As AI systems handle more tasks, business leaders are racing to build accountability and verification systems to catch mistakes before they cause damage.

What happened: Business executives gathered at Fortune Brainstorm Tech to discuss the core challenge of deploying agentic AI systems (AI that independently performs tasks). The consensus is that transparency and the ability to trace every step an AI takes—and understand why it made mistakes—is now a top priority. Companies like Thomson Reuters are building what they call 'fiduciary grade' products centered on transparency, data privacy, subject matter experts, and reliable content.

Why it matters: As AI systems take on more and more work, humans cannot keep up with verifying all of it. One panelist noted that 'you end up in this space where you've got so much work that's been done, so much work to audit, that you can't truly be accountable.' A common solution emerging is the 'LLM as a judge' technique—using one AI system to check the work of another, similar to a newsroom where an editor reviews a writer's output. The critical point is that separate AI systems must do the verification; as one executive stressed, 'you don't want AI to grade its own work.'

What to watch: Techniques from safety-critical industries (like aviation and nuclear power) developed decades ago are being imported into everyday AI practice. Computer coding is about one year ahead of other industries in developing these verification methods, suggesting the pattern will accelerate across sectors as agentic systems become more widespread.