From Models to Machines: Building AI That Actually Delivers with Ash Saxena
From early experiments with dismantled electronics to building AI systems that power real-world machines, Ash Saxena has spent decades at the intersection of research, entrepreneurship, and applied intelligence. Now, as Founder & Chief AI Officer of TorqueAGI, he is focused on one of the most ambitious challenges in technology: enabling robots to perform meaningful work in the physical world.
Ash brings a rare depth of experience, from his PhD work at Stanford alongside Andrew Ng to founding and scaling multiple AI-driven companies. His perspective cuts through the noise of today’s AI hype cycle, offering a grounded view on what is actually working, what is misunderstood, and where the real opportunities lie in robotics and embodied intelligence.
We explore how the shift from data-driven AI to reasoning-based systems is reshaping robotics, why most companies are approaching the problem the wrong way, and what it takes to move from impressive demos to reliable deployment in the real world.
Highlights:
Ash’s journey from building robots as a child to leading AI innovation across academia and industry, including early work on deep learning for robotics
Key inflection points that led him to found multiple companies, including applying AI to unlock access to credit through Catapult
Why “technology-first” companies often fail and the importance of aligning AI with real customer demand and ROI
The evolution of AI from statistical models to deep learning to today’s foundation models and reasoning-based systems
Why the biggest shift in AI is not better models, but dramatically faster time to deployment from years to days or weeks
What Torque AGI is actually building: end-to-end robotic “skills” that combine foundation models, agents, and real-time infrastructure
Why data collection at massive scale may not be the answer and how useful systems can be built with far less data than expected
The gap between AI demos and real-world deployment, and why most demonstrations fail outside controlled environments
A pragmatic roadmap for robotics adoption, from simple tasks today to more complex industrial automation over the next decade
Where Torque AGI fits in the stack as a modular layer that translates AI models into actionable robotic capabilities
The importance of interpretability, safety, and measurable performance when deploying AI into physical systems
The core technical bottleneck in robotics today: bridging deep learning with real-world physics and constraints
Why industrial robotics will see massive value creation in the next 5 to 10 years, while humanoids remain further out
A contrarian take on general-purpose systems: general AI will matter more than general-purpose robots
Where the industry is overhyping progress, especially around humanoid demos, and what is actually working today
Why AI-driven upgrades to existing robots could unlock 10x to 40x increases in productivity without new hardware
How to stay disciplined as a founder in a hype-driven market by focusing on real customer outcomes instead of funding cycles
What a successful deployment looks like, from quick demos to full operational integration in messy real-world environments
Organizational shifts required to adopt AI, including the rise of centralized AI teams driving cross-functional impact
How hiring has changed: agency, adaptability, and the ability to leverage AI tools now outweigh traditional credentials
Why experience at top companies is less important than curiosity, experimentation, and the ability to measure AI performance
Lessons for founders building in AI: stay focused on vertical value and be prepared for rapid shifts driven by large model providers
Learn more about TorqueAGI: LinkedIn | Twitter | Website
Connect with Ash Saxena: LinkedIn | Stanford
Connect with Greg Toroosian: https://www.linkedin.com/in/gregtoroosian/