Hebbia’s George Sivulka: “Bots will be most of the economy within decade”

Hebbia's CEO bets orchestration, not models, will run the AI economy

Hebbia is betting that the lasting value in AI lies in the layer that orchestrates many model calls into complete workflows, not in any single model. Complex professional work has many steps, so an agent that runs the whole process is more useful than a chatbot or search. Sivulka believes this extends well beyond finance and law, to a point where agents account for most of US GDP within a decade.

Takeaways

• Hebbia builds agents for multi-step work, not single tasks. Sivulka says most AI startups ship one-off transactional tools, while humans work through complex processes. Hebbia aims to handle those processes end to end, beyond a chatbot answer or a search, for tasks like investment evaluation.
• Value sits in the orchestration layer, not the model. Hebbia is LLM agnostic and sits on top of any model, which customers can bring themselves. It may orchestrate a thousand GPT-4 calls for one question. Sivulka compares it to a conductor or Excel, and Hebbia stopped training models around 2020.
• GPT-3 convinced him meta learning had arrived. Sivulka argues language is the only way logic is encoded, so LLMs became the obvious route once text, compute and data were sufficient. Playing with GPT-3 in summer 2020 led him to found Hebbia that August.
• Finance and law are the early markets. The first customers were hedge funds and asset allocators in private and public capital. Hebbia measured exposure across documents after the Silicon Valley Bank failure. Many law firms began using it recently for document review and negotiation support.
• Investors backed the company almost immediately. Peter Thiel was among the first to write a check, after a long breakfast in Los Angeles. A party-round pre-seed came in August 2020, then Index Ventures led the November seed and the Series A, and joined the Series B. Hebbia raised $130m in July.
• Agents will drive most of US GDP within a decade. Sivulka says computers already contribute a large share of GDP, and most human tasks remain unstructured. He expects AI to grow the economy rather than eliminate half of jobs, and describes himself as an accelerationist who takes risks seriously.
• The company's name and thesis come from neuroscience. Sivulka's Stanford PhD work in computational neuroscience included recording action potentials from five coral species with a thousand-electrode device. He trained a network with Hebbian learning, the simple brain learning rule that gave Hebbia its name.