High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / belief

Published · transcript-backed

Aishwarya Naresh Reganti: belief

11 Jan 2026 Lenny's Podcast Why most AI products fail: Lessons from 50+ AI deployments at OpenAI, Google, and Amazon

“Until now, I think LLMs have been our most commonly used modules, but as humans, we are multimodal creatures, I would say.”

— Aishwarya Naresh Reganti

Source trail

Everything needed to verify it.

Speaker
Aishwarya Naresh Reganti
Attribution
Verified speaker
Claim type
belief
Recorded
11 Jan 2026
Publisher
Lenny's Podcast

Transcript context

…That's so cool. So essentially agents anticipating what you want to do and getting ahead of you and I've solved these problems for you or I think this is going to crash your site. Maybe you should fix this thing right here or I see the spike here and let's refactor our database. Amazing. What a world. Okay. Ash, what do you got? I'm all in for multimodal experiences in 2026. I think we have done quite some progress in 2025, and not just in terms of generation, but also understanding. Until now, I think LLMs have been our most commonly used modules, but as humans, we are multimodal creatures, I would say. Language is probably one of our last forms of evolution. As the three of us are talking, I think we're constantly getting so many signals. I'm like, "Oh, Lenny's nodding his head, so probably I would go in this direction or Lenny's bored, so let me stop talking." So there's a chain of thought behind your chain of thought and you're constantly altering it with language that dimension of expression is not explored as well. So if we could build better multimodal experiences that would get us closer to human-like conversation richness. And you will also, just given the kind of models, there's a bunch of boring tasks as well, which are ripe for AI. If multimodal understanding gets better, there are so many handwritten documents and really messy PDFs that cannot be passed even by the best of the models as of today. And if it's possible, there'll be so much data that we can tap into. Awesome. I just saw Demis from DeepMind, AI, Google, whatever they call the whole org, talking about this where he thinks that's going to be a big part of where they're going, combining the image model work, the LLM, and also their world model stuff, Genie, I think is what it's called. Yes. So that's going to be a wild, wild time. Okay. Last question. If someone wants to just get better at building AI products, what's just maybe one skill or maybe two skills that you think they should lean into and develop?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.

Search evidence