Attio Workflows: The Complete Guide (2026)
•
Attio workflows are automations that run inside your CRM: a trigger fires, blocks execute, and records get updated, scored, and routed without anyone touching a spreadsheet. In June 2026, Attio rebuilt Workflows from the ground up around AI agents. You can now describe an automation in plain language and have Attio assemble it, run custom agents over your full customer context, execute JavaScript mid-run, and pay for AI by actual token usage instead of flat credits.
This guide covers the new Workflows engine, not the legacy builder it replaced. You will learn how workflows are structured, how to build your first one, what custom agents can and cannot do, what credits cost, seven workflows worth copying, how to migrate legacy workflows, and where Zapier, n8n, and Make still earn a place next to Attio. Everything here is verified against Attio's own documentation as of July 2026. If you are still evaluating the platform itself, start with our complete Attio guide.
What changed in June 2026
Attio launched the new Workflows on June 9, 2026, and describes it as an engine for orchestrating revenue agents alongside your team. The pitch, in co-founder Nicolas Sharp's words at launch: a system that thinks, researches, routes, and follows up across your entire revenue motion, on top of what Attio calls Universal Context, the layer that connects records, emails, call recordings, notes, and external tools. Sharp noted that hundreds of thousands of automations already run on Attio every day.
The practical differences from legacy workflows:
Capability | Legacy workflows | New Workflows (June 2026) |
|---|---|---|
AI in workflows | Fixed blocks (Classify, Summarize, Prompt completion) | Custom agents with tool access, web research, MCP, and structured outputs |
Building | Manual canvas only | Describe it in plain language and Ask Attio assembles it, or build on the canvas |
Code | Not available | Execute custom JavaScript in a sandboxed block |
Triggers | One per workflow | Multiple triggers per workflow |
Failures | Manual do-overs | Retry failed runs, pause, cancel, and resume mid-run |
Logic and lookups | 1 credit per block | Free |
AI pricing | Flat credits (10 per research block) | Token-based, pay for what the model actually uses |
A follow-up release on June 29 added structured outputs for custom agents, loops over arrays of objects, and clearer error messages that point to the exact block that received bad data. The canvas also now reads left to right, which sounds cosmetic until you build a workflow with three branches and can actually follow it.
One thing that did not change: workflows are available on every Attio plan, including Free. What varies by plan is the credit allowance that runs consume, covered below.
How Attio workflows are structured
Every workflow is a trigger plus one or more steps, chained on a canvas. Blocks take inputs, the values you configure, and produce outputs, which downstream blocks reference as variables.
Triggers start the run. The block library groups them into record triggers (record created, record updated, attribute value changed, a manual Record command button on any record), list triggers (record added to a list, list entry updated), schedules, and incoming webhooks. The new engine supports multiple triggers on one workflow, so a deal-scoring workflow can fire automatically when a deal hits Evaluation and also give reps a button to re-run it on demand.
Steps do the work. Logic blocks (If, Switch, Loop, Filter, Delay) route and gate the run. Data blocks find, create, and update records, entries, and tasks. Action blocks send Slack messages, enroll people in sequences, and make HTTP requests. AI blocks classify, summarize, and run agents. App blocks from partners like Notion, Linear, Fin, and lemlist push data into the rest of your stack, and developers can build their own with the Workflows App SDK.
Every run is logged. You can open any run, see what each block received, what each agent read and decided, and rerun a single failed run without redoing the rest. Workflows also inherit the permissions of the user you assign them to, which matters once automations start writing to records your interns should not touch.
How to build your first workflow
There are two ways to build: describe it, or drag it.
The fast way: Ask Attio. Open a new workflow and describe the goal in plain language. Something like:
Every Monday at 9am, find open deals with no activity in the last 14 days and send a summary to #sales in Slack.
Attio assembles the trigger and blocks for you. From there, tweak conversationally ("run it Tuesdays instead") or edit blocks directly. In my client work this is now the default starting point, even for complex builds. The first pass usually needs a tweak or two, but editing a mostly-right workflow beats assembling one from scratch.
Treat every Ask Attio build as a draft to verify, not a finished system. It does not know which attributes your reps actually maintain or where your pipeline diverges from the default, and a workflow that is 90 percent right can still write to the wrong field on every run. Attio gives the same advice in its own docs: check every block before you publish. Closing that last 10 percent, and designing the data model the workflow depends on, is exactly the work an Official Attio Expert Partner does.
