Shomik Ghosh, Partner at Sierra Ventures – Interview Series

Shomik Ghosh is a Partner at Sierra Ventures. He previously was a Partner at Boldstart Ventures, where he focused on investing at inception into technical founders building products to solve enterprise pain points, such as Cloudquery, Kiln AI, and Noded AI. Prior to that, he was a growth stage investor at Top Tier Capital, investing from Series B through pre-IPO, such as CircleCI, Anaplan, and Shape Security.

Sierra Ventures is an early-stage venture capital firm focused on backing innovative enterprise and deep technology startups. The firm invests primarily at the seed and Series A stages across areas such as artificial intelligence, cybersecurity, enterprise infrastructure, and cloud technologies. In recent years, Sierra Ventures has placed particular emphasis on early AI investments, supporting companies building foundational AI platforms, machine learning infrastructure, autonomous systems, and enterprise AI applications. Through its investment strategy and network of experienced industry advisors, the firm helps emerging technology companies refine product strategy, scale operations, and accelerate adoption of advanced AI solutions across industries.

You’ve transitioned from growth-stage investing at Top Tier Capital Partners to leading early-stage AI investments at Sierra Ventures, after years backing breakout companies at Boldstart Ventures. How has that journey shaped the way you distinguish frontier AI from applied AI today?

A lot has changed in that timeframe. AI is a massive enablement shift that has permeated industries faster than previous technology paradigm shifts as AI is standing on the shoulders of the shifts before. Cloud computing, PC/Mobile devices, and each wave of prior AI advancements has provided the building blocks for modern AI to spread rapidly. This is also why the impact feels so quick and sudden which drastic moves happening in the stock market and even impacting modern warfare. What we look for is founders who are taking a step into the future. They are embracing all the risks of building functionality and capabilities for a world that is yet to exist but in term are able to wow customers with never before seen results enabling faster scale. In both Frontier and Applied AI, this exists from robotics to vertical AI applications. 

In practical terms, how do you define “frontier AI” versus “applied AI” when evaluating early-stage startups, and where do you see the biggest misconceptions in how these categories are discussed across the broader AI narrative?

Frontier AI means leveraging technology to tackle problems at the edge of what’s possible. To date we have not had robots meaningfully enable industry outside of warehouses, we’ve not had new semiconductor chips or glasses designed using novel laser technology and raw materials. Applied AI means leveraging technology to tackle problems that are known today but heretofore were not possible to solve at the same extent. A good example is voice agents where companies like Smallest AI are helping customers deliver human like chat experiences and delivering outcomes to customers rather than a product that helps to achieve the outcome. This aspect of delivering outcomes vs helping improve process is a key shift that Applied AI is bringing to industries. 

From your vantage point working closely with founders across model innovation, robotics, and vertical AI, where are the most meaningful breakthroughs happening right now?

Breakthroughs are happening everywhere! AI Code gen is enabling faster product cycles than before. Models are delivering new capabilities with memory management and RL environments tailored to various use cases so that hallucinations are reducing rapidly while accuracy for knowledge work improves exponentially. This all feeds upon itself. In robotics, we’re seeing early signs that scaling laws are working just like in LLMs. This is a huge breakthrough as before LLMs were mostly text or image based but now models trained on the physical world that have to understand physics are showing similar scaling laws. New papers like the Recursive LLMs paper shows how models are able to improve themselves by working together. We’re seeing System 1 and 2 model structures start to emerge that are similar to dynamics we see in the brain. Domain specific models are becoming easier to train and distill from frontier OSS reasoning models to help Vertical AI builders deliver better outcomes to customers.

When assessing an early AI company, how do you balance technical novelty against product-market fit and real-world customer traction?

In the end, technical novelty in itself is typically more useful for the research field. In the foundation models for example, technical novelty could actually lead to a breakthrough that then presents a new scaling vertical. But for most startups, technical novelty is a means to an end to deliver a better outcome for customers. Startup founders should not build something just because it’s technically hard to build, but rather as a result of building in that manner, it leads to better customer results and also a better moat around the business that makes it harder for others to copy. In the age of AI code gen, a lot of the technical novelty can quickly be breached so more and more it’s about understanding outcome driven engineering vs just plain technical engineering.

