The facts: This week, Google, Microsoft, Salesforce, Snowflake and ServiceNow announced their support for a common infrastructure protocol for enterprise AI agents. What is at stake : the « plumbing » layer that determines how agents access data and business tools, and how they talk to each other. The implicit target is clear : Anthropic's Model Context Protocol (MCP), which in eighteen months has become the de facto standard for connecting agents to enterprise systems. In parallel, OpenAI is beefing up its on-the-ground offering : its consulting arm DeployCo, backed by more than 4 billion dollars, has just acquired Northslope, a firm of engineers who embed directly with clients to build their AI systems.

The Codito take: This announcement should be read for what it is : an admission that winning on models is no longer enough. When a connection standard takes hold, whoever controls it shapes the entire ecosystem built on top of it, much as USB or HTTP did in their day. Anthropic's MCP won that position quietly, tool by tool, integration by integration. The alliance of the five is a classic defensive move : offer an « open » alternative backed by a consortium, so that no single player owns the socket. For client companies, this war has one direct consequence : for a few quarters, two standards will coexist, and some tools will pick a side. Today's bad choices will become tomorrow's painful migrations.

What it means for you: An SME has no say in a standards war, but it lives with the consequences. The right stance is not to bet on the winner : it is to refuse lock-in. In practice, before signing for an AI tool or a custom build, ask your provider one simple question : « if I want to switch model provider or platform in two years, what happens ? ». If the answer is vague, or if it means rebuilding everything, you are not buying a tool : you are buying a dependency.

Number of the week Eighteen months. That is how long it took Anthropic's MCP to go from an obscure open specification to the de facto standard for connecting agents to data, to the point where five of the world's largest software vendors are teaming up to counter it. A useful reminder : in infrastructure, positions are won quickly and lost slowly. The architecture choices you make this year are the ones you will live with in 2029.

When two standards go head to head, large groups put their architects on watch and wait it out. An SME has neither the time nor the headcount for that. The good news : three simple habits are enough to get through this period without ending up on the wrong side of a forced migration.

1. Demand interoperability, not a promise about a standard

The question is not « does this tool support MCP or the alliance protocol ? » : both camps will evolve. The question is : is this tool built to talk to the outside world ? Documented APIs, complete exports, standard connectors. A tool that ticks those boxes will adapt to whichever protocol wins. A closed tool will lock you in with it, whatever side it picks.

2. Keep your data in formats that will outlive the tools

Protocols change, platforms merge, vendors disappear. What remains is your raw material : clients, quotes, history, documented know-how. If that material lives in open, exportable formats (SQL databases, structured files, standard documents), a migration is a project of a few weeks. If it lives only in a tool's proprietary format, a migration is a crisis. The simple test : export everything, today, and look at what you get. If the result is unusable, you now know where your real dependency lies.

3. Invest in your teams' autonomy, not just in the tool

This is the least obvious habit and the most profitable one. A team that understands what its AI system actually does (which data goes in, which rules apply, where the secrets live) can change it without drama : the provider becomes replaceable, the tool becomes replaceable, the knowledge stays. A team that uses a black box is hostage to that black box. The difference between the two is not decided by the choice of tool : it is decided by how the tool is deployed and handed over.

« You cannot choose the winning standard. You can choose to depend on none of them. »

This week's action : take your two or three main AI tools and put them through the three-habit test. Interoperability, exportability, autonomy. Every « no » is a quantifiable risk : the cost of getting out the day the market has made up its mind.

Book a strategy call

Discover Codito Support

To close this edition, the next instalment of the Gustave Rénovation story, and it illustrates precisely the third habit from Part 2.

A quick recap : in June, the first listening session with this high-end main contractor. In early July, delivery and demo of the Gustave Chantier 360 prototype. On Wednesday 15 July, session 2, and a change of register : this time, I am no longer the one building in front of Amaury. Amaury builds, and I guide.

What happened during the session

• A complete documentation kit was prepared beforehand. Business context, brand guidelines, data model, screen-by-screen specifications, source code of the approved prototype, and a series of ready-to-copy prompts, step by step. Everything the AI agent needs to know to build the right thing, and everything Amaury needs to understand to drive it.
• The working environment was set up live. Installing the development tools on Amaury's machine, initialising version control, securing access keys in environment files that are never shared. Amaury is not a developer : every new concept is explained the moment it comes up, with a simple analogy.
• Amaury now drives Claude Code himself. He launches the agent, gives it the steps one at a time, checks the result in his browser, approves, moves on to the next. He sets the pace. Our role shifts from builder to copilot : we step in on the blockers, not on every line.

This handover is not an educational bonus : it is the heart of the engagement. In a market where protocols are fighting it out and tools shift every quarter, the only thing that gives an SME lasting protection is a team that understands its own system. The day Gustave wants to change a component, a model or a provider, it will be a decision, not a crisis.

The thread running through this edition While the giants fight over the socket, an SME's best insurance is unchanged : understanding what it plugs into that socket. A self-sufficient client has nothing to fear from a standards war. That is exactly what Codito Support builds, month after month.