The manual way: the canvas. Pick a trigger, add blocks, wire the outputs of one block into the inputs of the next, and publish. Two habits will save you debugging time. First, test block by block as you build rather than publishing and praying. Second, before you publish anything that writes data, check which object and attributes each write block touches, because a workflow misfiring at scale makes a mess at scale.
One configuration detail most teams miss: workspace-level custom context. In workspace settings you can tell Attio's AI how your business runs, what your pipeline stages mean, and how you write. Both Ask Attio and your workflow agents use it, and five minutes here noticeably improves everything the AI produces.
Custom agents: AI judgment inside the workflow
The custom agent block is the headline feature. It runs an AI agent as a workflow step, with a prompt you write, variables from earlier blocks, and access to three kinds of context: your Attio data (records, notes, emails, meeting transcripts), the apps and integrations connected to your workspace (Granola notes or Apollo, for example), and live web research. Attio's launch post says agents can reach anything with an MCP server; in the builder, tool access is configured through the apps connected to your workspace, so check Attio's app store for the tools your play depends on before designing around them.
Agents return outputs, and since the June 29 update you can define a structured output schema so the agent returns exactly the fields your next blocks expect, rather than a blob of text. The agent itself does not write to your CRM. You pair it with a downstream update block that writes its output to records, which is the right design: the probabilistic part reasons, the deterministic part writes.
A concrete example from Attio's own solutions team: a CHAMP qualification workflow. When a deal moves to Evaluation, an agent reads the notes, emails, and call transcripts attached to the deal, scores challenges, authority, money, and prioritization, and returns each score in a structured output. An update block then writes those scores to deal attributes. The same pattern adapts to MEDDPICC or any framework your team runs, and it replaces the qualification fields reps never fill in.
Two limits worth knowing. Agents draft email copy but do not send email directly; you write drafts to an attribute and send through a sequence block, keeping a human in the loop. And agents are the expensive blocks in the new pricing, billed by token usage, so a prompt that tells the agent exactly which context to read costs less than one that lets it wander.
Code and SQL: keep the deterministic parts deterministic
The strongest builds I have shipped on the new engine split work by what each block is best at: agents for judgment, code for math, and rules for everything that must be identical every time.
The execute code block runs custom JavaScript in a sandboxed environment, taking values from upstream blocks as inputs. The canonical use case is weighted ICP scoring: a web agent gathers facts about a new company (sector, headcount, funding), then a code block applies your scoring algorithm. Same inputs, same score, every time, with the full reasoning visible in the output. Ask an agent to score leads instead and you can get a different answer on every run, which is exactly what you do not want feeding lead routing. You do not need to write the JavaScript yourself; Ask Attio has the workflow's full context and generates the code from a description.
Attio also previewed a SQL query block in its June 2026 workshop: joins, aggregations, and conditions running against your live records, for questions like "how long has each open deal been sitting in its stage." It extends the SQL capability already in Ask Attio, and it is a better answer than an agent for aggregating thousands of records, which gets slow and expensive in a large context window. As of July 2026 it is a preview, not generally available, so plan around it rather than on it.
Seven Attio workflow examples worth copying
These are patterns I build repeatedly for clients, all possible on the new engine today.
Deal qualification scoring. Trigger: deal enters a pipeline stage, plus a manual button. Agent reads notes, emails, and transcripts, scores against CHAMP or MEDDPICC with structured outputs, and an update block writes the scores to the deal.
Lead scoring and routing. Trigger: record created. Web agent researches the company, a code block applies weighted ICP scoring, an If block routes by score band, and an update block assigns the owner. High-fit leads can go straight into a sequence with a personalized opener drafted by the agent.
Pre-meeting briefs. Trigger: scheduled, or fired from a calendar event via webhook. Agent compiles deal history, recent emails, open threads, and company news into a brief, delivered to Slack before the call. Do not go hunting for a template library to start from: Attio's own team recommends describing plays like this to Ask Attio, because every workspace's objects and pipeline stages differ. That is also the honest case for an Official Attio Expert Partner: not templates, but plays designed around your specific data model.
Call follow-ups. Trigger: attribute value changed when a call completes. Agent summarizes the conversation, updates deal fields, creates tasks, and drafts the follow-up email to an attribute for the rep to review and send.
