Attio Just Turned Its CRM Into a Database for AI Agents [2026]
On October 14, Attio CEO Nicolas Sharp announced what the company calls the architecture for Agentic Revenue — a rebuild of the CRM's data layer around one idea: "To agents, Attio looks like a database." Four new capabilities shipped with the announcement: a SQL Engine, Postgres-compatible Data Connectors, Agent Scale reads, and a new Activities data type.
Sharp's framing is the sharpest one-liner of the agent-CRM year so far: "In the age of agents, systems of record have never mattered more. They've also never been less prepared for what's coming."
He's right about the problem. Every vendor we've covered this fall — Oracle, Salesforce, Outreach — is bolting agents onto a system of record designed for humans clicking through list views. Attio's bet is different: make the record system itself speak the language agents already know. That's genuinely interesting. It's also, in its most useful form, gated behind a custom-priced Enterprise plan.
Here's what actually shipped, what it costs to reach, and what it does and doesn't change for a sales team.
The Four Pieces
Per Attio's announcement and its new architecture page:
1. SQL Engine. Attio now answers SQL queries natively — joins, aggregations, and calculations run inside the CRM and return exact results, with every query left visible for inspection. The launch examples are pure RevOps: lead-source win rates on deals over $50,000, weighted pipeline by region and owner, open opportunities attached to churned customers. SQL also runs inside Attio's Workflows, so a query result can trigger an action — a sequence, a task, a Slack ping.
2. Data Connectors. A live, Postgres-compatible endpoint on your CRM data. Point Hex, Omni, Looker, Metabase, or dbt at Attio with standard PostgreSQL credentials and query it like any other database. Connections are read-only, with dedicated credentials workspace admins can revoke.
3. Agent Scale. The infrastructure claim: hundreds of millions of records per workspace, with thousands of agents working across the data simultaneously — all governed by the same permissions and access controls as human users. Attio's own framing of why this matters: "For every one record a human might check, an agent can read thousands."
4. Activities (beta). A new first-class data type for customer touchpoints that traditionally live outside the CRM — support tickets, product events, social DMs — with custom schemas. The launch example: a support escalation gets tagged before a renewal conversation happens.
One honesty note on maturity: the launch post describes all four capabilities as shipping "in beta," while the same post's get-started section applies the beta label only to Activities. Treat the whole stack as early and expect rough edges.
Why SQL Is the Headline
The entire agent-integration wave of 2026 has been about giving models access to CRM data. Salesforce put its CRM inside Claude. Salesloft and Clari shipped MCP servers. Gong, Apollo, and ZoomInfo built agent platforms on top of their data moats.
Almost all of it works the same way under the hood: the agent calls a REST API that returns pages of records, and the model does the math itself. That's slow, token-expensive, and — the part nobody puts in the launch post — unreliable, because LLMs are bad at arithmetic across a thousand rows of JSON.
A SQL engine changes the division of labor. The agent writes the query; the database does the computation; the agent gets back one exact number. When Claude's Dashboards beta shipped NL-to-SQL over Salesforce this month, it was the same insight from the model side. Attio is meeting it from the CRM side: don't make agents paginate through an API built for web apps — let them query.
The acquisition trail backs this up. Attio acqui-hired Inconvo in July — a Y Combinator-backed startup whose product was literally a conversational analytics API that translated natural-language questions into exact database answers. Three months later, Attio speaks SQL. That is a straight line.
The Fine Print: What's Actually Gated
Now the part the X threads skipped. From the announcement's own get-started section:
| Capability | Who gets it |
|---|---|
| SQL-powered questions in Ask Attio (chat) | All plans |
| SQL via REST API, MCP, and Workflows | Enterprise only |
| Data Connectors (Postgres endpoint) | Enterprise only |
| Agent Scale | Platform-level |
| Activities | Beta |
Read that table carefully. The demo of the agent-native CRM — asking questions in chat and getting SQL-backed answers — is on every plan, including the ones with published pricing: Plus at $35 and Pro at $79 per user per month on annual billing. The surface agents actually use — SQL over the API, over MCP, inside Workflows, and the Postgres endpoint — is Enterprise, which is quote-only.
