AI agents are moving in front of software, sketching out an internet built for agents

A series of product launches and executive comments over the past two months are being read as signs of a deeper shift in how people may use the internet. SpaceX released Grok Bot on Aug. 11 as an always-on AI coworker running on a cloud computer. Meta followed on Sept. 8 with Muse, a personal agent that can browse the web, fill forms and negotiate on a user’s behalf. According to the article, Muse hit No. 1 on the U.S. App Store within 10 days and coincided with a selloff in online travel stocks including Expedia, Airbnb and Booking. Anthropic then released Opus 5.5 on Sept. 22, while OpenAI cut the price of two new GPT-6 versions to half that of the prior generation. Around the same time, Salesforce executive Patrick Stokes said AI will break apart software interfaces and replace them. The article argues that these events point to the same structural change: agents are starting to interact with software and internet services for users. That shift, in the author’s view, is producing a “steamroller effect” for startups as frontier model advances erase narrow AI businesses, while also pushing software creation toward a more fragmented, workshop-like model where more people can build tools for themselves but fewer companies can sell broadly reusable software. In that setup, products may need to be designed less for humans and more for agents, with clean APIs, readable terms and structured data taking priority. The piece also says blockchain-based payments, stablecoins and tokenized incentive systems are likely to become core infrastructure for machine-to-machine transactions in an agent-driven web.