Software · Pillar guide
CRM AI Agents: What They Do, Native Options by Platform, and When to Build Custom
Written scope and quote after the call. NDA on request. You own the code.
Quick answer
CRM AI agents read and update records inside HubSpot, monday CRM, Pipedrive or GoHighLevel to do one job: log activity, clean data, review the pipeline, answer questions or draft follow-ups. Every one of the five vendors now ships some native AI, billed in credits or per sub-account. This hub compares the native agents from vendor docs checked Oct 31, 2026, explains when a custom agent is worth building, how our in-house engineers build one, the guardrails we put on it and what it costs to run.
Builds like this
Key takeaways
- A CRM AI agent is software that reads records, decides on a next step and writes back to the CRM (or drafts the write for a person to approve). The useful ones do one narrow job, such as logging calls, merging duplicates or flagging stalled deals.
- All four CRMs we work on ship native AI. HubSpot bills Breeze agents in HubSpot Credits at $0.010 each (Customer agent 50 credits per resolution, Prospecting agent 100 per lead, Data agent 10 per response for one record); monday lists its pre-built CRM agents for Pro and above and meters them in AI credits (a one-time 6,000 trial, 12,000 on Ultimate, then an add-on); Airtable gives 15,000 AI credits per billable collaborator a month on Team; GoHighLevel's AI Employee is $50 or $97 a month per sub-account (vendor pages, checked Oct 31, 2026).
- Pipedrive is the outlier: it has AI assistants (report creation, email writing, a Sales Assistant in beta, Nova meeting intelligence, and Pipedrive MCP for ChatGPT or Claude) but no agent that runs on its own, so a Pipedrive agent that acts on records unprompted is usually a custom build on the API.
- Build custom when the agent needs data from outside the CRM, has to follow business rules the native builder cannot express, must run on a plan that lacks the native feature, or would cost more in per-action credits than a custom build at your volume.
- Every agent we ship has a named owner, least-privilege API access, a log of every read and write, human approval on anything customer-facing or destructive, and throttling inside the CRM's API limits.
A CRM AI agent is a program that works on your CRM records with a language model in the loop: it notices a trigger, reads the deals, contacts or activities it needs, and writes a change back or drafts one for a person to approve. The agents that earn their keep do one narrow job, such as logging a call summary to the right deal, merging duplicate contacts, or telling a sales manager on Monday morning which deals have gone quiet. A typical custom build is one agent, one CRM, one job, with clear rules for what it may change.
This page is the hub for aibrevo’s CRM AI agent work on the four CRMs we implement: HubSpot, monday CRM, Pipedrive and GoHighLevel, with Airtable’s native AI compared alongside. It compares what each vendor ships natively, explains when a custom agent is worth the build, and links to a page for each platform and each agent job.
Sources: HubSpot Credits, monday.com support AI Credits, Airtable support AI billing and HighLevel AI product pricing, all checked Oct 31, 2026.
The jobs an AI agent does inside a CRM
It does work that a fixed workflow cannot, because the input is messy: a call transcript, a forwarded email, a half-filled record, a vague question from a manager. A workflow can say “when stage changes to Won, create an onboarding task.” An agent can read a 40-minute call transcript, pull out the next step and the decision maker, and propose updates to three fields on the deal.
The jobs we see asked for most, each with its own page:
| Agent job | Trigger | Reads | Writes or sends | Page |
|---|---|---|---|---|
| Data entry | Call ends, email arrives, form submitted | Transcript, email body, existing contact and deal | Activity note, field updates, new contact if none matches | AI CRM data entry agent |
| Data cleanup | Nightly or weekly schedule | Contacts, companies, deals | Merge proposals, standardised fields, missing-owner flags | AI CRM data cleanup agent |
| Pipeline review | Weekly schedule or stage change | Deals, activities, close dates | Stalled-deal flags, suggested next steps, a note to the owner | AI pipeline review agent |
| Reporting | A question in chat, or a weekly schedule | Deals, activities, owners, targets | A plain-English answer with the numbers and the filter used, a weekly summary | AI CRM reporting agent |
Source: aibrevo agent scoping checklist, Oct 31, 2026. Two jobs are deliberately missing from this list: general AI receptionists and outbound sales agents. Those are not CRM-specific, and we only build them here when the work is mostly about what lands in, and comes out of, the CRM.
What native AI agents does each CRM ship?