Product feedback capture. Trigger: deal moves to Lost. Agent extracts every piece of product feedback from the deal's context as an array, a loop iterates over it, and an app block files each item into Notion, Linear, or your feedback tool. Your product team gets structured loss reasons without anyone re-reading old calls.
Stuck deals that explain themselves. Trigger: daily schedule. A lookup finds open deals quiet for 14 days, an agent reads the last emails and call transcripts and classifies why each one stalled (ghosted, ball with the rep, or dead), and a Switch block runs a different play per reason: a re-engagement draft, a nudge to the rep, or a close with the loss reason logged. Pipeline review happens before anyone walks into the room.
Single-threaded deal detection. Trigger: weekly schedule across open deals. An agent counts engaged contacts per deal from email and call activity, flags deals running through one person, and the web agent maps likely stakeholders so the rep gets a drafted intro ask instead of a generic "multi-thread this" reminder.
Closed-lost revival. Trigger: weekly schedule over deals lost 90 or more days ago. The web agent checks each account for funding rounds, hiring spikes, and exec changes, and when a signal lands, an agent drafts a re-entry note referencing it and routes the account back to the owner. Net new pipeline from deals you already paid to create once.
One thing all seven have in common: they are only as good as the data model underneath them. Agents can only read context your workspace actually captures, so the objects, attributes, and pipeline stages you design up front determine the ceiling of every workflow you build later. That design work is where an expert pass pays for itself.
Attio workflow credits, explained
Workflow runs consume workspace credits, and the new engine reprices blocks in your favor. Attio's migration guide publishes the comparison:
Block type | Legacy cost | New Workflows cost |
|---|---|---|
Triggers (internal and external) | Free | Free |
Logic and control (If, Switch, Loop, Delay) | 1 credit | Free |
Data lookups (Find records, Calculation, Aggregate) | 1 credit | Free |
Data writes (create, update, delete) | 1 credit | 1 credit |
External pushes (Slack, webhooks, sequences) | 1 credit | 1 credit |
AI blocks (Classify, Summarize, Web agent, Custom agent) | 1 to 10 credits | Variable, token-based |
Code execution | Not available | Variable, complexity-based |
Attio's summary is that 9 in 10 block runs are free, with AI billed by actual model usage. In practice, rule-based workflows got cheaper, and AI-heavy workflows now cost what they actually consume. The trade-off: because AI blocks are token-based, the old per-run credit cap setting is gone, so watch your usage dashboard during the first weeks of an agent-heavy rollout, especially on workflows that loop.
One nuance worth knowing: workflow runs draw on workspace credits, while building with Ask Attio draws on your personal seat credits, so prompting your way to a workflow does not eat the allowance the workflow runs on.
Monthly workspace credit allowances by plan, per Attio's pricing page: Free gets 250, Plus ($29 per user/month billed annually, $36 monthly) gets 1,500, Pro ($69 annually, $86 monthly) gets 10,000, and Enterprise is custom. Add-on packs start at 5,000 extra credits for $70 per month billed annually ($85 monthly) and scale to 50,000 for $475 ($595 monthly). For most teams running agents on real volume, Pro's 10,000 credits is the practical floor. For the full plan-by-plan breakdown, see our guide to Attio pricing.
Attio workflows vs Zapier, n8n, and Make
The question I get most from clients: does the new engine replace our automation tool? For CRM-centric work, mostly yes. For cross-app plumbing beyond the CRM, the tools are complementary rather than competing, which is why Attio maintains a native Zapier integration in its own app store.
Native workflows win whenever the automation needs customer context. An external tool sees whatever fields an API call returns. An Attio agent sees the emails, call transcripts, and notes on the record, plus every related record. No Zapier workflow can CHAMP-score a deal off call transcripts it cannot read. Native also wins on operational grounds: run history lives on the record, workflows respect user permissions, and there is no per-task fee stacked on top of your CRM bill.
External tools win when Attio is one node in a longer chain. Zapier's native Attio app connects thousands of tools and is the fastest path for form intake (Typeform, Tally), enrichment handoffs (Clay, Apollo), and pushing CRM events into ops tools. Make suits heavier multi-app scenarios with visual data mapping. n8n is the pick for self-hosted control and complex transformation logic, connecting through Attio's REST API and webhooks, and it is what we deploy at Automation Jinn when a client needs orchestration they own end to end.
