Keep pulling the thread on Gokul Rajaram.
The concept of long-running AI agents has fundamentally changed product development within the last few months.
AI startups focused on simple applications are at high risk of being displaced by foundation model providers like Google and OpenAI, which offer their own agent-building tools like Gemini agent builder and ChatGPT Enterprise.
System-of-record companies are beginning to cut off API access to AI agent companies, as exemplified by Slack (owned by Salesforce) blocking Glean's access to its data.
Incumbent system-of-record companies are defending against AI agent startups by blocking API access, offering their own bundled agents for free, or charging prohibitive data access fees, such as $2 per API call.
The strategy of building an AI "system of action" on top of an incumbent's "system of record" is no longer viable; AI companies must now build the entire platform themselves.
Software companies with utility-based, per-seat pricing models, like Zendesk, are the most vulnerable to AI disruption as customers can replace human seats with AI agents.
Square's core business model innovation was to approve 95% of small business applicants and shift risk management from upfront underwriting to real-time, transaction-level analysis using machine learning.
ChatGPT possesses a unique advertising advantage by combining high-intent user data, similar to Google, with rich identity data, similar to Facebook.
The primary threat to dominant ad networks like Google and Facebook is a shift in consumer behavior toward using third-party AI agents, which would disintermediate the platforms from their users.
Product leaders are concerned that engineers using AI will generate a large volume of low-quality code, a phenomenon referred to as "AI slop".
The emergence of AI-powered developer tools like Claude Code and Claude Cowork has led to a significant increase in the ability of both technical and non-technical individuals to build software.
Due to the rapid advancement of AI model capabilities, traditional, prescriptive product development methods are becoming ineffective.