What an AI-native CRM is, and four things to check
An AI-native CRM is one where software writes most of the record: it reads email and meetings and updates fields. The label says nothing about whether you can trust the result. Four checks do: where a value came from, who approved what was sent, what the agent may not do, and how a change is undone.
Updated · 3 min read · Fonsent
In early access today. The AI model isn’t switched on yet. Field updates come from built-in rules, which pick up only sentences that state a value outright. An approved email is held in an outbox and isn’t delivered yet. Connecting a mailbox or a calendar isn’t open yet; import email, calendar files and transcripts as files. What works today
The phrase describes an architecture, not a result. Software that writes to your records can keep them current without anyone typing. It can also be wrong faster than a person ever was, in the same confident formatting. So judge one the way you would judge a new colleague with write access: by what you can check afterwards.
Four things to check
Run these in any product you are considering, Fonsent included. Each takes a few minutes with your own records.
- Open a value the software wrote and ask where it came from. Look for a sentence, not a summary.In Fonsent. Every value Fonsent writes from an email, a meeting or a transcript is stored with the exact sentence it came from. The database checks the quote against the source and refuses a value that has none.
- Find the last email that went out and ask who approved it, and what they saw when they did.In Fonsent. Nothing customer-facing leaves unless a person approved that exact message, and it leaves at most once.
- Ask what the agent may not do, however it is asked.In Fonsent. An agent can’t approve a draft. Only a person can.
- Change several records at once, then undo it.In Fonsent. Undo restores the earlier value only if nothing has edited it since. A change across several records is undone in one step, all or nothing.
How to read the answers
Listen for where the rule lives. “The model is told to cite its source” is a hope. “The database refuses a value without a quote” is a rule. The first holds until a prompt changes or a model has a bad day. The second holds whoever is writing, because the check sits where the value is stored.
The same test sorts the other three. An approval that can be given by the thing that wrote the draft is a formality. A list of what an agent may do alone is only as good as what happens at its edge. And an undo that restores an old value over newer work is a second mistake, not a way back.
What an agent does alone
Ask for the line in plain words, and expect it to be short enough to remember.
This is Fonsent’s. Fonsent reads freely and makes small, reversible changes on its own, each recorded with an Undo. Deleting, merging, bulk stage changes and new automations are drafted for a person. Email is always a draft.
Where the checks stop
A check tells you what a product does with what it was given. It can’t make a source say more than it says.
No quote, no change. If the source doesn’t state the value, Fonsent proposes nothing.
A source is matched to its contacts, its account and the account’s one open opportunity. If an account has several open opportunities the source is linked to none of them, and Fonsent says so. Nothing is guessed.
Each of the four has a page of its own, with the rule the database holds: see the product.