Is AI Entering Its Second Wave? A Conversation with Touring Capital’s Samir Kumar
After a year in which we saw a dramatic rise in the number of AI start ups, and increasingly high valuations (despite no or limited revenue) of those companies, some industry watchers were starting to talk of an 'AI bubble' similar to the dot-com era. Certainly history tends to repeat itself, and with new innovations, initial exuberance can outweigh a balanced approach to evaluating technology. That said, the pace of innovation in this AI paradigm shift does seem accelerated and more compressed than in prior tech innovation shifts.
As we enter the second quarter of 2025, could it be that we’ll see the beginning of the second wave?
To explore this, I sat down with Samir Kumar, General Partner at Touring Capital, who has seen multiple tech cycles play out. We discussed what’s real, what’s overhyped, and where AI is poised to deliver real economic impact over the next few years. As Samir shared:
“Humans tend to overestimate the impact of a new technology in the near term and underestimate it in the long run.”
Perhaps seeing that in the AI hype cycle now.
Here are some key takeaways from my conversation with Samir on emerging trends and his vision for the industry's trajectory over the next 2-5 years:
It has been an interesting arc to observe how AI solutions are being pitched. Post ChatGPT there has been a frenzy of companies building foundation models (both LLMs and diffusion models) along with those building UI/UX wrapped around API based access to models. These “API wrappers” were frowned upon as lacking sustainable differentiation. Ironically now what we are seeing is that the performance differentiation and half life of state-of-the-art models is shrinking -while the skepticism for API wrappers has reversed as agentic AI applications that capture the nuances of task and domain specific workflows become a source of differentiation. The recent Deepseek developments should accelerate focus shifting to the application layer.
I am personally excited about the potential for AI to be applied to scientific and engineering workflows. AI excels at navigating combinatorial complexity and the data deluge. From materials to drugs to biology, all are well positioned to see a dramatic acceleration in R&D from AI. AI enabled inverse design should allow us to discover novel solutions to scientific and engineering design problems that humans may have stumbled upon via trial and error or not at all! AI is also a major catalyst for faster and more robust simulations. Faster high quality simulations mean more (accurate) predictions of real world performance and more efficient physical testing before products can be commercialized.
I think this is the key question around the nature of the disruption to work and especially information work of which we are still in the very early innings. As agentic AI matures, becomes more reliable and trustworthy it will naturally be a catalyst for replacing human services and labor. As AI agents start doing the work of humans, that has consequences for the infrastructure they interact with and the rate at which information is processed and generated. This will be disruptive to the structure of teams, departments in organizations and also how organizations interact with other organizations on the outside. Taken to what sounds like an extreme today, imagine what happens to HR and recruiting in this scenario? Ultimately the optimistic view here is that agentic AI should allow humans to uplevel their productivity. Humans working with agents should be more like maestros directing the objectives and behaviors of agents and getting more done faster.
We should expect it to be widespread and across verticals. Low hanging fruit to date have been marketing, copyright, sales, customer support and all of these will see increasing adoption but I am excited about what lies ahead for product design, development and testing for both software and physical products
There is a lot of excitement and a flurry of activity in the robotics space as well (and of course robotics then applied to a wide set of verticals). The software tools we use to train robots to perform tasks will all benefit from genAI innovations in AI powered code gen, simulation, and synthetic video. Multimodal AI models can incorporate the sequences of mechatronic actions when physical robots move as another sequence modality to train on and optimize.
Yes absolutely. The second wave of the AI platform shift should see us go from the experimentation and tinkering phase to more crystallized use cases that can be commercialized with ROI metrics. AI4Science can span the spectrum of pure research and AI assisting in making headway in hard, or even problems thought to be previously intractable but also real, commercial applications that can be operationalized on the other end. Bringing new materials, drugs to market is a long, winding, costly process. The ROI metrics for compressing, speeding up these cycles are very clearly measurable. In the materials and chemicals domain, regulatory compliance, sustainability needs can become urgent business priorities and the role of AI in finding solutions more efficiently. Verticalized industry specific AI is something I am quite excited about and where I think a lot of economic value is still to be created.
