Four approaches can run a sales operation: a spreadsheet, a legacy CRM, a legacy CRM with AI bolted on, or an agentic CRM. They differ in one thing that decides everything else — whether the system is built for a person to type into or for an agent to work in.

Each of the first three is genuinely good at something. The point of this page is not that they are bad; it is where each one runs into a ceiling, and why the ceiling is where it is.

The spreadsheet

A spreadsheet is the fastest start there is. It is flexible, everyone already knows how to use one, and for a small pipeline it is often the right call — no setup, no cost, total control of the layout.

Its ceiling is structure. A spreadsheet does not know that a row is a deal, that a deal has a contact, that a contact belongs to a company. Nothing acts on its own. As the operation grows, the spreadsheet becomes a thing people maintain rather than a thing that works for them, and the moment two people share it, "which version is right" becomes a daily question.

The legacy CRM

A legacy CRM fixes exactly that. It gives records real structure, relationships, permissions, and a large ecosystem of tools built around it. This is mature, proven software, and for a great many teams it has been the right backbone for years.

Its ceiling is quieter, and it is built into the premise. A legacy CRM is a system for people to enter data into. Its records hold what a rep remembered to type after a call — which means the data is only as complete and as current as human discipline makes it, and human discipline under a full pipeline is thin and late. The structure is real; what fills it is not always.

Bolt-on AI

So AI arrives on top. An assistant beside the records that drafts an email, summarizes a thread, suggests a next step. On an investment your team already knows, that is a real convenience, and it is the path most of the market is taking.

Its ceiling is the one that matters most here, because it is invisible. The assistant sits on a model built for typing, so it inherits thin, late data — and an assistant reading incomplete records produces generic output, because it has nothing specific to work from. It can talk about the record. It cannot make the record better, because the record was never built for it to work in. No add-on lifts that ceiling; the ceiling is underneath the add-on.

The agentic CRM

An agentic CRM moves the same intelligence below the surface instead of on top of it. The data model is built for an agent to read and write directly, so the agent works the records the way a person would — and the records update as a side effect of that work rather than as a chore performed afterward. The context stays current because keeping it current is the work, not an extra step.

That is the whole difference between the last two. Bolt-on AI talks about the record. An agentic system changes it.

It is not free of trade-offs. It fits a sales motion that already exists — there has to be routine work before an agent can take it. And you shape the model to how your business actually runs, which is real work, even though an agent now does most of the assembling. Those conditions are covered on who Sonta AI is for.

The question that separates them

Strip away the detail and one question sorts all four: is the system something you type into, or something that gets worked?

A spreadsheet and a legacy CRM are typed into by people. A bolt-on assistant types into them faster. An agentic CRM is worked — by agents and people together, over the same records, with the model kept current by the doing. Everything else about Sonta follows from being on that side of the line.

Where to go next

For how that plays out in the product, the page on how Sonta AI fits together lays out the model end to end, and the principles behind Sonta AI covers the stance underneath it. To check the fit honestly, there is who Sonta AI is for.

How the Sonta approach differs