Your CRM runs itself
AI agents that live inside your Attio workspace. They clean your pipeline, read your call recordings, prep your meetings, and turn every call into follow-up tasks. You review and approve.
Agents are included in the 7-day sprint, or available as standalone builds.
Book a callWhat this replaces
Most teams have someone who does the CRM admin manually: chasing reps to update stages, cleaning duplicate records, compiling pipeline updates for leadership, writing follow-ups from call notes.
That work costs $2-5k a month of human time, runs inconsistently, and does not scale.
The agents do the same work on a schedule, on your existing Claude Code subscription, so there is no separate AI bill and no per-token cost. The human stays in the loop for anything irreversible.
Short answer
Attio AI agents are background automations that run inside your workspace on triggers (new lead, meeting end, daily schedule). They read and write records using Claude. We ship 6 named agents with every Sprint: Pipeline Hygiene, Deal Focus, Meeting Prep, Post-Call Action Items, Sales Coaching, and Content Mining. We have also shipped operations agents for Reporting, Backups, and Import de-duplication.
The six agents
Every Sprint ships all six, scoped to how your team actually sells.
Pipeline hygiene
Daily schedule
Sweeps your open deals every night. Flags merge candidates for review, moves stages that no longer match real activity, and creates a task for any deal that has gone quiet.
Stages that reflect reality. No duplicates, no deals parked in one stage for six months.
Deal focus
Weekly schedule, Monday
Reads the calls and emails on every open deal, looking for unresolved objections, a decision-maker going quiet, competitor mentions, timeline slippage. Scores each deal red, yellow or green with its reasoning on the record, then sends the five to focus on this week.
A forecast you can trust. Deals about to slip surface before they do.
Meeting prep
Daily schedule, evening
Reads tomorrow's calendar the night before, links each meeting to its Attio record, and writes a brief: deal snapshot, recent context, flags, talking points, and the outcome to aim for. Lands as a Slack DM and a note on the record.
Every meeting prepped, every night, without anyone remembering to do it.
Post-call action items
Per call, event-triggered
When a recording finishes, it finds the right record, pulls out every commitment made on the call, and creates owner-tagged tasks due in three days. Plus a recap note: decisions, commitments, next steps.
Nothing said on a call falls through. The follow-up list builds itself.
Sales coaching
Weekly schedule
Compares each rep's calls against your playbook and your closed-won patterns, then writes them a page: what worked, what to try next, the top three actions. Reps get their own page. The manager gets an anonymized rollup.
Gong-style coaching without the Gong bill. Coaching, not surveillance, and no rep ranking.
Content mining
Weekly schedule, Friday
Reads your call transcripts for content seeds: pain quotes, objections, before-and-after stories, customer vocabulary, new use cases. Each becomes a tagged row in a Notion backlog, with the top five posted to Slack on Friday.
A backlog that fills itself from real customer language. Nobody writes from a blank page.
Operations agents
The six above run your sales motion. These three run the workspace itself. All are shipped and working, available as standalone builds, and none edits a record without a deliberate separate step.
Reporting
Weekly / monthly / quarterly
Computes the numbers a dashboard cannot and writes them into a narrative report: revenue forecast by quarter, conversion and median time in stage, an at-risk list weighted by revenue, a renewal calendar. Optionally a hosted dashboard on your own Vercel. It reads and computes, never edits.
The reports your team rebuilds by hand each month, every number traced to a query and thin data flagged rather than fudged.
How it works →Backup
Nightly schedule
Attio has no user-facing record restore. Every night this exports the whole workspace to your own storage as CSV, with cross-reference columns so relationships survive a re-import, keeping a rolling five weeks. Read-only key, cents a month.
A recent, restorable copy on your own storage. The thing you will be glad you had after a bad bulk edit.
How it works →Import deduper
On demand, before every import
Attio's importer creates a new record when a row's format does not match, so one trailing space becomes a duplicate. This normalizes the CSV, removes duplicates inside it, then verifies every remaining row against the live workspace before calling anything a match.
Only genuinely new, clean rows reach an import. No more climbing back to tens of thousands of half-empty duplicates.
How it works →How we build them
1
We scope the agents
We find the ones with the highest impact for your workflow. Most teams start with pipeline hygiene and deal focus.
2
We wire them up
Each agent is a Claude Code skill on your own Claude Code, connected to Attio through the Attio MCP and to Notion, Slack, Gmail or Calendar as needed. No server to host, no API key to manage.
3
You review outputs
Nothing irreversible happens without a human. Mass updates, merges and sent messages go through approval first.
4
We tune them after handoff
The sprint includes 30 days of tuning as your process reveals edge cases. The operating layer keeps them sharp after that.
FAQ
What does it actually cost to run an agent?
The agents run as skills on your own Claude Code subscription, so there is no separate runtime to pay for and no usage-based bill from us. The cost is the one-time build, plus an optional monthly retainer if you want us watching and tuning the agents.
Do agents need a separate AI subscription?
No separate Anthropic API key and no hosted service. Each agent is a Claude Code skill that runs on your own Claude Code subscription, triggered by /schedule or by a call event. You own the runtime, the data, and the schedule. If we disappear tomorrow, the agents keep running.
What can agents do in Attio?
Read and write records, create tasks and notes, update fields, search emails, meetings, and call recordings. They cannot create new attributes or objects, and they cannot delete records. Those actions stay with a human.
How many agents are included in the sprint?
Six agents, scoped to your specific use cases. We identify them on days 1-2 and build on days 3-5.
Can we add agents after the build?
Yes. Additional agents can be scoped and built as standalone projects, or through the operating layer if you want a monthly engagement.
What if our process changes and the agents break?
The sprint includes a 30-day tuneup for exactly this reason. For longer-term coverage, the operating layer includes tuning as part of the monthly engagement.
Ready when you are.
Book a call and we will tell you honestly whether this is worth doing, or start with the free 48-hour audit and decide afterwards.