How we build applications is changing at an ever increasing speed. As frontier models get more capable, they dictate the requirements for the infra and tooling the next generation of apps will need. With reasoning and test-time compute surging, the tools these models can reliably call inside their reasoning loops are hardening to become the next-gen infrastructure layer.
We saw this in the first wave of agents (o3 and others) that embedded web search into their loops. 2025 brought a spike in search-API consumption, lifting mature providers like Brave and You.com and agent-focused startups like Exa to fresh revenue numbers and bringing to life a whole ecosystem of deep research agents.
Now a second wave (ChatGPT Agent, Perplexity Comet, and others) treats full browser use as a default step in the loop. The “browser wars” are back in Silicon Valley’s vocabulary, as we wait for the next move from OpenAI, Anthropic, and Google and companies like Browserbase, Browseruse, and E2B are benefiting from the shift to virtualized browsing.
If search was the first wave and browser use is the current wave, what will be the next wave?