Why a powerful AI model is no longer enough: a partner at Menlo Ventures on Anthropic and the new wave of startups

How Menlo Ventures Invested in Anthropic

Murphy’s portfolio at Menlo Ventures includes Anthropic, OpenRouter, Lovable, and Legora. In 2024, the fund invested over $500 million in Anthropic’s Series D round. Immediately following this round, Murphy served as a board observer at the company; he no longer holds this role but remains in touch with the team.

Murphy does not consider himself a “power user” of Claude. For daily work with the model, the fund has dedicated partners who build tools based on it, including for portfolio companies. Murphy uses Claude Cowork to process email and prepare quarterly summaries for distribution to colleagues on Slack. He also integrates it with Affinity, Slack, and Harmonic to track signals about promising founders and companies.

Menlo’s first investment in Anthropic occurred during the Series C round. The deal was unconventional and did not fit the fund’s typical parameters: the early-stage fund usually wrote checks of up to $15 million, while the later-stage fund required companies to have an established growth trajectory. At that time, Anthropic had neither revenue nor a clear product.

What proved decisive for Murphy were his personal impressions of the founders. He described Anthropic’s CEO, Dario Amodei, as a rare founder—friendly and energetic on one hand, yet possessing unique depth of knowledge regarding what needs to be built in AI and how to do it. Murphy had similar impressions of other co-founders, including Tom Brown and Daniela Amodei—the team had previously worked on early language models at OpenAI.

The bet was built on simple logic: following the explosive growth of ChatGPT, the market would inevitably desire a strong second player, and the team best equipped to handle this task was one that had already participated in creating a competing technology. By that time, the fund had been systematically investing in AI for about four years and had recently shifted the firm’s focus toward this sector.

Anthropic’s Growth Rates

By May 2026, Anthropic reached annual revenue of $47 billion, up from $9 billion in 2025. Murphy had not seen such growth rates in the internet era, the mobile technology era, or the first wave of cloud services, despite his 25 years in venture capital.

Why the Model Is Not the Main Advantage

Murphy believes that the high-quality model itself has never been Anthropic’s primary competitive advantage. The company managed to retain users by offering more than just API access to the model. This includes products like Claude Code, Claude Code for Life Sciences, and Claude Cowork, as well as the MCP protocol, which allows companies to connect their own data to the model, and the set of Claude Skills. According to Murphy, these add-ons transformed a powerful model into a full-fledged platform.

Open Models vs. Proprietary Models

Murphy separately commented on the balance between proprietary models and open-source solutions. He noted that the market in 2026 is moving toward a multi-model approach faster than expected: companies with large amounts of proprietary data are either training their own models or using open-source solutions to optimize token costs. The most technologically mature companies will distribute workloads—roughly half of requests will go to Claude, and the other half to open or customized models. Most companies, lacking resources for their own research teams, will likely continue to rely on models like Claude. The fund’s portfolio company, OpenRouter, helps companies optimize the choice between different models from the Claude family and other providers based on price, latency, and stability.

Criticism Surrounding Mythos

Murphy addressed the criticism surrounding the launch of the Mythos model line, which is focused on enhanced safety. Part of the AI safety community viewed this move more as a marketing stunt than a substantive security measure. Murphy disagrees: he sees it as a natural evolution of the company as its influence grows and acknowledges that some skepticism may stem from broader concerns about “regulatory capture” of the industry by major players. He noted that Anthropic’s leadership is mission-driven, not solely profit-driven, and explained that the company’s logic is to prepare for the consequences of powerful models in advance—before they fall into the hands of malicious users.

Lovable and Legora: The Fastest-Growing Projects in the Portfolio

Among other portfolio companies, Murphy highlighted Lovable and Legora as the fastest-growing startups he has seen in his 25 years of investment practice. Lovable, which started as an open-source project, enables people without technical skills to create and launch applications—analogous to how Uber once opened the private transportation market to non-professional drivers. Over the past year, the company’s revenue has grown several times over, and the platform has expanded: in addition to an app builder, Lovable now offers hosting and payment processing, and some businesses created on the platform are already valued at over $1 million. Legora, operating in the legal sector—a historically one of the most difficult tech markets for sales—exceeded $150 million in annual revenue by summer 2026.

How the Current Wave of AI Startups Differs from Past Waves

In response to questions about how the current wave of AI startups differs from previous technology cycles, Murphy cited the low barrier to entry for users as the key factor. Unlike the internet or mobile technologies, where mass adoption was preceded by years of changing user habits and device purchases, access to AI requires nothing more than a simple chat interface—already familiar to users from their experience with search engines. This made growth product-led rather than dependent on the scale of the sales department: the path from idea to a product people are willing to use has become shorter than ever. As evidence, he cited companies like Salesforce, which spend hundreds of millions of dollars on tokens.