Town pivots to neutral AI assistant to address adoption gap

Town, founded by Jean-Denis Grès, former CTO of Plaid, has pivoted from an SMB tax automation business to a horizontal AI assistant for knowledge workers. Grès stated that the tax product generated slightly below $1 million in first-year revenue with projections of $2.5–3 million, but was abandoned because incumbents could automate tax prep faster and the annual sales cycle prevented venture-scale growth. The new product, launched after product-market fit was found between December and April, integrates with email and calendars to automate meeting prep, email drafting, and business workflows. Town targets SMBs and mid-market firms for product-led growth while building enterprise security features.

Town positions itself as a 'neutral' platform that routes to the best model provider, including OpenAI, Anthropic, and open-weight models, rather than being locked to one lab. Grès argues that businesses should own their data and context assets to ensure low switching costs, a stance he contrasts with the Forward Deployed Engineer approach used by labs like OpenAI, which he believes creates lock-in and scales poorly as models improve. He asserts that neutral AI assistants are strategically superior to lab-specific tools, which may not represent the best intelligence over the next 3-5 years.

Grès also identifies a significant AI adoption gap driven by the lack of 'tinkers' in non-tech sectors. AI adoption is concentrated among individuals with the time and skill to experiment with technology, while the broader workforce struggles to deploy AI effectively. Many users treat AI as a replacement for Google search rather than a tool for automation because the product experience is often just a text box. Grès argues that the primary barrier to AI diffusion is not intelligence but product experience. Companies must move beyond text-box interfaces to proactively demonstrate and execute value for non-technical users, allowing them to achieve the same results as tinkers without building custom integrations.