Why the agentic era needs a business operating system

Most businesses run on a stack of tools. A CRM for customers, an invoicing tool for money, a spreadsheet for stock, a shared inbox, a chat. Each one does its job well, and each one was designed around the same picture: a person sits in front of a screen, types, and clicks.

In that picture, people are the glue. They carry what they learned in one tool to the next, in their heads. They know that the customer who just called is the one with the overdue invoice, and that the quote was promised a discount.

Now a new kind of user has arrived: the AI agent. An agent is an assistant that does not only answer questions. It looks things up, drafts the email, updates the deal, follows up next week, and asks you when it is unsure. People call this the agentic era. It is early, and nobody knows exactly how far it will go. But one thing is already clear: an agent has no head to carry things in. It can only work with what it can reach.

What is changing

From recording to doing. Business software used to store what happened while people did the work. Now more of the routine work, such as looking things up, drafting, updating and following up, can be handed to an agent. So the software has to be somewhere an agent can work, and not only somewhere a person types.

From one tool to the whole business. An agent is most useful when it can see how things connect: the customer, the quote, the invoice, the payment, the stock. The questions that matter, like which customers are overdue, and what did we promise them?, cross the borders between tools.

From trusting people to trusting a process. When a colleague changes a record, you trust their judgment. When an agent does, you want to know what it was allowed to do, who it acted for, and what exactly it changed.

Where a stack of separate tools falls short

Each tool is good at what it was made for. The gaps appear when an agent joins the team. These are tendencies, not a verdict on any one product.

Each assistant sees a slice. In many products, AI arrives as a feature inside that product’s screens: a chat box, a button that drafts a reply. The one in your CRM sees customers, the one in your invoicing tool sees invoices, and none of them sees the business.

The model is theirs, not yours. A tool comes with its own idea of what a lead, a contact or a deal is, and lets you add a few fields. An agent has to work in your terms: your stages, your products, your promises. A fixed shape makes it guess, or makes you bend the business to fit the tool.

Control was designed for people. Permissions usually answer one question: which user can see or edit this? An agent raises more. What may it do, and as whom? What must wait for a yes? Is there a clear record of everything it changed? Without those, you either trust it too much or switch it off.

Your data and your costs live elsewhere. Many tools run in the vendor’s cloud and charge per user. An agent touches more of your data, more often, and does not fit a way of counting that assumes one person per seat. Where your data lives, who sees it, and what you pay as the work grows become real questions.

You wait for the roadmap. This area moves fast. If a tool lacks something you need, you wait for the vendor to build it, or you work around it.

What it takes: an operating system for the business

Think about what an operating system does for a computer. It gives every program the same files to work with, runs work in the background, decides who may do what, and lets you add programs without starting over.

A business needs the same thing. One place where the records of the whole business live, so the customer, the quote, the invoice and the stock are linked, not copied between tools. A place where the work runs on those records, whether a person or an agent starts it. Limits that say who, or what, may do which things. Connections to the services you already use. And room to add what is missing. The CRM, the invoicing and the stock are then not separate worlds. They are programs on one system, sharing what the business knows.

What to look for

Whatever you use, a few questions tell you whether it is ready for agents:

• Can the assistant see the whole business, or one tool’s slice of it?
• Does it work in your words and your steps, or the vendor’s?
• Does it ask before it changes anything, and can you see exactly what it did and for whom?
• Do you decide where your data lives and which AI provider is used?
• If something is missing, can you add it yourself?

If the answer to most of these is no, an assistant will feel like a gimmick on top of the product, however good its demo is.

Devias is built as that operating system, and to answer yes to all five. We explain how in How Devias is built for the agentic era.