Agents that do the weekly work you were about to hire for
Marketing, QA, customer follow-up and finance sync, done by Claude agents on a schedule. We find the work worth automating, build the agents on your accounts and keep them working every month.
- 25
- Claude agents shipped into client workspaces
- 3×
- a week our SEO agent publishes on our own products
- 24×
- a day our ops agent checks our own products' health
- $0
- of servers hosted by us: agents run on your accounts
Short answer
Craftt builds and runs Claude agents for companies of 20 to 100 people: marketing, QA, customer follow-up and finance sync. A one-week AI audit is $1,500, each agent is from $3,000 to build, and monthly operation is from $1,500. Agents run on your own Claude, GitHub and cloud accounts, and you keep them if you leave.
What the agents do
Agents marked running work today on our own products or for clients. Agents marked built per client we build to your setup from the audit.
Marketing
Site improvement
Built per clientVisits your site several times a day, finds broken links, weak copy and slow pages, and opens a pull request in your repo with the fix. You merge or close it.
SEO writer
RunningPicks the next keyword from your plan, writes the article in your voice, checks it against your style rules and publishes on a schedule.
AI visibility check
RunningEvery week asks AI search the questions your buyers ask and records whether you are named, who is named instead, and which page it cited.
Content from calls
RunningReads your call transcripts for pain quotes, objections and customer vocabulary, and turns them into a backlog of post drafts.
Product and QA
Weekly QA
Built per clientReads every change merged this week, writes test cases for each, runs them against staging and reports what broke, with steps to reproduce. Built for teams where a human checks a few changes out of a hundred.
Health monitor
RunningChecks your app, crons and integrations every hour and sends one message when something stops, with what failed and since when.
Release notes
Built per clientTurns merged pull requests into a customer-facing changelog and a short internal summary for sales and support.
Customers and finance
Activation and churn watch
RunningEvery day finds users who signed up and stalled, or stopped using the product, and drafts a personal email for each. You approve before anything is sent.
Accounting sync
RunningKeeps invoices and payments from Xero or QuickBooks on the right customer record, so sales sees who has paid without asking finance.
Quotes and RFQs
RunningBuilds a quote or supplier request from the specs on a record and leaves it as an email draft for each recipient.
Sales and CRM agents live on a separate page: Attio agents.
Three steps, three prices
AI audit
$1,500
One week, paid upfront
We interview the people who do the repeating work, list every task that recurs weekly, and pick the three agents that would save the most hours. You get the list with the hours per agent and what each one needs access to.
Agent build
From $3,000
Per agent, 50% upfront
One agent built, tested on your real data and handed over with written instructions. It runs on your Claude subscription, in your GitHub or your cloud account. You own the code and the prompts.
AI operations
From $1,500/mo
Month to month
Agents break when your site, product or process changes. Every month we monitor every run, fix what broke, retune prompts and add one new skill. You get a short report each week: what the agents did and what they got wrong.
The audit is yours to keep. Build the agents with us, with someone else or yourself.
How it runs
1. Audit the repeating work
A week of interviews and access to the tools where the work happens. The output is a ranked list of agents with hours and cost per agent, not a strategy deck.
2. Build the first agent
We start with the agent that saves the most hours. It runs in draft mode first: it proposes, a person approves. It writes on its own only after you have seen two weeks of output.
3. Run and extend
On the monthly plan we watch every run, fix breakages within one business day, and add the next agent from the audit list when the first one is stable.
FAQ
How is this different from your Attio agents?
The Attio agents on /agents work inside the CRM: pipeline hygiene, deal focus, meeting prep. This service covers the rest of the company: marketing, QA, customer success and finance. Many clients start with the CRM and add these once the CRM is clean.
Who is this for?
Companies of 20 to 100 people where someone spends hours every week on work that repeats: writing content, checking releases, chasing stalled users, moving numbers between systems. If the work happens once a quarter, an agent does not pay back.
Where do the agents run?
On your accounts: your Claude subscription, your GitHub, your cloud. We never host your agents or your data on our servers. If you stop working with us, the agents keep running and the code stays yours.
What does it cost to run an agent?
Most agents run on the Claude subscription you already pay for, so there is no separate AI bill. Agents that call the API directly cost a few dollars a month; the audit names the running cost of each agent before you decide.
Can an agent send emails or change things on its own?
Only after you have approved two weeks of its drafts. Until then every agent proposes and a person decides. Anything irreversible, like sending to a customer or deleting data, stays behind an approval for as long as you want.
Why a monthly plan and not a one-off build?
An agent built once and left alone breaks within months: a site changes, an API changes, a prompt starts drifting. The monthly plan is what keeps the agent doing the work. You can cancel any month and keep everything built so far.
Why is the audit paid upfront?
It is a week of interviews and analysis, and you keep the result even if you build the agents yourself or with someone else.
A 30-minute call, or a free audit in 48 hours.
On the call we tell you whether Attio is worth it for your team. The audit is a written list of what to fix in a workspace you already have. Both are free, and you decide afterwards.