You give a request to the lead agent. It hands the Google Ads, Meta, page and analytics work to specialists, and each one does its part in the same editors your team uses. What comes back is a set of drafts you can open, check and change, each with a note on why it was built that way.
Nørgaard VVS opens in Odense next month. Set up search and a landing page like we did for København.
The lead agent plans the work but doesn't edit campaigns or pages itself. It passes each job to the specialist that owns that surface. There is only ever one writer per surface per client, so two runs can never change the same campaign at the same time.
Keyword research, campaigns, ad groups, keywords, responsive search ads, negatives and budgets.
Campaigns, ad sets, ads, creatives and audiences, including every combination of copy and image you want to test.
Sites, pages, sections, copy and test variants, written into the page editor.
Sessions, sources, goals, search terms and ad performance. It reports what it finds to the others and changes nothing itself.
Called as a tool by the page builder and Meta Ads. Every image it makes is saved to the client's asset library.
Create an agent with its own instructions, documents, skills and model, then choose which specialists are allowed to call it. A custom agent is consulted and gives an answer back. The specialist that owns the campaign or page still makes the change, so your reviewer never ends up rewriting the ads itself.
Give it your scoring rubric and have the specialists check their work with it before anything is staged for you.
Holds a client's tone and approved phrases, and gets asked before ad copy or page text is written.
Knows what matters in one industry, like regulated claims for clinics or authorisation numbers for electricians, and flags what's missing.
Your knowledge base is one library of folders and files: voice, offers, account structures, bidding rules, page patterns. Instead of pasting everything into every prompt, you grant each agent the parts it needs, either always included or fetched when it's relevant.
The reviewer reads the rubric. The Google Ads specialist reads the bidding rules. Neither has to wade through the other's documents to find what matters.
| Document | Meta | Pages | Reviewer | |
|---|---|---|---|---|
| How we writeAgency | Always | Always | Always | On request |
| VVS search structureAgency | Always | Off | Off | On request |
| Bidding & budget SOPAgency | Always | On request | Off | Off |
| Landing page patternsAgency | Off | Off | Always | On request |
| VVS review rubricAgency | Off | Off | Off | Always |
| Odense offer sheetClient file | On request | On request | Always | Off |
Documents, agent settings and custom agents made at agency level apply to all client accounts without copying anything.
A client account can add its own offer sheet, disable a document that doesn't fit, or replace one outright.
Client rules are added on top of the agency instructions, or replace them when an account needs something completely different.
Autonomy is a setting on each agent, not something guessed from how a message was worded. On the same client, the Google Ads specialist can clean up search terms automatically while the page builder waits for approval on every change.
Your limits sit underneath all of it. They are checked by the platform when a change is written, so a change that breaks one is rejected whatever the agent decided.
It stays here until someone decides, even if you close the tab.
Limits
Checked by the platform on every write, for agents, schedules and API calls alike.
The agent acts when someone asks and does the task it was given, nothing more.
It reads performance and tells you what to pause, exclude, rewrite or test, without changing anything.
It prepares the changes and stages them. Nothing goes live until someone on your team approves.
It makes the changes itself, inside your limits, and writes the reason for each one in the log.
A trigger starts a named agent with a prompt you wrote. It can run every Monday morning, every second Tuesday, or the moment a page passes 500 visits or a campaign passes a spend threshold.
Each run leaves a thread you can open later, with anything that needs approval waiting inside it. The monthly client report works the same way. How reports are written
Every Monday at 07:00 Google Ads
"Go through last week's search terms and add negatives for anything unrelated to the services we sell."
Last run Mon 07:04 · 18 negatives addedEvery second Tuesday Meta Ads
"Draft three new angles from the current offer sheet and set them up as a test."
When /akut-odense passes 500 visits Page builder
"Start an A/B test on the hero headline using the second angle from the brief."
First weekday of the month, 08:00 Lead agent
"Write the monthly report for Nørgaard VVS from the change log and send me the draft."
The agents, documents, approvals and limits on this page are also available over REST and MCP, so your intake forms, ops tools and your own agents can start the work without anyone opening a tab.
Start a named agent on a named client from your own software, then poll for the result or get a webhook when the drafts are ready.
Another model sends a brief to the lead agent or a specialist. The work runs here, in that client's workspace, under that client's rules.
Connect an agent you already run and give it the tool set, with the client's documents and limits attached. It calls the tools and we carry them out.
Every write made this way is logged with the API key or connection that made it.
Google and Meta on one screen, with an editor that opens the whole account and publishes only what changed.
See the ads workspace → WebsitesFull sites, landing pages and campaign subdomains that the agent builds and your team edits visually.
See the website editor → ReportsClient reports written from the account's change log, in your agency's voice, on a schedule you set.
See how reports work →Start a 14-day trial, connect Google Ads and Meta, and ask the lead agent for something you'd normally spend an afternoon on.