Episode 21: GenAI breakthroughs, roadblocks and what to learn next

This week on Harnham’s Data & AI Podcast, Senior Growth Manager Jayme Oshaben is joined by Ashish Bansal, founder of StarSpark.AI and former Google leader, for a grounded look at where generative AI is really heading.

Ashish reflects on the breakthroughs that brought us here: from seq2seq models to Transformers and self-supervised learning, and why cost, scale and reliability still determine which ideas make it into production. He explains the common stumbling block teams face with GenAI: it’s easy to get to a promising proof of concept, but closing the gap to a production-ready system demands stronger engineering, cleaner data, and a clear product mindset.

The conversation looks ahead to what’s next:

• Smarter, in-product support that avoids the “please hold” loop.
• AI and robotics shifting from virtual experiments to physical tasks.
• Education reimagined with affordable tutors that encourage curiosity.
• Why GenAI is probabilistic, not deterministic, and why that matters.

On skills, Ashish is direct: embrace new tools, but go beyond them. The advantage now lies in pairing technical depth with communication, product sense, and a growth mindset. As founder of StarSpark, the first AI math teacher designed for K–12 mastery, Ashish shares how education can be transformed. StarSpark uniquely combines alignment with state standards, adaptive grade-level progression, and 100% solving accuracy with personalized, proactive teaching that no other platform delivers.

Whether you’re testing your first agent, scaling a platform, or planning your next role, this episode offers a clear-eyed view on turning GenAI from demo into lasting value.