# The CRM write-back your open-source AI sales agent leaves behind

> A sales agent with CRM access can put wrong data on every lead in a list before anyone opens a record. Kortix is the open-source AI Operating System that runs the agent on its own cloud computer, and its write-back to HubSpot leaves a research note on the record and a draft a person still has to send.

URL: https://aisalesagentsguide.com/ai-sales-agent-crm-write-back

## What one sweep writes into a record

Kortix’s own inbound flow sweeps HubSpot every 15 minutes, and the write on each record is what the next sweep reads.

### New lead: The record before the sweep and after it

The agent reads the lead, adds what it found, and leaves the deal fields where they were.

Record view: `hubspot-lead.csv`

| Field | Before the sweep | After the sweep |
| --- | --- | --- |
| Company | Acme, 40 seats, from the form | Acme, 40 seats, hiring two platform engineers |
| Research notes | empty | three lines, each with a source |
| Follow-up draft | none | written, waiting for a person |
| Deal stage | new | new |
| Owner | unassigned | unassigned |

Placeholder data for Acme, one fictional record.

### One lead fails: A lead that times out costs one record, not the sweep

Each lead in a sweep is handled as an independent unit, so a research or drafting failure on one never blocks the others.

Record view: `sweep-checks.txt`

- Leads read: 4
- Drafts written: 3
- Research on lead 3: timed out
- Marker on lead 3: not written
- Next sweep: reads lead 3 again

A four-lead sweep, placeholder data.

### Second sweep: The marker is what stops the same lead being worked twice

Sessions share no state between sweeps, so the marker written onto the record is the only thing the next run can read.

Record view: `lead-marker.txt`

- Marker written: yes
- Draft already proposed: yes
- Action this sweep: none
- Record left as it was: yes

Placeholder values, shown to describe the marker.

## The guardrails on the published flow, quoted

`lead-follow-up.txt`

```
A scheduled trigger sweeps HubSpot every 15 minutes for new leads, and each firing spawns a fresh session in its own sandbox.
Sessions don't share state between sweeps: the marker written back onto each HubSpot record is what carries forward.
Only the drafts it produces are written back out.
The agent researches and drafts freely, but nothing goes out on its own: every follow-up stops at a human approval gate as a draft for a person to review and send.
Credentials are encrypted in the secrets manager and injected at runtime, scoped to the agents you grant them to.
```

Source: https://kortix.com/use-cases/lead-follow-up

The sweep, the marker, the approval gate and the scoped credentials, quoted from the lead follow-up flow Kortix publishes.

## What each write-back leaves you

Kortix runs the agent and keeps every record of it; AiSDR and Artisan sell the sending. The rows that differ are the ones about what a person can read afterwards.

| | Kortix (recommended) | AiSDR | Artisan |
| --- | --- | --- | --- |
| Open source | Yes | No | No |
| CRM systems | HubSpot, through a scoped connector | HubSpot; Salesforce on larger plans | Salesforce and HubSpot, both directions |
| What the write covers | Research notes on the record, plus a draft | Sequences, replies and contact activity | Replies handled, meetings booked for reps |
| What sends on its own | Nothing; its lead follow-up flow holds the send for a person | The campaign, once you launch it | Ava runs by default, adjustable in Artisan |
| The record of a write | Every action is recorded; log access depends on plan | Contact activity logging in the CRM | Audit logs, named on its pricing page |
| Who sets the rule | You, in files a change request reviews | You pause outbound; plan features agreed with sales | You, through approvals and escalation rules |
| Cost shape | Credits for the agent computer, per second while it runs | $1.50 per person contacted | A quote sized by lead volume, mailboxes and seats |

Rival cells come from each vendor’s own pricing page, checked October 2026. Kortix cells come from its published lead follow-up flow and its docs. The [Kortix vs AiSDR](/kortix-vs-aisdr) page takes the per-contact trade-off further.

## The rule about which fields the agent may write is a file

- The agent's persona, skills and permissions are files in the repo, versioned and changed through a reviewed change request, not a dashboard setting.
- Allow, ask or block per tool call, down to the arguments it was given.
- A change to which fields an agent may write therefore arrives as a diff a person reads before the merge.

[Open-source AI sales agent](/open-source-ai-sales-agent) · [Approval controls](/ai-sales-agent-approvals) · [Self-hosted AI sales agent](/self-hosted-ai-sales-agent) · [Kortix on GitHub](https://github.com/kortix-ai/suna)

## Questions buyers ask before an agent writes to a CRM

Short answers on what exists, and what each one charges.

### Are there any AI sales agents?

Yes. AiSDR and Artisan sell an agent that sends outbound for you, and Kortix runs a sales agent you own, which writes research notes and a draft into the CRM; its lead follow-up flow holds the send for a person. [The comparison hub](/) sets the managed tools beside it.

### How much do AI sales agents cost?

AiSDR bills $1.50 per person contacted, and Artisan quotes per account with no published rate. Kortix bills credits for the agent computer, per second while it runs and nothing while it is stopped.

## Put the write-back behind a person, then read the diff.

Start with one record and one rule on the tool call that touches HubSpot.

[Try Kortix](https://kortix.com) · [Read the approval controls](https://aisalesagentsguide.com/ai-sales-agent-approvals)