Where Startups Can Win Beyond OpenAI and Anthropic | HubSpot’s Dharmesh Shah

HubSpot Strategic Positioning in the AI Era

• Chad GPT and Cloud Cowork are characterized as the "new OS" or "new browsers," effectively serving as the interface layer while value resides in the underlying work and data.
• The core thesis is that substantial software opportunities exist "outside the path" frontier AI labs are racing down, avoiding direct competition with their infrastructure models.
• HubSpot is restructuring its 20-year-old architecture to accommodate a world where AI agents navigate unstructured data rather than just humans reading structured relational databases.
• Dharmesh Shah distinguishes between "structured data stores" (for known facts) and a "system of context" (unstructured data/graphs), advocating for a dual-store architecture to support agentic workflows.

Architectural & Product Evolution

• HubSpot moved from a suite of marketing apps to an all-in-one CRM platform to solve customer intimidation regarding software fragmentation; this "all-in-one" philosophy remains central.
• The CRM was launched not as a Day 0 plan but as a strategic response to a survey showing 60% of HubSpot's customers had no CRM, driven by the belief that existing tools were too expensive and complex.
• The company is rebuilding its platform around "agents," moving from deterministic queries for human navigation to probabilistic inference for AI agents.
• A new framework is being designed for the next 20 years, utilizing a "system of context" (potentially a graph structure) alongside structured records to enable AI to drive outcomes.

Innovation & "HubSpot Next" Strategy

• HubSpot established "HubSpot Next," an innovation lab operating as a "startup within a startup" with full freedom to experiment outside core stack and branding rules.
• A specific project launched via this arm is "uSpot," a one-person CRM for solopreneurs, released 30 days prior to the interview.
• Successful experiments in the innovation lab (e.g., Agent.ai) can be absorbed into the main product via an internal acquisition-like process, allowing the core team to pick and choose features.
• The innovation lab allows for "cheaper learning" by isolating risk and testing new markets or technologies before integrating them into the main platform.

Market Data & Economic Trends

• U.S. new company formation is projected to reach a record 5.6–6 million in 2025, with 4 million already formed by August of the current year.
• Over 70% of these new companies are expected to remain one-person entities, representing a massive, underserved market for products like uSpot.
• HubSpot successfully launched uSpot with an early monetization strategy using a consumption-based model, testing user willingness to pay immediately.
• Dharmesh Shah counters the "SaaS apocalypse" narrative, asserting that the total addressable market for software is larger now due to AI solving previously unsolvable problems, particularly in marketing.

Competitive Landscape & Builder Advice

• Frontiers AI labs face "incentive constraints" that make it unlikely they will enter specific vertical markets like CRM, despite having the capability, because their business model relies on selling inference/tokens rather than vertical SaaS.
• Dharmesh advises entrepreneurs against building products that can be dismissed by saying "yet" to a frontier model's capabilities; instead, they should target areas with structural or incentive barriers for incumbents.
• AI models are trending toward commoditization and fungibility (similar to AWS in 2005), reducing the risk of being dependent on a single provider for core ingredients.
• A key architectural rule is to use deterministic queries for deterministic questions rather than inferring answers via LLMs; inference should be reserved for unstructured or complex cognitive tasks.
• Developers should focus on "Time to Joy" for the ecosystem, compressing the loop from API integration to making the first dollar for partners.

Future Outlook & Philosophy

• The "Yellow Brick Road" metaphor warns against blindly following the shiny path of frontier AI development where value may be diluted by the sheer volume of traffic and infrastructure costs.
• Dharmesh notes that AI is "just software" and a new energy source, allowing builders to iterate 10x faster and test ideas that previously took years in months.
• HubSpot is utilizing AI to simulate "alternate universes" of code, allowing the team to refactor legacy architecture (e.g., testing a Rust rewrite) by running large-scale evals on new models to measure trade-offs in latency and quality.
• The "best time to build" is currently, characterized by high agency, low barriers to entry due to abstraction layers, and an explosion of new business formations.
• Dharmesh states the best day at HubSpot occurred "about a week ago" when the strategic vision for the AI transition began to coalesce.