So "to agents, Attio looks like a database" has an unstated second clause: if you're on the plan with custom pricing. We flagged the same pattern when Salesloft and Clari gated their MCP servers behind $125–165 quote-only agent seats: the open-protocol story and the packaging story point in opposite directions. Attio is friendlier than most — Ask Attio on free and low tiers is real access, and read-only Postgres credentials that admins can revoke is a cleaner security model than most MCP deployments. But if you're a 10-person sales team on Pro imagining your AI SDR agent running win-rate SQL through the API, you're not in the addressable market yet. Budget for a sales conversation first.
It's also worth saying what didn't change on October 14: pricing. No new per-agent, per-query, or per-action meter was announced. After a year of watching Agentforce's $0.10-per-action math and Microsoft's credit packs turn agent demos into procurement spreadsheets, "it's in your Enterprise plan" is refreshingly simple — until the Enterprise quote arrives pre-loaded with the new architecture's value.
How Attio Got Here
The October announcement is the fifth move in a year-long sequence, and the sequence is the strategy:
- February: granular permissions and Ask Attio, the natural-language chat layer.
- April: connections into Claude, ChatGPT, and Notion — meet the agents where they live.
- June: Workflows, pitched explicitly as "orchestrate revenue agents, on every account."
- July: the Inconvo acqui-hire (SQL/analytics) — and, quietly the same month, a price raise across Plus and Pro.
- October: the architecture announcement that ties it together.
Attio raised a $52M Series B in August 2025 with around 5,000 paying customers, and its current customer logos skew exactly the way you'd expect for this pitch — AI-native startups like Granola and Modal, companies whose "sales team" already includes agents. The positioning war here isn't against legacy Salesforce seats; it's to be the default system of record for companies that are agent-first from day one. Clay is running the adjacent play from the data side with Sculpt; Attio is running it from the record side.
What an Agent-Queryable CRM Doesn't Solve
Here's the thing the architecture page can't say: a CRM that agents can query at infinite scale is still only as useful as what's in it.
Every example in Attio's launch — win-rate analysis, weighted pipeline, churn-linked opportunities, renewal-risk tagging — is an agent working records you already have. That's the same boundary we hit with Oracle's Fusion Claw agents and monday.com's Harmony voice agents: the execution layer keeps getting dramatically better, and the targeting layer — which accounts should be in the database in the first place, and which ones are in-market right now — stays someone else's problem.
For a B2B sales team, the honest sequencing is:
- Signal and targeting first. Who's visiting your site, who matches your ICP, who's showing intent this week. If this layer is weak, SQL over your CRM returns exact answers about the wrong accounts.
- Then the system of record. Attio's architecture makes this layer genuinely better for agents — exact aggregates instead of hallucinated math is a real upgrade.
- Then execution agents. Outreach, voice, demos — the layer every vendor is shipping this quarter.
MarketBetter lives at layer one and three: identifying in-market accounts from visitor and intent signals, then running the outreach loop on them. An agent-queryable CRM in the middle makes that stack stronger, not redundant — which is why we'd call Attio's announcement good news for the category and insufficient on its own.
Bottom Line
Attio's architecture bet is the most technically coherent answer yet to a real problem: agents are terrible at consuming CRMs built for human clicks, and REST-pagination-plus-LLM-math is a dead end. A native SQL engine with permissions behind the query is the right shape, and the Inconvo acquisition shows the plan was deliberate.
But match the pitch to the packaging before you move. The agent-facing surface is Enterprise-gated and quote-only, every component is early (some or all of it beta, depending on which Attio page you read), and the architecture solves how agents read your pipeline — not how the pipeline gets filled.
If the part you actually need is filling it — knowing which accounts are in-market and getting agents to work them — book a MarketBetter demo and we'll show you the targeting layer that makes any CRM, agent-native or not, worth querying.