Dharmesh Shah

Overview

Dharmesh Shah is the co-founder and CTO of HubSpot, and in this knowledge base he appears as a prominent voice on how AI products should evolve from human-only software into systems built for human-agent collaboration. Across recent mentions, he is associated with launching HubSpot's private-beta Agent CLI, expanding HubSpot's agentic product surface, and advocating that the real product advantage in AI often comes not just from the model, but from the surrounding harness: tools, memory, context, workflows, and integrations.

For AI Product Managers, Shah matters because his commentary consistently connects frontier model capabilities to product design choices. He emphasizes agent-ready interfaces, discoverable APIs and CLIs, deep proprietary context, closed-loop data systems, and operational models that help customers realize value faster. His perspective is especially relevant for PMs building AI-native SaaS, developer tools, and enterprise platforms that must support both human users and software agents.

Key Developments

• 2026-04-10: Launched jsondata.com, a free AI-powered tool for viewing, filtering, compressing, and manipulating JSON data in a nested interface.
• 2026-04-22: Proposed user-defined AI prompt files like MESSAGES.md and INVITES.md for LinkedIn to automatically classify and handle messages and invites.
• 2026-04-22: Argued that closed-loop systems that feed deal and outcome data back into AI are more valuable than isolated success metrics because they improve future performance.
• 2026-04-27: Shipped a major update to HubCode, HubSpot's agentic coding tool for building HubSpot apps, after encountering timeout constraints in AI-driven endpoint calls.
• 2026-05-01: Suggested reframing FDEs as Forward Deployed Experts, expanding the concept beyond engineers to domain specialists such as lawyers, consultants, and teachers who can accelerate customer value.
• 2026-05-03: Argued that durable differentiation is harder to sustain with only a frontier model plus harness, and that deeper long-term advantage comes from accumulated proprietary data and context.
• 2026-05-03: Expressed skepticism that the industry is necessarily heading toward an AI "mageddon," signaling a more measured view of competitive collapse.
• 2026-05-18: Argued that legacy APIs, MCPs, and CLIs were designed for human developers, and now need to become more discoverable, legible, and forgiving for AI agents as primary users.
• 2026-05-18: Highlighted the importance of agent readiness, arguing that software must support both excellent human UX and robust agentic experience (AX).
• 2026-05-19: Said HubSpot is launching an agentic experience (AX) that enables AI agents to configure the platform, create dashboards, and manage CRM workflows directly rather than relying on a human UX layer.
• 2026-05-20: Shared that Andrej Karpathy joined Anthropic to use Claude to accelerate AI research, underscoring the leverage created by AI-powered research loops.
• 2026-05-25: Argued that the real usability layer in AI products is the harness—platforms like ChatGPT or Claude Cowork that provide tools, memory, skills, and context around the base model.
• 2026-05-28: Launched HubSpot's private-beta Agent CLI, a next-generation command-line tool for agentic workflows, and argued that the future of software lies in collaboration between humans and AI agents.

Relevance to AI PMs

1. Design for agents, not just humans. Shah's push for agentic experience, agent-ready APIs, and agent-friendly CLIs is a practical reminder that PMs should define product requirements for machine users as well as human users. That means better discoverability, structured outputs, forgiving interfaces, and workflows that agents can execute reliably.

2. Build moat through context and feedback loops. His emphasis on proprietary data, accumulated context, and closed-loop systems is highly tactical for PMs deciding where defensibility comes from. Instead of over-focusing on model selection, PMs should prioritize systems that capture customer context, usage history, outcomes, and feedback to improve future agent performance.

3. Treat the harness as product strategy. Shah repeatedly points to tools, memory, skills, and orchestration as what turns a capable model into a usable product. For PMs, this means roadmap decisions should cover not only model quality, but also context management, tool access, workflow control, onboarding, observability, and safety.

Related

• HubSpot: Shah's primary company and the platform where many of these agentic ideas are being operationalized through Agent CLI, HubCode, CRM automation, and agentic experience.
• Agent CLI: A private-beta HubSpot command-line tool tied closely to Shah's thesis that humans and AI agents will collaborate in software workflows.
• HubCode: HubSpot's agentic coding tool for app development, illustrating Shah's interest in developer productivity and agent-native tooling.
• ChatGPT and Claude / Claude Cowork: Referenced by Shah as examples of the harness layer that makes foundation models usable through tools, memory, and context.
• Anthropic and OpenAI: Connected through Shah's commentary on frontier models, product harnesses, and the broader competitive landscape.
• MCP, APIs, and CLIs: Central to Shah's argument that product interfaces need redesign for a world where agents increasingly act as first-class users.
• jsondata.com: An example of Shah shipping practical AI-powered tooling outside HubSpot.
• Andrej Karpathy, Sam Altman, Reid Hoffman, Guillermo Rauch: Related ecosystem figures whose work intersects with Shah's interests in AI platforms, agentic software, and product strategy.