Solutions · Marketing
It sounds like youbecause your voice is a file it reads first.
AI copy reads like AI copy because nothing told it how you write. In Machine, your voice, the claims you're allowed to make and the words you never use live in your repo as a skill. It's versioned, reviewed, and loaded into every session before a single word gets written.
Your voice is a file · your claims are a file · every draft is a diff
- Voice lives in
- A skill file in your repo
- Every session
- Reads it before it writes
- Lands as
- A draft in a change request
- Gets better by
- Editing the file, not re-prompting
01The handoff
Hand off the production line. Keep the ideas.
Strategy isn't the bottleneck. The bottleneck is that one positioning decision becomes nine assets, and by the third one the tone has drifted and nobody has time to notice.
Job 01
One decision, all nine assets
The launch note, the page section, the changelog entry, the social copy, the email. All written from the same source of truth in one session, so they agree with each other. That's usually where things break, not in the quality of any single piece.
Job 02
A content brief with the research done
What's already been published on the subject, what your own site says about it, and what you've said before that this would contradict. Your brief starts from what exists, not from a blank page.
Job 03
A claim audit across your own site
It reads every page and finds where the copy claims something the product doesn't do, or says the same thing three different ways. Nobody schedules this job and everybody needs it. When it's done, it's a diff.
Job 04
Competitive watch, with no made-up findings
A scheduled session reads what has publicly changed and reports what's genuinely new. When nothing is, it says so. A weekly digest that invents a headline every week gets muted by week five.
Job 05
One long piece, in every format it deserves
The talk becomes a post, the post becomes a thread, the thread becomes an email. Each one is written for where it will live, not the same text pasted four times with the line breaks moved.
Job 06
The campaign read-out
It pulls what ran and what happened, writes the honest version (including what didn't work), and drafts what it would change. Attribution stays your call. It presents the numbers; it doesn't decide what caused them.
02What comes back
A draft on a branch, reviewed as a diff.
Most marketing review means pasting a document into a comment thread and losing track of which version won. Here every draft is a commit, so the review is a diff and the history is real.
Scheduled sessions: launch note draft
- Voice
- skills/voice/SKILL.md
- Claims
- skills/claims/SKILL.md
- Status
- Draft · for review
- Flagged
- 2 claims, unverified
A trigger fires at 07:00. A session starts, does the work on its own machine, and opens a change request before anyone is at a desk. That’s the whole feature.
Two lines in this draft are flagged, not written: a throughput figure and a comparison with a competitor’s scheduling. Neither traces to a source in the repo, so the agent marked the claim instead of filling in the sentence.
Banned by the voice file, so absent here: “seamless”, “unlock”, “revolutionary”, and every superlative except the one form the file allows.
Your voice is code, so fixing it is a commit
When a draft comes back wrong, you edit the skill file, not the prompt. Every future session, in every project that loads that skill, picks up the fix. And the history shows exactly when the tone changed.
It flags what it can't back up
A claims file lists what you're allowed to say and what needs a source. When an agent hits an unsupported claim, it marks it instead of writing a confident sentence. It's the single most useful behaviour you can set up in marketing.
Review a diff, not a comment thread
Two drafts of the same page are two commits. What changed between them is a diff you can read in ten seconds, not a document with eleven suggestion bubbles and no clear current version.
03Where it reaches
Where your words live, and where they need to go.
Connect each source once for the whole project. The credential is decrypted on our side and attached to the outbound call. It never lands in the machine the model is driving.
Notion and Google Drive
The brief, the positioning doc, six months of decisions nobody wrote down anywhere else. It reads them as prior art before writing, then puts the finished draft back where your team looks.
GitHub
If your site is a repo, your copy is code. A copy change becomes a change request with a preview, reviewed just like a code change by whoever owns the words.
Your email and campaign tools
The marketing stack is wide, and most of it is in the Easy connect catalogue: pick the app, click through its OAuth screen, done. Set sending to Ask and nothing reaches a list without a person.
Google Sheets
The campaign tracker, the keyword sheet, the calendar your whole team actually edits. It reads them as input and writes the read-out back into them.
Slack
The one live channel. Ask for a variant in the thread where the launch is being debated, and the draft comes back to that thread as a file.
Easy connect covers 3,000+ apps through their own OAuth screens. If a tool isn't in the catalogue, you can reach it through OpenAPI, a Postman collection, GraphQL, raw HTTP or a remote MCP server. Same three-minute setup, one step less magic.
04How it runs
Let it draft now. Never let it send on its own.
Same session machinery, started three ways. In marketing, be most careful with the automated one, because the output is public.
01 · On demand
"Give me the four variants"
Ask in a thread. The session reads the source, the voice file and the claims file, then posts the drafts back to the same thread.
02 · Human-assisted
It stops before anything is published
Set publish and send actions to Ask. The run pauses at the call with the exact payload in front of you (the list, the subject, the body) and picks up from there when you approve.
03 · Automated
Research on a schedule. Publishing, never.
Run the competitive watch and the claim audit on a cron and let them land as change requests. Automating the reading is an easy win. Automating the publishing is how a wrong claim goes out at 03:00 with nobody awake.
05Control
Nothing published, nothing claimed, without a person.
In marketing, an agent's mistake is permanent and public. So here are the defaults, stated precisely rather than reassuringly.
01
Approval gates are off until you turn them on
Out of the box, Machine is permissive: an action runs unless you've said otherwise. In a marketing project, your first change is setting every publish and every send to Ask. It isn't done for you, and we'd rather say so.
02
Merge is default-deny
A draft lands on the session branch and reaches main through a change request. An agent can't merge unless an admin grants project.cr.merge in your project config, and widening that grant is itself a reviewed change. Nothing goes live just because a session decided it was done.
03
The claims file is what really holds the line
Permissions stop an agent from doing something. A claims skill stops it from saying something. Both live in the repo, and the second is what keeps an unverifiable number off your homepage.
04
Connector credentials never enter the machine
The sandbox holds one project-scoped Machine token and no third-party keys. Your campaign tool's credential is decrypted server-side and attached to the outbound request.
05
Every call it made, and who allowed it
The connector and exact action, the agent, the person or trigger behind the session, the outcome, and who released a held call. If something went out, the record says who let it.
06Other teams
Same machine. Every other team.
One project, one set of connectors, one memory that keeps growing. Each team writes the skills for its own work. Nobody builds a second system.
- SalesResearch, drafts and CRM cleanup, held for your yesOpen
- ProductFeedback turned into specs, with the evidence attachedOpen
- EngineeringReproduce it, patch it, open the change requestOpen
- FinanceThe close, the reconciliation and the variance note, with the working shownOpen
- PeopleScheduling, kits and onboarding, never the hiring decisionOpen
- ITRunbooks that run, and a platform that passes your reviewOpen
- Data ScienceA real machine, a real query, an analysis you can re-runOpen
- All solutionsEvery team on one page, plus the rules that never change between them.Open
Marketing
Put your voice in the repo.Let the production line run.
One governed platform for your marketing agents. Any model, your keys, every action on the record.
- One machine per session
- Lands as a change request
- Every action recorded