The good news is you do not have to choose. Attio workflows trigger from incoming webhooks and end with HTTP request blocks, and Attio emits webhooks on record changes, calls ending, and notes being created. The pattern that works: let Attio handle everything that touches CRM context and judgment, let your orchestrator handle cross-app plumbing, and connect the two with webhooks in whichever direction the data flows.
Migrating legacy workflows
Legacy workflows keep running on their old pricing, but new capabilities only land in the new engine, so migrate deliberately rather than never.
The fastest path is Ask Attio: point it at a published legacy workflow and ask it to recreate the workflow in the new engine. Attio recommends this route and so do I, with one hard rule: review every block before publishing. AI migration is a draft, not a guarantee. Several blocks were renamed or reworked (Attribute changed is now Attribute value changed, If/else is now If, Prompt completion folded into the custom agent block), and variable references between blocks deserve special attention.
Two practical notes from doing these migrations. First, you have to publish a new workflow to test it, and the legacy version is still listening for the same trigger, so disable one side or you will double-fire writes. Second, do not force a one-to-one mapping. The new engine has blocks the old one lacked, and many legacy workflows are better rebuilt around an agent than translated literally. If you are moving to Attio from another CRM entirely, our walkthrough on migrating from HubSpot to Attio covers the data-model side of that project.
Honest limitations
The new engine is excellent and I recommend building on it, but go in with clear eyes.
There is no test or sandbox mode yet. You publish to test, so workflows that write data need careful staging; trigger filters scoped to a test record are the practical stopgap. Attio has confirmed a safe test environment is on the roadmap.
AI costs are variable by design. Token-based pricing means you pay for what agents actually consume, which works in your favor on well-scoped prompts and against you on vague ones that loop over big lists. Attio applies guardrails against runaway runs, and pairing those with a weekly glance at the usage dashboard keeps costs predictable.
Agents reason, blocks write. Agents do not send email directly or take destructive actions on their own; they return outputs that deterministic blocks act on. For a system writing to revenue data, that is the right call, and it means the human-in-the-loop checkpoints are yours to design.
A few conveniences from the legacy builder, like canvas notes, have not carried over yet, and the SQL block is still in preview. Given that Attio shipped two workflow releases in June alone, I expect this list to shrink quickly.
None of these change the verdict. The rebuilt Workflows is the strongest reason yet to run your GTM motion inside Attio rather than duct-taping it on top.
Frequently asked questions
What are Attio workflows?
A workflow is a set of automated rules that runs inside Attio: when something happens, take these actions. Each workflow pairs a trigger, like a record update, schedule, or webhook, with blocks that update records, send Slack messages, run AI agents, or execute code. Workflows are included on every Attio plan.
How much does Attio cost?
Attio's plans are Free ($0, up to 3 seats), Plus at $29 per user/month billed annually ($36 monthly), Pro at $69 ($86 monthly), and a custom-priced Enterprise tier. Workflows are available on every plan; what changes is the monthly workspace credit allowance runs consume: 250, 1,500, 10,000, or custom.
What are Attio workflow credits?
Credits are the usage currency workflow runs draw from your workspace allowance. In the new engine, triggers, logic blocks, and data lookups are free, record writes and external pushes cost one credit, and AI blocks are billed by token usage. Attio reports that 9 in 10 block runs are free.
Do Attio workflows replace Zapier?
For CRM-centric automation, mostly yes: native workflows read your full customer context and run AI agents external tools cannot replicate. Zapier, n8n, and Make still win for long chains across many non-CRM apps or tools without Attio blocks. Many teams run both, connected through webhooks and HTTP blocks.
Can Attio workflows send emails automatically?
Custom agents draft email copy but do not send it directly, a guardrail that keeps a human in the loop. For automated sending, write the agent's draft to an attribute and enroll the contact through a sequence block, which sends from your connected inbox on a schedule you control.
How do I migrate legacy Attio workflows to the new builder?
Ask Attio can read a published legacy workflow and recreate it in the new engine, which Attio recommends as the fastest path. Review every block before publishing, since several were renamed or reworked, then disable the legacy version so both workflows are not listening for the same trigger.
Sparsh Gupta, Founder of Automation Jinn and an Official Attio Expert Partner, helps B2B SaaS teams design Attio workspaces and the AI workflows that run on them. If you want lead routing, scoring, and follow-up running on agents instead of your reps' to-do lists, book a discovery call.