Beyond the technology itself, what do you specifically look for in a founder building an AI company at the earliest stages?

We want to see founders who are building for the future that has not yet happened. Making calculated bets on agents, models, and hardware improvements that are likely to happen in the near term and building products that capitalize on that. We then want the founder to explain why that future will happen and why building for that future now will improve customers’ lives 10x in the future by preparing for that now. We also want founders who are completely embracing AI. If you’re not using Cursor, Codex, Claude Code to experiment and learn, it is hard to envision the future given the pace of improvement that those products are making to the software universe. Those changes have downstream impact on the hardware universe since increasingly hardware and software is tightly integrated to enable autonomous decision making by the hardware to deliver better customer outcomes.

What signals suggest that a technically ambitious AI company has the potential to evolve into a scalable, enterprise-ready business rather than remaining a research-driven effort?

Usually founders who have started in a research driven effort have an end goal in sight. They may want to continue doing research to advance the field, but they also understand that monetization the applied research helps to provide the fuel for those advancements. So really we’re just trying to understand how a founder who is currently in research mode is thinking about the applications of that research and what hypotheses they have to test progress of the research in the world along the way to derisk the research stage.

For founders building highly technical platforms without immediate revenue visibility, how should they structure milestones and investor conversations differently from startups with clearer near-term monetization paths?

Very hard to say. Each startup has unique aspects to it. A robotics company may not have revenue visibility for a long time but milestones along the way could be emergent capabilities, scaling laws in models, actions that were previously unable to occur. In AI infrastructure, it may be delivering to 2-3 design partners a product that provides outcomes teams are delighted by and willing to take the bet of using the product in production even though it’s early. In vertical AI, you typically have a clearer near term path to monetization because if you deliver a customer outcome in a vertical that solves a big pain point, customers are willing to pay for that typically immediately. 

There has been significant momentum around startups building AI agents—what is your perspective on the long-term success potential of companies focused primarily on autonomous agents within enterprise environments?

Jensen Huang of Nvidia said Openclaw was the ChatGPT moment for the agentic era. I think that says it all. The timeline for agents in the enterprise is no longer a long-term bet but one that is quickly coming to fruition whether enterprises want it to or not as computer use, browser, and personal agents proliferate through the orgs starting from bottoms up usage. The era is here, enterprises are going to be adopting agents in every aspect of the org, and they will need governance, security, monitoring, infrastructure, compute, and data rails to serve all of this.

What patterns are you seeing in the types of AI founders or domains that are attracting sustained investor confidence versus those that may be over-indexing on narrative?

I think there’s too much areas of opportunity for founders to be building in the same areas. Legal tech like Harvey, Legora, Eudia have all done well but yet there’s still new companies coming out every day in this area. My message for founders would be AI is a massive enablement shift. It is impacting every aspect of the world. Given that, the surface area of products to build and problems to solve is unlimited. Think bigger than just going after a successful company that you’ve seen raise a lot of money. We can use AI to deliver life-changing outcomes, and so I would encourage founders to spend more time thinking about problems they would love to solve and then working backwards on how AI could help solve them.

Looking ahead, how do you expect early-stage AI company building to evolve over the next five years as capabilities mature and markets stabilize?

I’m not sure “market stabilization” is a term I would use. I think AI is improving exponentially and as a result there will be lots of disruption. But disruption creates endless opportunities and so early-stage company building is entering one of the most dynamic eras we have seen in recent years. We forget but companies like OpenAI and Anthropic are less than a decade old and are already considered mega cap tech companies. There is a window of time as capabilities expand rapidly for so many massive companies to be built. It is one of the most exciting times in tech that I’ve experienced in my life.

Thank you for the great interview, entrepreneurs who wish to learn more should visit Sierra Ventures.