Sora is Dead and Linear's CEO on the future of Agent-Driven Development
OpenAI shut down Sora after just months due to broken unit economics, with downloads falling 66% and compute costs reaching $15 million daily at peak. Linear's CEO argues AI agents are ending traditional issue tracking in favor of context-driven development workflows.
Summary
OpenAI officially discontinued Sora, its AI video generation platform, after a brief but expensive run that saw initial success followed by rapid decline. Despite hitting the top of iOS App Store charts and securing a $1 billion Disney partnership, downloads plummeted from 3.3 million in November to 1.1 million by February, generating only $2.1 million in total revenue while burning an estimated $10-15 million daily in compute costs. The flat subscription pricing model proved unsustainable for high-compute video generation, leading OpenAI to pivot toward a desktop super app strategy combining ChatGPT, Codex, and browser functionality. Meanwhile, Linear's CEO declared the end of traditional issue tracking, arguing that AI agents are replacing handoff-based development workflows with context-driven systems. Coinbase exemplified this shift by having engineers abandon code for two weeks to understand agent-driven development, resulting in continuous development cycles where agents work overnight and humans review in the morning. The discussion extends to broader implications including the war against AI-generated content across platforms like Wikipedia and Reddit, Meta's release of Tribe V2 for predicting brain responses to stimuli, and companies like Cisco saving millions by replacing SaaS tools with AI agents while Ramp demonstrates self-maintaining codebases through automated monitoring and bug fixes.
Key Insights
- Sora's flat subscription pricing model was fundamentally broken for video generation, with compute costs reaching $15 million per day at peak while generating only $2.1 million total revenue
- Linear CEO argues that issue tracking is finished because AI agents installed on 75% of enterprise workspaces are replacing handoff-based development with context-driven workflows
- Coinbase's head of engineering discovered that answering questions creates a hidden tax that compounds to slow companies down, leading to a two-week no-code experiment
- Meta's Tribe V2 model can predict human brain responses to stimuli using 1,000 hours of brain scan data, potentially revolutionizing how product teams test user experiences
- Cisco replaced presentation software with AI agents saving $5 million annually, with their VP asking which applications can become automated workflows that eliminate the need for apps entirely
Topics
Full transcript available for MurmurCast members
Sign Up to Access