Your Tools Learn to Talk

Google handed its agent protocol to the same neutral foundation that houses Anthropic's. Gemini moved into Google Chat as a command line for your workday. And Perplexity tripled its revenue selling an agent that drives a computer for you. This was the week AI stopped being a destination and started becoming plumbing.

The Bottom Line

For two years every AI tool you bought was an island — its own login, its own connectors, its own dead end. This week the two protocols that let agents talk to your data and to each other moved under one neutral roof, an assistant landed inside the chat app your team already pays for, and buyers pushed one agent company past $750 million a year. The bottleneck moved from can it work to will you let it. So tell me in the comments: would you hand an AI agent the keyboard on your own machine — or does a human keep the mouse?

See what we build →

Top 3 Stories This Week

01 · Linux Foundation · A2A + MCP · Agent Standards

Your AI Tools Just Agreed to Speak the Same Language

Every business owner who has paid for two AI tools knows the tax: they don't talk. On August 20, that got materially better. Google moved Agent2Agent (A2A) — its protocol for how one AI agent asks another to do a job — into the Agentic AI Foundation, the neutral body the Linux Foundation launched in December 2025. A2A now sits beside Anthropic's Model Context Protocol (MCP), the standard for how an agent reaches your database, your API, your files. Two halves of the same wiring, one governing body, no single vendor holding the pen. In under a year the foundation grew from fewer than 40 members to more than 250, with AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI all paying in at the top tier. Read that member list again. Those companies agree on almost nothing — except that agents walled inside one vendor's garden are worth less than agents that roam. For you, that means the tool you buy in 2027 is far less likely to hold your data hostage.

Microservice Opportunity: Stop building integrations twice. Wire your systems — CRM, invoicing, scheduling, inventory — to an MCP server once, and every agent you adopt afterward plugs into that same socket instead of demanding a fresh custom build. A single MCP layer over your three core systems takes roughly two weeks. The payoff arrives the day you swap vendors and change one line instead of rebuilding six connections.

02 · Google Workspace · Ask Gemini · Productivity

Gemini Moved Into the App Your Team Already Argues In

Most AI tools fail for the same dull reason: nobody opens a second tab. Google just removed the tab. Starting August 26, Ask Gemini replaces the old side panel in Google Chat and turns the message box into a command line for the whole Workspace. From inside a thread, your team can search Gmail, Drive and Calendar, catch up on conversations they slept through, draft the update nobody wants to write, schedule the meeting, and file the task — without leaving the chat. It rolls out to Business Standard and Plus, Enterprise Standard and Plus, and Google AI Pro for Education, with up to 15 days before it shows up for everyone. Two things to tell your team before they file a ticket: the old Gemini side panel in Chat disappears with this launch, and the conversation history inside it does not migrate. Then the part with a clock on it — Google is handing out promotional higher usage limits through October 1. After that, caps apply. If your company runs on Workspace, you have five weeks of cheap room to learn what your team actually asks for.

Microservice Opportunity: Ask Gemini knows Workspace. It does not know your job-costing system, your point of sale, or the scheduling tool that runs your crews. Build the bridge: a microservice — a small AI tool that does one specific job — that exposes your operational data to Chat, so "what did we quote the Rivera job?" gets answered in the thread instead of in a phone call. Use the free window to log which questions your team asks. Those logs are your build list.

03 · Perplexity · Perplexity Computer · Agent Economy

An Agent That Drives Your Computer Just Tripled a Company's Revenue

Talk is cheap. Invoices are not. On August 23, The Information reported that Nvidia is in talks to invest in Perplexity at a valuation above $30 billion — more than a 50% jump from the roughly $20 billion it carried a year ago. The number underneath is the one worth studying: Perplexity's annualized revenue has climbed past $750 million, up from under $250 million at the start of 2026. Roughly triple, in eight months. A meaningful share of that growth traces to Perplexity Computer, a cloud-based agent that operates a computer on a professional's behalf — clicking, filling, pulling, filing. Not chatting about work. Doing it. Companies are now writing real checks for software that takes the mouse. That is a different purchase than a chatbot, and it carries a different risk: an agent with your credentials can act faster than you can object.

Microservice Opportunity: Do not buy a general-purpose agent and point it at your business. Scope it. Pick one repetitive screen-driving task your staff hates — pulling remittance details out of a bank portal, re-keying orders from a supplier site, chasing certificates of insurance — and build an agent that does only that, with read-only credentials and a human approving anything that moves money. Narrow agents ship in days and fail in ways you can see.

This week in numbers: 250+ — Members inside the Agentic AI Foundation — up from under 40 a year ago. MCP and A2A both live there. · Oct 1 — Free higher limits on Ask Gemini in Google Chat expire. Usage caps follow. · 3x — Perplexity's annualized revenue growth in eight months — under $250M to over $750M.

Sources

TL;DR — Key Takeaways

Top 3 Stories This Week

Your AI Tools Just Agreed to Speak the Same Language

Gemini Moved Into the App Your Team Already Argues In

An Agent That Drives Your Computer Just Tripled a Company's Revenue

Related answers