How to set up Attio AI agents without code

Short answer: Attio AI agents are background workflows that read and write CRM records on a trigger or a schedule. You set them up without writing code by combining an Attio workspace, a Claude prompt or a small Claude Code skill, and an orchestrator: Attio Workflows for native triggers, Make or n8n for cross-tool steps, or a scheduled Claude Code skill for the simplest path. We ship six agents on every Sprint at Craftt. The full setup typically takes 2-4 hours per agent for someone with no coding background, and runs at $2-$5/day in token and automation costs across all six.

What an Attio AI agent actually is

An Attio AI agent is a small automation with three parts:

1. A trigger. Either a Workspace event (record created, field changed, meeting ended) or a schedule (daily 7am, weekly Monday).
2. A read step. Pull the relevant Attio records: a deal, a list, a person, a recent meeting transcript.
3. A think-and-write step. Ask Claude to do something with that data: summarize, classify, draft, score, deduplicate, and then write the result back as a note, a task, a comment, an update, or a Slack message.

That's the entire shape. Everything else is variation in trigger, scope, or output target.

The reason this works without code is that Attio's MCP server (covered in our Attio MCP guide) gives Claude direct read-and-write access to the workspace with no API key, no SDK, and no custom integration. Claude does the thinking. The orchestrator decides when. Attio holds the data.

The four orchestration options

You have four ways to schedule an Attio agent. Pick the one that fits the trigger and the team's stack.

Option 1: Attio Workflows (native)

Best for: anything triggered by a workspace event (new record, field update, list assignment, status change).

Attio has a native Workflows builder. Triggers include record created, attribute updated, list entry added or removed, and incoming webhook. Actions include AI-classify, AI-summarize, create record, update record, send Slack, send email, and call HTTP.

When this is enough: a single workspace, no external tools, simple if-then logic.

When it stops being enough: cross-tool steps (read from Gmail, write to Attio), multi-step branching, or anything that needs Claude to make a judgment call across more than one record.

Option 2: Make or n8n (cross-tool orchestration)

Best for: agents that pull data from outside Attio (Gmail, Calendar, Slack, webhooks) or push to multiple destinations (Attio + Slack + Notion).

Make is a hosted no-code automation tool with hundreds of integrations. n8n is the self-hosted equivalent. Both expose Attio as a connector and Claude as a connector. You drag triggers and actions onto a canvas, wire them with a flow, and ship.

When this is enough: any scheduled or event-driven agent with 2-5 steps and a defined branching logic.

When it stops being enough: when the agent needs to make multiple sequential decisions, hold state across runs, or iterate over hundreds of records with judgment calls per record.

Option 3: Claude Code skills (scheduled agents)

Best for: complex agents that iterate over many records, need long context, or do work that resembles "a human analyst goes through each open deal and writes a note."

Claude Code is Anthropic's CLI. Skills are reusable prompt-plus-tool packages that Claude Code can invoke. You can schedule a skill to run on a cron (daily 7am, Monday 9am) and have it use Attio's MCP server to read and write directly.

This is the orchestration layer we use for the harder agents at Craftt. There's no Python, no API key, no infrastructure. Claude Code IS the runtime. You write a Markdown skill file describing the job, and a /schedule command schedules it.

When this is enough: anywhere you'd describe the agent as "think through the full pipeline and write something thoughtful per deal."

When it stops being enough: when the work needs sub-second latency, real-time UI, or strict deterministic logic with no LLM judgment.

Option 4: Mix and match

Most production agent setups use two or three of these. A typical pattern: Attio Workflow catches the trigger, calls a webhook to Make, Make assembles context, hands off to a Claude Code skill for the thinking, and the skill writes back to Attio via MCP.

The six agents we ship on every Sprint

Each of these can be set up in a Sprint week or built standalone after the workspace is live. Below is what each one does, how it's triggered, and which orchestration option fits best.

Pipeline Hygiene (daily, 7am)

Reads every open deal. Flags duplicates for human merge review. Advances stages on deals with confirmed activity. Surfaces deals with no recent activity for a check-in task. Cleans empty required fields.

Orchestration: Claude Code skill on a daily schedule, reading via MCP.

Cost: ~$0.20/day for a 200-deal pipeline. Roughly $6/month.

Setup work: a 1-page prompt describing your stage definitions, your "stale" threshold, and what counts as legitimate activity. About 1 hour with a customer to nail down rules.

Deal Focus (weekly, Monday 6am)

Reads call recordings and email history for every open deal in the last week. Scores deal health red, yellow, green, with one sentence of why. Surfaces the five highest-leverage deals to focus on this Monday. Drops the list as a Slack message and an Attio note on each deal.

Orchestration: Claude Code skill, weekly schedule. Reads calls via Attio MCP, writes notes back via MCP.

Cost: ~$1/run with a 100-deal pipeline. Once a week. Roughly $4-5/month.

Setup work: define what "deal health" means for your business (we walk through this in the Sprint). The prompt then turns into a few hundred words of judgment criteria. About 2 hours per customer.

Meeting Prep (event: 18:00 evening before)

Reads tomorrow's calendar. For each meeting, pulls the company and contact records from Attio, recent emails and notes, open tasks for the attendees, and the last meeting transcript if there is one. Writes a 1-page prep brief: snapshot, context, suggested talking points, the outcome to aim for.

Orchestration: Make or n8n (needs to read Google Calendar and Gmail), Claude as the writing step, Attio as the data source and the output target.

