This week, we feature an excerpt from a post by George Sivulka, CEO and co-founder of Radical Ventures portfolio company Hebbia, on why vertical software still wins in the age of foundation models. Read the full essay on X.
Right now everyone in this industry is asking some version of the same question: in a world where Anthropic, OpenAI, and Google are pouring hundreds of billions of dollars into building the most powerful general-purpose AI systems in the world, why does vertical software need to exist? How does a company like Hebbia earn a right to win? Why don’t the foundation models just eat everything?
I’ve been reading all the discourse about this in the last few weeks—the arguments about moats being dead, the rebuttals, the counter-rebuttals. I’ve heard the pitch that LLMs destroy learned interfaces, vaporize business logic, commoditize data access, and in the end leave vertical software companies with nowhere to hide. And I agree with more of it than you’d expect.
But I think the problem with all of this is that it misses the point.
At the most basic level, enterprise software is just code; it’s an accessible interface and a backend connected to important data and systems of record. But if you think that’s where the value of software comes from, you don’t understand why vertical software companies succeed. The value of enterprise software comes from understanding the process and the organization well enough to make the software do exactly the right thing.
And I think that process engineering, and the network effects that arise from it, will continue to be the foundational advantage for vertical software.
Read the full essay on X.
AI News This Week
-
Data Centres in Space: Less Crazy Than You Think (The Economist)
Of the global data center capacity due this year, 30-50% could be delayed due to permit battles, grid connections, and soaring electricity demand. With current technology, a 1GW orbital data center would cost $51bn versus $16bn on Earth. But if SpaceX’s Starship rocket achieves its promised cost reductions to $200/kg, orbital data centers would cost $12bn and become cheaper than terrestrial ones. Starcloud, a startup founded in 2024, tested the concept by launching a satellite with an Nvidia H100 GPU that successfully trained a language model. They found that standard AI chips can operate reliably in orbital conditions despite radiation exposure, with lower failure rates than anticipated. Companies are already building the supporting infrastructure needed for orbital computing, including Radical Ventures portfolio company Muon Space, which is developing satellite constellations that space-based data centers would require.
-
The Era of Doctor AI is Already Here (Axios)
More than 40 million people ask ChatGPT health-related questions every day, and one in four of its 800 million regular users submits a health prompt weekly. Consumers are already using AI to explain lab results, prep for doctor visits, and access medical guidance. Answer quality still depends on how questions are asked and other inputs, though models are continuously retrained and improving.
-
Where are China’s A.I. Doomers? (NYT)
Chinese consumers are among the most enthusiastic AI adopters in the world, with 69% saying the technology’s benefits outweigh its risks, compared to 35% of Americans. American companies have largely pursued frontier-model development, while Chinese companies have built for everyday life. Driverless taxis operate across dozens of cities, medical chatbots help patients skip hospital queues, and leading models are free to use. Beijing has set targets for AI to reach 90% of Chinese society by 2030, framing the technology as an economic engine rather than an existential risk.
-
AI Fears Temper Interest as Private Equity Firms Weigh Data Company Deals (Reuters)
Financial data providers and research firms have seen sharp stock declines in recent months, making them attractive takeover targets on paper, but PE firms are hesitant to move forward. Bankers say they cannot accurately value companies as executives struggle to predict whether their business models will survive AI disruption. “Public market investors are trying to figure out where the world goes,” said Radical Ventures co-founder and Managing Partner Jordan Jacobs. “AI is such a new technology, and the improvements in new application areas are so dramatic, and the new opportunities are happening so quickly, that it’s very hard to predict things years in advance.”
-
Research: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games (MIT et al)
Researchers developed AI GAMESTORE, a benchmark that uses 100 simplified web games to evaluate AI performance against human performance. Games are automatically generated in 30 minutes using Claude Sonnet 4.5 and Gemini Flash 2.5, creating a scalable evaluation platform. The platform’s continuous generation of new game variants addresses benchmark saturation, offering a novel approach to measuring general intelligence through diverse cognitive tasks that resemble unstructured real-world human activities.
Radical Reads is edited by Ebin Tomy (Analyst, Radical Ventures)
