CoCounsel

Jake Heller explains how Casetext chose legal work to automate, built CoCounsel around lawyers’ real workflows, and sold the company to Thomson Reuters for $650 million. He recommends breaking expert work into specific steps, using prompts where needed, and testing each step with evaluations grounded in domain expertise and customer examples. For sales, he advises pricing around customer value, building trust through comparisons and pilots, and training customers so they use the product successfully. His central lesson is to stay focused on product quality and solve large, valuable problems.

Jake Heller explains how Casetext spent years building legal research tools before GPT-4 made a much larger product possible. After seeing the model, he and his co-founder quickly committed the company to CoCounsel, an AI legal assistant, and used customer reactions to win over employees and skeptics. He describes building reliable legal workflows by breaking tasks into steps, testing prompts against gold-standard answers, and handling practical details such as document integrations and OCR. The talk argues that vertical AI products can create substantial value by combining models with domain expertise, application logic, and careful evaluation.

Jake Heller recounts Casetext’s 10-year path from a crowdsourced legal research site to an AI legal assistant, including false starts with different customer segments and products. The talk explains how access to GPT-4 enabled CoCounsel, whose demo showed lawyers how document review and research could be compressed from days into minutes. It emphasizes matching technology to customer needs, building accuracy and scale around raw models, and staying ready through years of work with the same customers. Heller also argues that LLMs open opportunities across the infrastructure and application layers, and encourages founders to build with them.