If anyone follows Jason Haddix, you've probably heard about Perplexity offering Google 34.5B for Chrome.
This got my attention, so I dug further.
Here is what I learned:
Perplexity just lobbed a 34.5B offer at Google for Chrome. Whether this closes or not, the signal is loud: the browser is turning into the AI control plane for the internet.
New internet: tabs become tasks, pages become APIs, your agent negotiates with services on your behalf. Chrome, or any dominant browser, becomes the runtime for agents, not just HTML.
Protocols that matter:
• MCP, Model Context Protocol, gives models a standard way to plug into tools and data, think USB for AI. This is how agents get real capabilities without duct tape.
• A2A, Agent-to-Agent, lets agents talk to each other directly, coordinate, and transact across vendors. Interop kills lock-in, it also explodes the attack surface.
Infinite extensibility of data: with MCP-style connectors, anything you can read or write becomes callable, databases, calendars, POS, cameras. Context stops being a static prompt, it becomes a live graph that your agent can extend on the fly. Power goes up, governance better be ready.
Ultra-personalized ads: the unit of advertising shifts from a banner to an instruction. Agent-to-agent bidding for your next action, not your next click. Your agent will broker your attention, price it, and enforce preferences, or you will get steamrolled by dark-pattern agent scripts.
Security, the part most folks will ignore until it bites:
• Supply chain, every MCP server is a new dependency. Sign everything, sandbox hard, monitor egress.
• Prompt injection moves into the DOM, the browser becomes the blast radius.
• Memory theft, if your agent keeps long-term memory, that store is the new crown jewels. Local first, encrypted by default, consent receipts, easy revocation.
Open questions for you all:
Should a browser that controls distribution for search and agents live outside a search giant, or is this possible sale a cure worse than the disease?
What baseline rights should a personal agent enforce, ad-free mode, default deny on data sharing, verifiable consent, rate limits on agent-to-agent requests?
If MCP and A2A win, what is the minimal security standard, signed connectors, capability-scoped keys, per-task sandboxes, audit logs by default?
If Chrome became AI-native tomorrow, what would you build first, and what would you block by policy on day one?