Interview, Fireside Chat
Former Chief Scientist at Salesforce, Richard Socher | You.com, LLMs, AI Agents, Complex Work
- u.com secured a $50 million Series B financing round, led by Georgian with participation from NVIDIA, Salesforce Ventures, SoftBank Ventures Asia (now SBVA), DuckDuckGo, Gen Digital, and Day One Ventures, bringing total funding to $99 million.
- The company is pivoting its market positioning from a search engine to an "AI-driven productivity engine" focused on complex knowledge work, citing superior accuracy over competitors for high-stakes career decisions.
- Revenue is doubling every quarter in 2024, driven by a dual business model comprising a direct-to-consumer subscription line and a high-volume API line of business.
- API traffic now exceeds end-user traffic on the u.com property, indicating that businesses are integrating u.com's web-connected, model-agnostic capabilities into their own products.
- Subscription pricing offers access to multiple LLMs including models from Anthropic, Google Gemini, and Llama 3.1/3.2 at a lower cost than competitors, bundling them into a single package with web connectivity.
- Enterprise features now support uploading entire data rooms (e.g., 50 MB of files including PDFs, CSVs, and PowerPoints) for AI-driven due diligence and internal data analysis, surpassing standard context window limitations of competitors like Google Gemini.
- New "Multiplayer" capabilities allow teams to share custom workflows, agents, and projects with shared underlying data, enabling collaborative research and diligence within secure, private environments.
- AIX Ventures, co-founded by Richard Socher, has invested in over 30 AI startups, including Hugging Face, Perplexity, Whisper, and Weights & Biases, leveraging a team of full-time AI practitioners to identify emerging trends.
- Richard Socher's background includes inventing prompt engineering in 2018, founding Metamind (exited to Salesforce in 2015), and serving as Chief Scientist at Salesforce where he led AI efforts.
- The company differentiates through a proprietary search stack and "citations that scroll," which highlight the exact source of facts on long documents, addressing the "garbage in, garbage out" problem prevalent in RAG-only competitors.
- Pricing models are being aligned with enterprise usage through a hybrid of seat-based and consumption-based models to address the issue of low adoption rates (often <10%) in traditional seat-licensing deals.
- Target customer profiles include hedge funds (e.g., a top 20 global fund), biotech firms, cybersecurity unicorns (e.g., Mindcast), and publishers (e.g., Working Bit).
- Socher forecasts a Jevons Paradox for AI, where the decreasing marginal cost of intelligence leads to increased consumption of AI across sectors like marketing, flight scheduling, and customer support, rather than reduced usage.
- The future of AI agents is predicted to shift from prototype to production, with the expectation that agents will eventually surf the web and execute complex workflows, fundamentally changing the internet and ad economy.
- Socher warns against "hype bubbles" and inflated valuations, noting that while AI is transformative, achieving robust, error-free automation of long sequential workflows (e.g., 15-step processes) requires significant engineering to avoid compounding failure rates.
- AIX Ventures strategy avoids "unicorn seed rounds," preferring to back companies with revenue and concrete milestones to balance high-risk early-stage ventures with later-stage returns.
- u.com is actively hiring across all functions, including engineering, VP of Engineering, sales, marketing, and enterprise roles, to support its scaling roadmap.
- Future technology roadmaps include an "10X Research Mode" and enhanced capabilities for agents to perform code execution and web interactions, aiming to automate roles rather than just assist them.
- Socher identifies a coming explosion in biology applications, predicting AI will transform biology from a science of observation to an engineering discipline capable of writing novel proteins for medicine.
- Key competitive disadvantage for others is cited as reliance on "fake citations" and random links, whereas u.com's accuracy is critical for customers where incorrect data leads to career or financial risk.
- The "Genius Mode" (now standard agents) utilizes retrieval-augmented generation to decide when to search, write code, or follow up on queries, moving beyond simple Q&A to complex workflow automation.