Attio AI in 2026: every feature, what it does, what it costs

"Does Attio have AI?" stopped being the right question in 2026. The right question is which of Attio's AI features are worth turning on, which ones eat through credits, and which ones replace a real workflow.

This post walks through every AI feature Attio ships today. What it does, when to use it, what it costs, and the edges where it still struggles.

The shape of Attio AI

Attio's AI splits into four parts. Knowing the shape makes the pricing and the workflows make sense.

• AI Attributes. Fields on records that run an AI model to fill themselves in. Four types: Research, Classify, Summarize, Prompt Completion.
• Research Agent. A standalone AI agent that can search the web and your workspace to answer open questions, and drop structured answers back on a record.
• Call Intelligence. Meeting recorder plus transcription plus AI summary, pulled straight into the record for whoever the call was with.
• AI in Automations. AI steps you can drop into a workflow to classify, research, summarize, or decide, and then branch on the result.

Everything else people call "Attio AI" is a combination of these four. There is no fifth thing hiding behind a sales call.

1. AI Attributes

AI Attributes were the feature that turned Attio from a flexible CRM into an AI-native one. They sit on records the same way a text field or a select field does, except the value is generated by an AI model.

Four types ship today.

Research

Runs a web search scoped to a specific question and writes the answer back into the field. Good for fields like "What do they sell?", "Most recent funding round", "Core tech stack", or "Key competitors".

The Research attribute is the same engine the Research Agent uses, scoped to one question instead of an open brief.

Classify Record

Picks one or more options from a set list based on everything Attio already knows about the record. The obvious use is ICP scoring: give the model five segments and let it pick. Less obvious uses include lead source normalization, industry tagging, and pipeline stage suggestion.

Classify is the cheapest of the four in practice because you can often run it once per record and cache the answer.

Summarize Record

Collapses every note, email, meeting, and linked record into a paragraph. Useful for the top of a company page where you want context at a glance without scrolling through six months of history.

The catch: summaries go stale. If the record changes a lot, the summary needs to re-run, which costs credits every time.

Prompt Completion

The flexible one. Write any prompt, reference any attribute on the record or on linked records, get a text answer back. This is how you build "Draft a personalized opener" or "Suggest the next step" or "Write a one-line deal risk note".

Most of the interesting AI-native Attio setups are 80% Prompt Completion attributes with a few Classify attributes around them.

2. Research Agent

The Research Agent is the highest-profile AI feature Attio shipped in 2025 and it is still the one teams use wrong most often.

It works like this. You give it an open question. It searches the web, reads what it finds, cross-references against what Attio already knows about the record, and writes back a structured answer. Not a paragraph of fluff. Specific fields you asked for, filled in with citations.

Example brief: "Research this company. Return their funding history, current headcount, tech stack, and two relevant recent news items." Ten seconds later those four fields are filled in on the company record.

The important part most teams miss: the Research Agent can be triggered inside a workflow. New company created, run the Research Agent, route to the correct segment based on what it found, assign to the right owner. That is the part that replaces a research hour per lead.

When to use it

• New inbound leads where you want context before the first reply.
• Outbound lists where you want qualification before the first touch.
• Quarterly refresh of existing accounts: what has changed in the last 90 days.
• Event follow-up where you need fast context on 40 new records at once.

When not to use it

• Records where the data is already in your CRM or your product database. Use a workflow block, not an agent.
• Cases where a cheaper Research attribute with a tight question would do the job. Agents cost more credits.

3. Call Intelligence

Attio's call recorder joins Zoom, Google Meet, or Microsoft Teams, records the call, transcribes it, and generates a summary with action items. The summary lands on the linked company or person record automatically.

Call Intelligence is a Pro-plan feature. If you are on Plus, you do not get it. For outbound sales teams that live in calls, this is the single biggest reason to upgrade.

What it does well: summarizing a 45-minute call into a tight paragraph plus next steps, and surfacing themes across all calls with a single account.

What it does not do yet: coaching-grade scoring (talk-to-listen ratios, filler words, objection handling) at the level of a dedicated product like Gong.

4. AI in Automations

Every AI attribute can also live inside a workflow. That is where the compounding starts.

Three building blocks matter.

• AI classify step. Looks at a record, picks an option, branches the workflow.
• AI research step. Runs the Research Agent or a scoped research query inside a flow.
• AI prompt step. Generates text (an email, a summary, a note) and uses it downstream.

You string those together to get real automations. Three examples that land on every implementation I ship.

Inbound lead routing. New company created, AI research step pulls funding and headcount, AI classify step picks the segment, workflow branches to the right owner and Slack channel.

ICP drift monitor. Weekly recurring workflow reruns ICP Classify on every open deal, flags deals where the ICP score dropped, posts a weekly digest.

