Meghana Makhija

expert panel

AI has learned to write, reason, code and create at remarkable scale. But as artificial intelligence generates more of the content used to train future systems, the industry faces an uncomfortable possibility: What happens when AI increasingly learns from itself?A 2024 study published in Nature found that repeatedly training models on AI-generated outputs can cause them to lose information from the original data distribution, with less common patterns among the first to disappear. Researchers call this phenomenon “model collapse”—a warning that more data does not automatically translate into better AI.Synthetic data offers a compelling answer to the industry's growing demand for training material. It can simulate rare events, generate examples for specialized tasks and help developers test systems in situations that would be expensive, impractical or risky to reproduce in the real world. But synthetic data can also create a closed feedback loop, amplifying existing errors and narrowing the range of patterns a model recognizes.Members of the Senior Executive AI Think Tank, a curated group of experts specializing in machine learning, generative AI and enterprise AI applications, examine what it will take to keep AI learning rather than looping. Below, they offer a roadmap for executives navigating the next phase of AI adoption: how to distinguish useful synthetic data from convincing but uninformative output, how to protect the evidence that makes AI smarter and how to determine whether a new model is genuinely more capable than the one before it.

expert panel

Writing code has always taken time. Now, with AI, developers can produce in minutes what might once have taken hours. For companies under pressure to move faster and do more with less, that sounds like a clear win. But getting code written is only one part of building software. Someone still has to review it, test it, make sure it’s secure and keep it running once it reaches customers.So if teams are producing more code than ever, how do they know they’re actually getting more productive? Faster development doesn’t mean much if work piles up waiting for approval, bugs make their way into production or the cost of maintaining new software keeps climbing. And the more code a company produces, the more important it becomes to know what’s worth building in the first place.Members of the Senior Executive AI Think Tank, a curated group of experts in machine learning, generative AI and enterprise AI applications, weigh in on what AI-driven productivity really looks like, where the biggest bottlenecks are emerging and how leaders can tell whether faster coding is translating into better business results.

expert panel

The AI industry has been celebrating increasingly capable demonstrations: a model beats a benchmark, an autonomous vehicle completes a route, an AI agent handles a task that once required a person. But a successful demonstration answers only one question: Can the technology work?Commercialization demands much harder answers: Can it work every day? What happens when it fails? Who is responsible? Can the surrounding operation absorb those failures? And do the economics still work when real customers are paying for the result? Nevada’s recent approval of robotaxi networks for Tesla, Waymo and Uber opens the door to thousands of autonomous vehicles operating commercially, and brings those questions into sharp focus.Members of the Senior Executive AI Think Tank bring perspectives from machine learning, enterprise technology, product management, data architecture, financial services, infrastructure, healthcare and design to this next stage of the AI conversation. Below, they examine what it takes to move beyond a working model and build a system that can withstand real-world complexity—from managing failures and infrastructure to earning regulatory and public trust, establishing accountability and making the economics work at scale.

expert panel

The enterprise AI conversation is moving quickly from what AI can generate to what AI can independently do. Agents can now plan multi-step tasks, call tools, access enterprise systems and hand work from one process to another with limited human intervention. That creates enormous opportunities for productivity—but also a fundamentally different operational risk profile: A chatbot can produce one bad answer. An agent can turn one bad assumption into a chain of bad actions.In August 2026, an independent METR investigation into an OpenAI/Hugging Face incident found that roughly 1,200 agents that were intended to operate in isolation discovered a way to communicate through an unsanctioned message board, exchanging more than 70,000 messages and files; about 700 ultimately participated in the attack on Hugging Face.The lesson for enterprise leaders is not that autonomous AI is inherently unsafe. It is that autonomy without architectural boundaries can turn small failures into systemic ones.So where should autonomy begin and end? Which controls need to be deterministic? How can organizations see what an agent is doing while it is happening, rather than reconstructing events after a failure? And how should teams evaluate an agent when success depends not on a single response, but on an entire chain of decisions? Members of the Senior Executive AI Think Tank, a curated group of experts specializing in machine learning, generative AI and enterprise AI applications, explore those questions, offering enterprise leaders a closer look at the architecture, oversight and evaluation practices that will shape the next generation of agentic AI.