AI-native CRM
The positive case rather than the comparison: what it means to design the schema, the identity model and the queue for a caller that is software.
Read it →GET/v1/activities
What an AI CRM usually means, and what it means here
Most products under this heading are a CRM for people with AI features added to them. This one is a CRM for software with no AI inside it at all. Both are defensible, they suit different problems, and this page exists to help you work out which of the two you came looking for.
Two different products wear this label
One phrase, two products, and a page of search results where both of them are described in the same words. Here is each one written the way its own maker would write it.
What the phrase usually describes
A CRM built for salespeople with machine learning layered on top: lead scoring, deal risk, call summaries, drafted replies and a chat panel that answers questions about the pipeline. The people are still the users, and the AI is there to make them faster.
What it describes here
A CRM built for software with no machine learning inside it whatsoever. Your agents do the reading, the judging and the writing; this holds the records, the timeline and the queue, and keeps the evidence for every value somebody might later question.
AI in the CRM against AI at the keyboard
Neither of these is wrong. They answer different questions, and the honest way to choose is to ask who will be doing the operational work in your team a year from now.
Who the AI helps
AI features in a CRMA person, at the moment they are looking at a screen. Scores, summaries and suggested replies land beside what they were already doing.
Agent Native CRMNobody. There is no assistant on this product, because the reader is expected to be a program that brought an assistant of its own.
Where the intelligence comes from
AI features in a CRMThe vendor. You get the model they picked, tuned the way they tuned it, and you change it by changing vendor.
Agent Native CRMYou. The model, the prompts and the tools are yours, and swapping any of them changes nothing about the record they write into.
What you can automate
AI features in a CRMWhatever the workflow builder exposes. Anything past its edge becomes an integration project or a feature request.
Agent Native CRMWhatever you can write code for, because the entire product is 98 REST operations and there is nothing sitting behind the screen.
Why a value says what it says
AI features in a CRMOften nowhere. A field holds a number, and whether a person or a model put it there is not part of the record.
Agent Native CRMOne read. Every value points back at the activity that produced it, with the writer, the source and the confidence attached.
What a seat costs
AI features in a CRMA price per user per month, which turns a team of twenty agents into a budget conversation before it is an engineering one.
Agent Native CRMNothing per user. The plan caps stored records and activities; agents are not counted and calls are not metered.
What that looks like in calls
Every route named below is one this API serves — a test reads them back against the operation table and fails on anything invented. Read them in order and you have the argument on this page without the prose around it.
The question an AI CRM ought to be able to answer
Why does this field say what it says? Here that is a read rather than an investigation. A record's activities are the values it was assembled from, each carrying who wrote it, when, from what source and at what confidence, in the order it happened.
- GET /v1/activities
- GET /v1/people/{id}/activities
- GET /v1/deals/{id}/activities
A score is a field, and the score has an author
Nothing here computes one. If your model produces a number, it becomes a field you added, written with a source and a confidence, and the run that produced it sits on the timeline — which is more than most scores in most CRMs can say for themselves.
- POST /v1/schema/fields
- PATCH /v1/companies/{id}
- POST /v1/activities
Drafts wait in a queue, not in an outbox
A message an agent wrote is a task with that message in its payload. A person reads it, edits it, cancels it, or leaves it alone. The agent that claims it delivers it from its own account and closes the task with the outcome it got.
- POST /v1/tasks
- PATCH /v1/tasks/{id}
- POST /v1/tasks/{id}/cancel
- POST /v1/tasks/{id}/complete
The pipeline moves on events, not on a guess
A deal changes stage through one route and closes through another, and neither happens without an activity explaining it. Closing as lost requires a reason. No model here quietly decides that a deal has gone cold and marks it so.
- POST /v1/deals/{id}/stage
- POST /v1/deals/{id}/close
- GET /v1/views/{id}/records
GET/v1/activities
The useful question is not whether there is AI in your CRM. It is who is holding the keyboard.