Outbound personalization. Before a sequence fires, AI prompt step generates a one-line opener using the prospect's most recent news. The opener drops into the first email.

These are not futuristic. They are three-hour builds inside Attio today.

How credits actually work

Attio bills AI usage in workspace credits. This is where it pays to know the numbers.

Two cost rules to keep in mind.

• Research Agent consumes about 10 credits per run. 10,000 credits on Pro is roughly 1,000 agent runs a month.
• Cheap AI attributes (Classify, short Prompt Completion) cost 1 to 2 credits each. You get thousands of those per month before the ceiling hits.

Top-up packs exist: 5,000 credits for $85, 10,000 for $150, 25,000 for $225. Most teams who actually build workflows end up on one of these packs in month three.

The trap: turning on AI attributes across every record without scoping them to a view or a trigger. A Summarize Record attribute that runs every time any linked record updates will burn through credits before anyone has seen a benefit. Always scope AI attributes to a view or a trigger, not a global recompute.

Five AI workflows worth shipping on day one

These are the five I install in every AI-native Attio setup. They pay for the credits inside the first month.

1. Inbound research. Any new company, run the Research Agent, fill six structured fields, assign to the right rep. Replaces a 20-minute manual lookup per lead.
2. ICP score. Classify every company against 4 to 6 ICP segments, surface the score on every view. Replaces a spreadsheet that nobody updates.
3. Deal risk note. On every open deal, a Prompt Completion attribute reads the last 30 days of activity and writes a one-line risk flag. Replaces the "can you give me a status?" Slack message.
4. Call follow-up draft. Call Intelligence plus a Prompt Completion step generates the follow-up email draft within 60 seconds of the call ending. Replaces the 10-minute post-call writeup.
5. Weekly pipeline digest. Recurring workflow summarizes every open deal into a one-page Slack post every Monday. Replaces the stand-up nobody reads.

None of these need code. All of them need careful prompt design and a clear field schema. That is most of the implementation work.

What Attio AI still does not do

Three honest gaps.

• Bulk historical runs. Attio's AI is built to run per record, on trigger. Running it across 20,000 existing records in one go is possible but credit-expensive and slow. For big historical enrichments, a scripted run outside Attio is often cheaper.
• Coaching-grade call analytics. Call Intelligence summarizes well. It does not score reps, benchmark talk ratios, or flag deal risk from voice patterns.
• Custom model selection. You use Attio's model choices. If you want to pick Claude over GPT, or use a fine-tuned model for a specific field, you cannot do that inside the UI today.

None of these are deal-breakers for most teams. They are worth knowing before you commit.

Who Attio AI is actually for

A few clear fits.

• Sales teams with high inbound volume where the research time per lead compounds.
• Outbound teams who need personalization at scale without a dedicated researcher.
• Customer success teams who want deal risk flags and auto-generated account summaries.
• Ops teams who build the workflows once and let the rest of the company consume them.

The fit is less obvious for:

• Solo founders on the Free or Plus plan with low lead volume. The manual research is cheaper than the credit spend.
• Service businesses with a small account list where personalization is already one-to-one.
• Enterprise teams already paying for Clay, Clearbit, Gong, or a full data stack. Much of Attio AI overlaps and the double spend rarely makes sense.

The honest take

Attio AI in 2026 is the strongest AI layer of any CRM under $100 per user per month. Not because the individual features are unique, every CRM is racing to ship the same four building blocks, but because Attio's data model is flexible enough that the AI attributes and the Research Agent fit naturally on any record type.

The catch is the same catch as the rest of Attio: flexibility is a tax if you do not know what you want. Teams who turn on every AI feature and hope for insights burn credits without building anything. Teams who design three specific workflows and run them against specific triggers pay back the Pro upgrade inside the first month.

That design step is what most AI-native Attio implementations are buying. The features are included. The judgment about which ones to turn on, with which prompts, against which triggers, is not.

Sources

• Attio: Embedded intelligence
• Attio: Introducing the AI-powered Research Agent
• Attio: Automations and workflows
• Attio: Research agent help center
• Attio: Go to market, supercharged by AI
• Attio: Pricing
• Attio: Plan and pricing changes
• CheckThat: Attio pricing breakdown
• Lightfield: Attio pricing, real costs
• Stacksync: Attio CRM 2026 review

Free audit of your Attio workspace

If you want a second pair of eyes on which AI features are worth turning on for your workspace, I run a free 48-hour audit. You add me as an Attio expert, no extra seat and no billing. I send back a one-page teardown ranked by impact, the three highest-leverage AI workflows to ship first, and a 5-minute Loom walking through the top one. No call, no pitch. 5 slots a week.

Get your free Attio audit