. It is the quiet agreement engineers have lived by for 20+ years that tests equal truth. That if the green lights turn on, the system deserves our trust. Agentic AI does not respect that agreement. It writes faster than we can read, edits without asking, and ships logic that looks correct while behaving like it learned its ethics from a speed run. Code is no longer scarce. Understanding is. And the old rituals of validation are starting to feel ceremonial.

That tension is what the Engineering Leadership Community is walking straight into on Feb 18, 2026, with a small group virtual roundtable called Test and Validation in the Age of Agentic AI. This is Part II of the Engineering Judgment in the Age of Agentic AI series, and the title alone signals the shift. Judgment is back in the building. Not as vibes, but as a discipline. Around this moment, governments are publishing research saying agent testing is still a developing science. Conferences are circling the topic. Everyone agrees the problem is real. Few are willing to admit how unfinished the answers are.

The room is intentionally tight. No stages. No panels. Just engineering leaders who are already living with agents in their systems and feeling the drag between velocity and confidence. The kind of conversation where people speak in complete sentences because the stakes demand it. ELC has spent years curating these rooms across San Francisco, Seattle, New York, and Istanbul, quietly assembling more than 8,000 leaders from 3,000 companies who prefer signal over spectacle. This is that culture, pointed directly at validation.

Hosting the conversation is Ben Segal, ex Sr. Director at Swift Navigation, the engineer who led the team behind the Skylark Cloud GNSS Correction Service. Skylark is a system where mistakes do not just throw errors, they move objects in the physical world. It delivers centimeter level accuracy and carries ISO 26262 certification because failure has consequences. Ben Segal holds multiple patents in advanced positioning technologies, but more importantly, he has lived inside environments where testing is not a checkbox. It is a survival instinct.

What this roundtable is really asking is uncomfortable and necessary. When agents write the code, what exactly are we validating? Behavior. Intent. Boundaries. Trust itself. Testing is moving up the stack, and so is responsibility. This is not about frameworks. It is about whether engineering leadership can evolve fast enough to stay credible in a world where machines are junior developers who never sleep and never doubt themselves. The answer is forming in rooms like this, quietly, before it shows up as policy or doctrine.