Is AI Entering Its Second Wave? A Conversation with Touring Capital’s Samir Kumar
After a year in which we saw a dramatic rise in the number of AI start ups, and increasingly high valuations (despite no or limited revenue) of those companies, some industry watchers were starting to talk of an 'AI bubble' similar to the dot-com era. Certainly history tends to repeat itself, and with new innovations, initial exuberance can outweigh a balanced approach to evaluating technology. That said, the pace of innovation in this AI paradigm shift does seem accelerated and more compressed than in prior tech innovation shifts.
As we enter the second quarter of 2025, could it be that we’ll see the beginning of the second wave?
To explore this, I sat down with Samir Kumar, General Partner at Touring Capital, who has seen multiple tech cycles play out. We discussed what’s real, what’s overhyped, and where AI is poised to deliver real economic impact over the next few years. As Samir shared:
“Humans tend to overestimate the impact of a new technology in the near term and underestimate it in the long run.”
Perhaps seeing that in the AI hype cycle now.
Here are some key takeaways from my conversation with Samir on emerging trends and his vision for the industry's trajectory over the next 2-5 years:
It has been an interesting arc to observe how AI solutions are being pitched. Post ChatGPT there has been a frenzy of companies building foundation models (both LLMs and diffusion models) along with those building UI/UX wrapped around API based access to models. These “API wrappers” were frowned upon as lacking sustainable differentiation. Ironically now what we are seeing is that the performance differentiation and half life of state-of-the-art models is shrinking -while the skepticism for API wrappers has reversed as agentic AI applications that capture the nuances of task and domain specific workflows become a source of differentiation. The recent Deepseek developments should accelerate focus shifting to the application layer.
I am personally excited about the potential for AI to be applied to scientific and engineering workflows. AI excels at navigating combinatorial complexity and the data deluge. From materials to drugs to biology, all are well positioned to see a dramatic acceleration in R&D from AI. AI enabled inverse design should allow us to discover novel solutions to scientific and engineering design problems that humans may have stumbled upon via trial and error or not at all! AI is also a major catalyst for faster and more robust simulations. Faster high quality simulations mean more (accurate) predictions of real world performance and more efficient physical testing before products can be commercialized.
I think this is the key question around the nature of the disruption to work and especially information work of which we are still in the very early innings. As agentic AI matures, becomes more reliable and trustworthy it will naturally be a catalyst for replacing human services and labor. As AI agents start doing the work of humans, that has consequences for the infrastructure they interact with and the rate at which information is processed and generated. This will be disruptive to the structure of teams, departments in organizations and also how organizations interact with other organizations on the outside. Taken to what sounds like an extreme today, imagine what happens to HR and recruiting in this scenario? Ultimately the optimistic view here is that agentic AI should allow humans to uplevel their productivity. Humans working with agents should be more like maestros directing the objectives and behaviors of agents and getting more done faster.
We should expect it to be widespread and across verticals. Low hanging fruit to date have been marketing, copyright, sales, customer support and all of these will see increasing adoption but I am excited about what lies ahead for product design, development and testing for both software and physical products
There is a lot of excitement and a flurry of activity in the robotics space as well (and of course robotics then applied to a wide set of verticals). The software tools we use to train robots to perform tasks will all benefit from genAI innovations in AI powered code gen, simulation, and synthetic video. Multimodal AI models can incorporate the sequences of mechatronic actions when physical robots move as another sequence modality to train on and optimize.
Yes absolutely. The second wave of the AI platform shift should see us go from the experimentation and tinkering phase to more crystallized use cases that can be commercialized with ROI metrics. AI4Science can span the spectrum of pure research and AI assisting in making headway in hard, or even problems thought to be previously intractable but also real, commercial applications that can be operationalized on the other end. Bringing new materials, drugs to market is a long, winding, costly process. The ROI metrics for compressing, speeding up these cycles are very clearly measurable. In the materials and chemicals domain, regulatory compliance, sustainability needs can become urgent business priorities and the role of AI in finding solutions more efficiently. Verticalized industry specific AI is something I am quite excited about and where I think a lot of economic value is still to be created.