Everyone is talking about AI agents. Very few are thinking about what those agents will be fed.
The Signal Most People Underestimated
The news sounds technical: Google releases GWS, a CLI that allows scripts and AI agents to interact with Drive, Gmail, Calendar, and all Workspace services through structured commands and JSON responses. No more manual REST calls, no more fragile curl scripts to maintain.
In reality, it is the final piece of a much larger design. And to understand it, it is worth taking a step back.
Google’s Strategic Blueprint: One Year to Build the Agentic Stack
Over the past twelve months, Google has released a series of protocols and tools that, taken individually, look like technical updates. Taken together, they reveal a precise strategy: to become the reference infrastructure for the agentic world.
April 2025 — Agent2Agent Protocol (A2A). Google launches an open standard for AI agents to communicate with each other, regardless of framework or vendor. Not a product: a networking layer for the agentic world. At launch, over 50 partners — from Salesforce to SAP, from Atlassian to McKinsey. Today, more than 100.
December 2025 — Managed MCP Servers. Google releases its own MCP servers for Maps, BigQuery, Compute Engine, and other Cloud services, with security managed through IAM and Model Armor. The stated goal: make Google “agent-ready by design”. And with Apigee, any existing enterprise API can become a tool accessible to agents — with existing governance controls already in place.
February 2026 — WebMCP. The Chrome team, in collaboration with Microsoft and incubated within the W3C, launches a browser protocol that allows any website to expose structured actions directly to AI agents. Instead of having an agent try to figure out how to use a website by reading the DOM and guessing where to click, the website itself explicitly declares what it can do and how.
Today — GWS CLI for Workspace. The tools people use every day enter the agentic ecosystem with a stable, structured interface accessible to any framework.
This is the same role that HTTP and TCP/IP played in the 1990s. Whoever defines the protocol defines the rules of the game.
What This Means for Companies: The Window Is Closing
Gartner estimates that 40% of enterprise applications will have integrated AI agents by the end of 2026, up from less than 5% in 2025. The infrastructure to connect these agents to enterprise systems is consolidating around a few defined layers — and Google is building a significant portion of them.
For those working in the industry, this opens three concrete implications:
- For developers and system integrators: every new standard released is a window of opportunity. Those who implement WebMCP today or integrate A2A into their architectures have months of advantage over those who wait for the technology to stabilize.
- For CTOs and IT leaders: architectural decisions made today will carry weight three years from now. Adopting open, interoperable protocols means not being locked into a single vendor when the market consolidates.
- For product managers and entrepreneurs: “works with AI agents” will become a competitive feature, not a curiosity. Products that implement these standards early will appear in agentic workflows by default.
But there is one dimension of this shift that risks going unnoticed — and it is the most important one.
The Infrastructure Paradox: When Technology Becomes Commodity, Data Becomes Strategic
Let us reason by analogy.
When cloud computing became commodity, the companies that won were not those with the best cloud. They were those who knew what to do with it. When CRMs became widespread, the winner was not whoever had Salesforce. The winner was whoever had the richest, cleanest, most contextualised customer data.
The same is about to happen with AI agents.
Protocols can be copied. Models can be rented. Integrations can be built in weeks. But a company’s historical data cannot be bought or replicated.
Two companies. Same industry. Same agents. Same protocols. Same model. One has years of historical data on its operations, customers, and processes — structured, maintained, accessible. The other has the same technological tools, but data scattered across silos that do not talk to each other.
The result will not be slightly different. It will be radically different.
Because an AI agent is only as powerful as the context it has access to. Without high-quality proprietary data, even the most sophisticated agent works in the dark — and amplifies chaos instead of reducing it.
The Real Competitive Stronghold: What to Do Now
This is not a call to slow down AI adoption. It is exactly the opposite.
Now is the time to work in parallel on two fronts:
- Technology front: explore the new protocols, train your teams, build the first integrations. The agentic infrastructure is ready — those who wait fall behind.
- Data front: invest in governance, quality, and accessibility of enterprise data. Not as a separate project, but as an enabling condition for everything else.
Those who arrive at the agentic era with well-governed, accessible, and contextualised data will start from a position of structural advantage. Those who arrive with chaotic data will feed their agents exactly that chaos — automated and scaled.
A company’s historical data — how it has served its customers, the anomalies it has learned to recognise, the decisions it has made and their outcomes — is the product of years of operations. It is the memory of the organisation. It is the only asset that competitors cannot replicate, even with the same technology stack.
Conclusion
Google is building the agentic infrastructure. AI models are becoming more powerful and less expensive. Protocols are being standardised. Access to technology is being democratised.
In this scenario, artificial intelligence does not create value from nothing. It extracts it from what already exists.
The question is not “do you have AI agents?”. The question is “what will you feed them?”
The answer to that question is already inside your company. The challenge is making it accessible.
PROSSIMI PASSI
Vuoi trasformare i tuoi dati dormienti in intelligence attiva?