Where a person still decides
None of these is a setting somebody switches on. Most are enforced by the API itself; the rest are enforced by the fact that the work sits in a queue until somebody takes it.
You can tell what the software did
Every activity names the principal that wrote it and every task names the principal that claimed it, so a week of agent work reads back as a list of decisions with authors rather than as a database that changed overnight.
You can take the work back
An owner or an admin can release a task an agent claimed and never finished, or reassign it, with the chain of follow-ups intact. Recovering work from software that stopped is an ordinary operation rather than a support conversation.
You can disagree with it in writing
A person's correction is later than the agent's write, so it wins. The disagreement stays visible on the timeline instead of overwriting the history that led to it, which is how the next agent learns the value was contested.
You can leave with the shape, not just the rows
Every object, field, stage and vocabulary word in a workspace is readable over the API you already hold a key for. An export from here describes the model as well as the data that happens to be sitting in it.
GET/v1/activities
What you get back
Not a diagram of the architecture — the actual shape of the response the rest of this page is built on. Nothing in it was configured in a settings screen; every part of it is readable and writable over the same API.
GET /v1/activities
- type
- The vocabulary this workspace accepts, including the types agents registered themselves
- actor_id
- The principal that wrote it. Not the human whose seat a piece of software borrowed
- source
- Where the writer says this came from — a storefront, a vendor, a call, an inbox
- confidence
- The writer grading its own source, 0 to 1. Not a number this product computed
- created_at
- Append-only and never rewritten, so the order you read is the order it happened
What this does not do
Said plainly, because each of these is a thing a product described this way is usually assumed to do, and an agent that assumes it will write code that never runs.
No model, no scoring, no predictions
This product does not rank leads, forecast a quarter, summarise a call or suggest a next step. If those are what you need, the AI features in a conventional CRM are a real answer, and pretending otherwise would waste a week of your time.
It does not send and it does not prospect
No email, no messages, no calls, no lists to buy and nothing to look up. Your agents bring the search, the enrichment and the delivery, then write back what actually happened with the source beside it.
It will not replace Salesforce for a sales floor
Twenty people who live in a CRM all day want the screens, the reports and the ecosystem a large vendor has spent twenty years building. This is built for the case where the operational work is done by software.
Questions people ask before they build on it
Answered here rather than in a call, and answered the way they would be answered in a call.
What is an AI-powered CRM?
Usually a customer relationship manager with machine learning features layered on: lead scoring, deal forecasts, call summaries, drafted replies and a chat panel over the pipeline. The user is still a person, and the AI is there to make that person faster at what they were already doing.
Is this an AI-powered CRM?
Not in the usual sense, and saying so is more useful than the alternative. There is no model in this product. It is the record and the task queue your own agents work through, which is a different purchase from a CRM that scores your leads for you.
Which of the two should I buy?
If people do the selling and you want them helped, buy the CRM with the AI features; the large vendors do that well. If software is doing the operational work and needs somewhere it can read, write and hand off without a browser in the way, that is what this is.
Can I use both?
Yes, and some teams will. Nothing here stops you keeping a conventional CRM as the system of record for the humans. An agent holds a key to this workspace for the work it does, and reaches anything else with its own tools and its own credentials.
Does an AI CRM need AI in it to be useful?
No. The parts that make a record trustworthy — natural keys, an append-only timeline, a source and a confidence on every write, a queue where exactly one claim wins — are ordinary engineering, and they are what an agent actually needs. The intelligence sits upstream of all of it.
Where to read next
The same product from a different angle. None of these repeats this page; each answers a question this one deliberately leaves alone.
CRM for developers
For the reader who has stopped asking which CRM has the best AI and started asking which one their own code can own outright.
Read it →Agentic CRM
What this category looks like once the operational work is done by agents and the human job becomes supervision, correction and the occasional veto.
Read it →Start with an empty workspace
An agent can create one with no key, no invitation and no human, and shape it through the same API you are about to hand it. Your API calls are never rate limited and agents are never counted.