All insights
AI Product Building Future of AI AI Agents

Agents running the platform vs. agents on the platform — the operator shift changes what software must be

The shift from agents as features inside your product to agents as operators running your product — passengers become pilots

@dharmesh (Dharmesh Shah, HubSpot CTO) · · 3 connections

Dharmesh Shah’s distinction from his internal HubSpot Slack message: “Being agentic is not just about agents running on our platform, it’s about agents running our platform (being able to operate it). That’s how you take AI from being a simple tool to a savvy teammate.”

The ON → RUNNING shift has three implications:

  1. API completeness: agents ON a platform need read access. Agents RUNNING a platform need full CRUD plus workflow orchestration — every operation a human operator performs must be programmatically available.

  2. Decision delegation: agents ON a platform execute predefined tasks. Agents RUNNING a platform make operational decisions — routing, prioritizing, escalating. The platform must expose decision points, not just data endpoints.

  3. Trust infrastructure: agents ON a platform need authentication. Agents RUNNING a platform need authorization boundaries, audit trails, and reversibility mechanisms — the same governance humans get.

This is the trajectory from Autopilots capture the work budget — six dollars in services for every one in software made concrete: the tool budget is “agents on the platform.” The work budget is “agents running the platform.” HubSpot is explicitly targeting the latter.

Connects to Self-disruption follows the value chain downward — software companies must eat their own agent layer before someone else does — building agent-operable platforms means disrupting your own human-facing product before someone else does.