Cost: ~$0.10 per meeting. Most teams run this on 5-15 meetings/day = $15-50/month.

Setup work: the connector setup (Google account auth, Slack channel for delivery) takes about 30 minutes; the prompt is ~150 lines and is mostly the same across customers. ~1.5 hours per customer.

Post-Call Action Items (event: call ends)

Reads the call transcript and any notes the rep typed. Extracts every commitment made on the call ("I'll send the proposal by Friday", "we'll loop in finance next week", "send case studies by Monday"). Creates Attio tasks for each, tagged to the right owner. Writes a recap note attached to the deal.

Orchestration: Attio Workflow trigger (call completed) → webhook to Make → Claude reads transcript → MCP writes tasks back to Attio.

Cost: ~$0.05 per call. For a 20-call-per-day team: ~$30/month.

Setup work: define what counts as a commitment (specific phrasing, required components) and who tasks default to if owner isn't named. ~1 hour per customer.

Sales Coaching (weekly, Friday 4pm)

Compares each rep's calls this week against your sales playbook and against closed-won patterns from the last 90 days. Writes one page of feedback per rep: what they're doing well, where they're drifting, one specific suggestion. Drops an anonymized rollup for the manager that surfaces team-wide patterns.

Orchestration: Claude Code skill, weekly schedule. Heavy MCP read activity (10-50 calls per rep), then a long-form write.

Cost: ~$2/run for a 5-rep team. Roughly $8-10/month.

Setup work: you need an actual playbook document and at least 5 closed-won deals to anchor on. About 2-3 hours including the playbook codification.

Content Mining (weekly, Friday)

Reads every call transcript from the week. Extracts pain quotes, objections, customer vocabulary, and use-case seeds. Drops them into a Notion backlog tagged by theme. Optionally drafts 3 short LinkedIn posts from the strongest material.

Orchestration: Claude Code skill, weekly schedule. MCP reads transcripts, Notion API writes the backlog, optional Slack notification.

Cost: ~$1.50/run. Roughly $6/month.

Setup work: define theme tags, the Notion database schema, and the LinkedIn voice. ~1.5 hours per customer.

The setup pattern that works for every agent

After ~50 agents shipped, the pattern is the same every time:

1. Define the trigger precisely. "Daily" is not enough. Pick 7:00am UTC, Tuesday-Friday only, and write that down.
2. Write the read step as a single sentence. "Pull every deal with stage != Closed and last_activity_at < 14 days ago." Vague reads produce vague outputs.
3. Write the judgment criteria as prose, not bullets. Claude understands rules better when they're explained, not enumerated. Tell it the goal, the constraints, and what the right answer looks like.
4. Specify the write target exactly. "Create a Task on the deal with owner = deal owner, title = 'Check in: ', due = end of next business day." Don't leave Claude to guess where to put the output.
5. Add a guardrail. "If you are not confident, leave the field blank and flag for human review" beats forcing a low-confidence write every time.
6. Run for a week in dry-run mode. Before any writes go live, have the agent write its proposed outputs to a Notion page or a Slack channel instead of Attio. Read what it produces. Adjust the prompt. Then turn on writes.

The reason these agents work without code is that every component of the pattern above is plain English. The orchestrator handles the plumbing. Claude handles the judgment. Attio holds the data. You're not engineering — you're writing a job description.

Cost math

The headline number for a customer running all six agents:

What this replaces, in our customers' words: one part-time SDR or RevOps coordinator. The agents don't replace a human — they remove the work the human shouldn't have been doing. The team gets one cleaner pipeline, better-prepared meetings, and follow-ups that actually happen.

What the no-code path looks like in practice

If you wanted to ship Pipeline Hygiene tomorrow with no engineer:

1. Install Claude Code on your machine (15 minutes, one Homebrew command).
2. Set up the Attio MCP connector (5 minutes via OAuth).
3. Drop a pipeline-hygiene.md skill file in ~/.claude/skills/ describing what the agent should do.
4. Run /schedule pipeline-hygiene and pick a daily time.
5. Let it run for a week in dry-run mode (the skill writes proposed actions to a Slack channel, not Attio).
6. Review, adjust the prompt, then flip the "write to Attio" switch.

That's it. No infrastructure to provision, no API keys to manage, no Python to maintain.

We package this whole sequence as the AI-Native Attio Sprint at Craftt: 7 days of focused work, six agents tuned to your specific process, and a 30-day tuneup after handoff. If you'd rather build them yourself, our Attio MCP guide covers the MCP-side setup in full.

What can go wrong

Three failure modes show up in every implementation:

1. Vague triggers. "Daily" or "when a deal changes" without specificity produces an agent that fires too often or at the wrong time. The fix is in step 1 of the pattern above.

2. No guardrails on writes. The first version of any agent should overwrite nothing. It should propose, not act. Writes get turned on after a week of clean dry-runs.

3. Drift over time. A working agent stops working when your sales process changes and the prompt doesn't. Build a Friday check into someone's calendar to read the agent outputs once a week for the first month.

The ongoing CRM manager service exists for exactly this reason.

Free audit of your Attio workspace

If you want a written review of whether your workspace is agent-ready (and which agents would have the most leverage on your team), we run a free 48-hour audit. No call, no pitch, 5 slots a week.

Or if you want the full six agents set up for you, the AI-Native Attio Sprint is the path: 7-day flat-priced build, $2,997, six agents tuned to your team, 30-day tuneup included.