Every one of the five vendors now has native AI, but “agent” means different things on each, and each meters it differently. The table is from each vendor’s own pages, read on Oct 31, 2026. Rates change often, so check the linked page before you plan spend.
| CRM | Native agents and AI | Plans | How it is billed | Source |
|---|---|---|---|---|
| HubSpot | Breeze Customer agent, Prospecting agent, Data agent, Breeze Assistant; agent builder for custom agents | Customer and Prospecting agents: Professional or Enterprise. Data agent: Starter and above. Agent Hub and the agent builder: HubSpot’s own pages disagree on Starter; confirm in your account | HubSpot Credits at $0.010 each ($10 per 1,000). Customer agent 50 credits per resolution (or per voice minute), Prospecting agent 100 per lead, Data agent 10 per response to one prompt for one record; custom agents consume credits per action unit | Breeze agents, Credits |
| monday CRM | Pre-built Lead Sourcing, Calling, Meeting Prep, Pipeline Monitor, Campaign Intelligence and Campaign Builder agents; no-code agent builder; sidekick assistant | Pre-built CRM agents listed for Pro and above on monday’s AI agents page; AI credits are a one-time 6,000 trial (12,000 on Ultimate), then an AI credits add-on | AI credits: agent runs about 10 to 250+ credits each, sidekick about 10 to 150+ per message, AI blocks 8 per action | CRM AI agents, AI Credits |
| Pipedrive | AI report creation, AI import assistant, Sales Assistant (beta), AI email creation, AI notifications, email summaries (beta); Nova meeting intelligence (drafts CRM updates you approve); Pipedrive MCP for ChatGPT, Claude and other MCP clients; no background agent | Report creation, Sales Assistant, Nova and MCP on all plans; email creation and notifications on Premium and Ultimate | No AI credit meter stated; MCP usage draws on the account’s daily API token budget | Pipedrive AI, Nova, MCP |
| Airtable | Omni (builds apps and agents, answers questions), Field Agents that run per record (categorise, web search, document analysis) | AI credits on every plan, including Free | Monthly credits: 500 per editor on Free, 15,000 per billable collaborator on Team, 20,000 per paid user on Business, 25,000 per paid user on Enterprise Scale. Omni building is free; an Omni question is about 10 credits; a field agent web search about 10 per run; a 10-page contract analysis about 200. Packs from $20 a month for 10,000 | AI billing |
| GoHighLevel | AI Employee: Conversation AI, Voice AI, Reviews AI, Content AI, Managed Agents; Agent Studio for custom flows | Per sub-account | Pay-per-use, Growth $50 a month or Unlimited $97 a month per sub-account; Agent Studio pay-per-use on every plan; phone charges always extra | AI product pricing |
All figures checked Oct 31, 2026. Two things to note. First, HubSpot’s Breeze page says standard workflows do not consume credits, while its knowledge base says a Breeze AI action inside a workflow can, so ask which actions in your design are metered. Second, monday’s AI Credits article covers the work platform; treat it as a guide for CRM and confirm in your Credits Usage tab. Third, Pipedrive’s AI features help a person work faster; none of them acts on records on its own schedule.
The per-platform pages go deeper on each: HubSpot AI agents, monday AI agents and Pipedrive AI agents. For GoHighLevel, start with what GoHighLevel AI Studio is and how much GoHighLevel AI agents cost, which break down the AI Employee plans and the per-minute and per-token rates.
When is the native agent enough, and when should you build a custom one?
Use the native agent when it does your exact job, on a plan you already pay for, and the per-action maths still works at your volume. Most teams should try it first: it already has permission to your data, the vendor maintains it, and switching it off costs nothing.
Build custom when one of these is true:
| Situation | Why native falls short | What a custom agent does instead |
|---|---|---|
| The answer needs data outside the CRM (billing, ERP, product usage, a support desk) | Native agents mostly see what is in the CRM plus the web | Reads the other system’s API and writes a combined result to the CRM |
| Your rules are specific (who owns a lead, when a deal counts as stalled, which duplicates may be merged) | Native builders accept a prompt, not your full rule set | Applies coded rules and uses the model only for the messy part |
| The feature is not on your plan | monday lists its pre-built CRM agents for Pro and above; Pipedrive has no background agent | Runs on the API you already have |
| Volume makes per-action credits expensive | A metered rate times thousands of records adds up | Batches work and calls the model only where judgement is needed |
| You need an audit trail or an approval queue | Native logs vary by vendor | Logs every read and write and holds risky actions for sign-off |
| You run two CRMs, or are mid-migration | Each native agent sees only its own CRM | One agent works across both, with one set of rules |
Source: aibrevo, Oct 31, 2026, from the vendor limits above. If the agent mostly needs a screen inside the CRM rather than autonomy, the work is closer to an app: see HubSpot app development, monday app development and internal tools development. If it mostly moves data between systems, it is an integration, covered on CRM integration services.
How our in-house team builds a CRM AI agent
Our in-house engineering team, 16 engineers led by co-founder and CTO Shivam, builds the agent on the CRM’s own API and the vendor’s app framework where one exists. The steps are the same on every platform.
1. Pick one job
Name the trigger, the records read, the fields written and who owns the result. You can check: a one-page job spec.
2. Map data and access
List objects, fields and the minimum API scopes. The agent gets its own credentials, never a person's. You can check: a scope list you approve.
3. Build
Coded rules for the deterministic parts, the model for reading free text and drafting. You can check: a staging run on sample records.
4. Test on real records
Run against a copy or a sandbox with your real data and compare outputs with what a person would do. You can check: a test log with the misses.
5. Guardrails and go-live
Approval queue, logging, throttling, alerts on failure. You can check: the first week's log.
Guardrails every CRM AI agent needs
The model will sometimes be wrong, so the design has to make a wrong answer easy to catch and undo. These are the guardrails we put on every agent:
- Least-privilege access. A dedicated private app, token or service account with only the scopes the job needs. A reporting agent gets read-only access.
- Human approval for risky actions. Anything sent to a customer, any merge or delete, and any change to amount, stage or owner goes to a queue first until the error rate on real records is known. monday’s own guide to building a sales agent recommends the same: role-based permissions, approval on customer-facing actions and one named owner.
- A log of every read and write, with the record ID, the old value, the new value and the reason, so any change can be traced and reversed.
- Throttling inside API limits. Airtable allows 5 requests per second per base, monday caps daily calls at 1,000, 10,000 or 25,000 by plan, and Pipedrive spends a daily token budget per seat; the full table is in our CRM API rate limits comparison.
- PII handling. Send the model only the fields it needs, keep secrets out of prompts, and check where the model provider stores data. For health data, see whether your CRM can be used at all, for example is Pipedrive HIPAA compliant and is monday.com HIPAA compliant.
- Numbers computed by code. When an agent reports a total or a count, code calculates it from the API result; the model only writes the sentence around it.
What does a CRM AI agent cost to run?
Running cost has up to three lines: the vendor’s AI meter (if you use native AI), model usage (for custom agents), and hosting. The vendor meter is the one people forget. Illustrative arithmetic from the published rates, for 1,000 actions a month:
| Native option | Rate (checked Oct 31, 2026) | 1,000 actions a month |
|---|---|---|
| HubSpot Data agent | 10 credits per response for one record, at $0.010 | 10,000 credits for 1,000 single-record answers, $100 beyond any included credits; a prompt that spans many records costs 10 credits for each |
| HubSpot Prospecting agent | 100 credits per lead | 100,000 credits, $1,000 beyond any included credits |
| monday agent runs | About 10 to 250+ credits per run | About 10,000 to 250,000+ credits |
| Airtable Omni questions | About 10 credits each | About 10,000 credits, inside the 15,000 one Team billable collaborator brings |
| GoHighLevel AI Employee Unlimited | $97 a month per sub-account | $97, plus Agent Studio usage and phone charges |
Arithmetic only, before tax and before any credits your plan includes. HubSpot says included credits vary by product and tier; monday’s AI credits work differently again. Sources: the vendor pages in the table above.
A custom agent swaps the vendor meter for model tokens and hosting, which depend on the model and how much text each run reads. We estimate those from a sample of your records during scoping. Projects are scoped and quoted per build after a free 30-minute call; we do not publish a price list for agent work.
How long does a CRM AI agent take to build?
The ranges we use when scoping: a single-job agent on one CRM, with read access and draft-only writes, is two to four weeks including testing on real records. An agent that reads two or more systems, writes on its own or needs an approval queue is typically four to eight weeks. What moves the date most is not build hours but how quickly API access is granted and how messy the records are. If the data is in poor shape, the data cleanup agent is often the first build, because every other agent reads the same records.
Who is a CRM AI agent for, and who should skip it?
It is for teams already on one of the four CRMs with a repeated, expensive chore: reps who skip logging calls, a manager who spends Monday morning exporting deals to a spreadsheet, a database full of duplicates from three imports. It is a poor fit when the CRM itself is not set up yet (fix the pipeline first, see monday CRM implementation services or our CRM setup services), when the job happens a few times a month, or when nobody will own the agent after launch.
Before you commit, check three things: that the job is written down in one paragraph, that someone can say what “correct” looks like for ten sample records, and that the plan you are on allows the API access or native feature the design needs.
Scoping a CRM AI agent? Book a free 30-minute call with your CRM, plan and the one job you want handled, or send a message. We will tell you whether the native agent is enough before we quote a custom build.
Sources
All pages opened on Oct 31, 2026.
- HubSpot, Breeze AI agents: named agents, Agent Hub availability, custom agents on credits.
- HubSpot, HubSpot Credits: $0.010 per credit, per-agent rates.
- HubSpot Knowledge Base, Understand HubSpot Credits and billing: included credits vary by tier; Breeze actions in workflows can consume credits.
- monday.com, CRM AI agents: pre-built agents, Pro plan and above, builder.
- monday.com support, Available plan types on monday CRM: 6,000 trial AI credits (12,000 on Ultimate), then an add-on.
- HubSpot, Product and services catalog: Credits rate sheet, Data agent 10 credits per response to one prompt for one record.
- Pipedrive, Nova and Pipedrive MCP: meeting intelligence and MCP access on all plans.
- monday.com support, AI Credits: credit consumption by capability.
- monday.com, How to build an AI sales agent in monday CRM: permissions and approval recommendations.
- Pipedrive support, Pipedrive AI: AI features by plan and status.
- Airtable support, Airtable AI billing: credits per plan, consumption examples, packs.
- HighLevel, AI product pricing: AI Employee plans and exclusions.
- aibrevo, CRM API rate limits compared, 2026: API limits, Oct 17, 2026.
All pages in this guide
- AI CRM Data Cleanup and Enrichment Agent: Dedupe, Standardise and Fill Gaps With Approval Solution
An AI CRM data cleanup agent finds likely duplicates and proposes merges, standardises names, phones, countries and job titles, fills missing company data from public sources, and flags bad emails and phone numbers, with a person approving every merge. This page sets out the native dedupe and enrichment tools in HubSpot, monday CRM, Pipedrive and GoHighLevel (sourced, checked Oct 31, 2026), where they stop, and how our in-house engineers build a custom agent. - AI CRM Data Entry Agent: Turn Emails, Calls and Forms Into CRM Updates Solution
An AI CRM data entry agent reads emails, call transcripts, meeting notes and form submissions, then creates or updates contacts, deals, notes and next steps in HubSpot, monday CRM, Pipedrive or GoHighLevel. This page compares what each CRM's built-in AI already fills in (sourced, checked Oct 31, 2026) with a custom agent built by our in-house engineers, and covers guardrails, running costs and scoping. - AI CRM Reporting Agent: Ask Your CRM in Plain English and Get a Weekly Pipeline Summary Solution
An AI CRM reporting agent answers plain-English questions about your deals and activity (which deals over $20k have had no activity in 14 days?) and sends a weekly pipeline and activity summary to Slack or email. This page covers what the agent reads, what native reporting AI HubSpot, monday CRM and Pipedrive already ship (checked Oct 31, 2026), when a custom agent is worth building, how we keep the numbers correct, and what it costs to run. - AI Pipeline Review Agent for HubSpot, monday CRM and Pipedrive Solution
An AI pipeline review agent reads every open deal in your CRM on a schedule, flags stale deals, missing next steps and close dates, and deals stuck in a stage, drafts follow-up tasks for the deal owner to send, and gives managers a weekly deal review. This page compares it with Pipedrive's rotting feature, HubSpot deal scores and Breeze, monday CRM automations and AI credits, and Airtable field agents, all checked against vendor docs on Oct 31, 2026. - HubSpot AI Agents: Breeze Agent Setup vs a Custom Agent Built on HubSpot's APIs Solution
HubSpot AI agents come in two forms. Breeze agents (customer, prospecting, data and others) run inside HubSpot on HubSpot Credits, from 50 credits per resolved conversation. Custom agents are either built in HubSpot's agent builder or run outside HubSpot and write back through the API, workflows, custom coded actions and webhooks. This page explains what each Breeze agent does, which tiers it needs, what it costs to run, and when our in-house team builds a custom agent instead. - monday.com AI Agents for monday CRM: Built-In Agents, AI Credits and Custom Builds Solution
monday.com AI agents are monday's own agents (the AI Lead Agent, the AI Sales Agent and the agent builder) plus custom agents built on the monday API, webhooks and automations. They run on AI credits: about 10 to 250+ credits per agent run, 150 credits per five-minute AI Sales Agent call, with a one-time trial of 6,000 credits on most monday CRM plans. This page shows what each option does inside monday CRM, when the built-in agent is enough, when a custom build is worth it, what it costs to run and how we build one. - Pipedrive AI Agents: Built-In Pipedrive AI vs a Custom Agent on the API (2026) Solution
What Pipedrive's own AI does today (Sales Assistant, AI email creation, AI notifications, Nova meeting intelligence, MCP access for ChatGPT and Claude), which plans include each feature, and when a business needs a custom Pipedrive AI agent built on the API and webhooks instead. Includes the jobs a custom agent does, the records it reads and writes, how we build it with human approval and token budgets, and what it costs to run.
