A custom agent is a role you set up once that then works on campaigns for you: it comes up with ideas, plans them, and can take them live and review them through your connected tools. You decide which channel and tools it can touch, how often it runs, and whether it acts on its own or checks with you first.
Before you start
- To let an agent use a tool beyond the built-in campaign tools, connect that integration first in Settings, under Integrations.
Create an agent
- Open Agents from the left-hand sidebar.
- Select New agent.
- In Name, give it something you’ll recognise, such as “Social campaigns” or “SEO”.
- Optional: in Instructions, describe how it should approach its campaigns. The agent follows these every time it runs.
- Optional: pick a Channel from the built-in list to focus the agent on.
- The built-in Growth Method campaign tools are always on for every agent. Under Integrations, tick each integration the agent may use; a ticked integration gives the agent all of its tools. The list only offers tools your team has enabled. If your team has many integrations, switch between Enabled and All to show the agent’s integrations or the whole list.
- Optional: select an integration’s row to open its tool list and tick individual tools. All tools at the top ticks or clears the whole list, and a dash on an integration’s tick box means the agent has some of its tools, not all. The token cost next to each tool shows how much of the AI’s working memory it takes up on every run, so trim the tools an agent doesn’t need and it stays focused. Leave everything ticked and the agent gets every tool, including any the integration adds later.
- Set Runs to how often the agent should run: On demand, Every 6 hours, Twice daily, Daily, Weekly, or Monthly. Scheduled runs happen on the hour, UK time: daily at 6am, weekly on Mondays at 6am, monthly on the first of the month at 6am, twice daily at 6am and 6pm, and every 6 hours at midnight, 6am, noon, and 6pm.
- Choose its Autonomy (see below).
- Select Create agent.
Your agent appears in the list with a Custom badge, alongside Ada, the built-in chat agent. The Runs column shows how each agent runs at the moment: its schedule, On demand if it has none, or Off if you’ve switched it off. Autonomous agents also show a lightning-bolt icon next to their name, so you can see at a glance which ones run on their own. The built-in campaign agents that power each stage stay out of the way by default; turn on Show campaign agents above the list to see them.
How it runs your campaigns
Every campaign moves through the same stages, in the same order, every time:
Backlog → Planning → Live → Analysing → Complete.
These stages are fixed. An agent decides what happens within a stage, the plan, the copy, the results that matter, but it never chooses the order and never invents a step. Growth Method moves a campaign to the next stage in code, not by asking the AI, and only ever forward. It won’t run a campaign through the same stage twice, and an agent can’t skip ahead or loop back. This is the same track your team already uses for campaigns you run by hand.
That’s what makes an agent safe to leave running. However it plans or words a campaign, it can only move it along a track you already trust, one stage at a time and only forward. Even an autonomous agent acting in your connected tools can’t loop, skip ahead, or run a campaign through the same stage twice.
How far an agent takes a campaign
On a schedule, how far an agent takes a campaign depends on its autonomy:
- A Human in the loop agent drafts a campaign idea and stops. It lands in your Backlog for you to review. If, while researching the idea, it reaches a step in a connected tool that needs approval, it pauses and emails your team a link to the Approve step; approve it and the run carries on.
- An Autonomous agent writes the plan and moves the campaign to Planning. If you have accepted that it can act in your connected tools without review (see below), it never stops to ask at any step: it carries on and takes the campaign Live on its own, working through the plan with your connected tools as it goes.
An autonomous agent only takes a campaign live when every step can be done without you. If a step needs a person, something only you can do such as briefing a colleague, or if a tool action fails, it stops at Planning and leaves a note of what is left, rather than launching a half-finished campaign. From Live, a campaign moves on to Analysing by itself once it has run for its set duration.
When that happens, the same agent reviews the results on its own: it pulls the numbers from your connected analytics tools, weighs them against the campaign’s hypothesis, and writes its findings to the campaign. The campaign owner gets an email when the analysis is ready, and the run’s conversation records every step, so the whole campaign, from idea to verdict, is on the record. If the review reaches a step that needs approval (the campaign was created by you, or by an agent that is human in the loop), it pauses and the owner’s email says so, with a link to the paused conversation.
When you work with an agent in chat, you stay in the loop whatever its autonomy: you take each next step, you approve each action that touches a connected tool, and you decide when the campaign goes live. An agent moves a campaign forward on its own only during a scheduled run or Run now. When it reviews a campaign it waits for real results, so it won’t rush to a verdict or invent early numbers. It reviews a thirty-day test after thirty days, not on day one.
Tip: set Runs to On demand while you’re trying an agent out, then switch it to a schedule once you’re happy with how it works.
Choose how much it does on its own
Autonomy controls whether the agent acts by itself or checks with you first:
- Human in the loop: the agent does the work and a person approves each action in a connected tool. It publishes nothing without your say-so. On a schedule, it leaves a draft in your Backlog, and emails your team if it pauses for an approval along the way.
- Autonomous: the agent acts without stopping to ask, including using your connected tools. On a schedule, once you have accepted the responsibility below, it takes a campaign all the way to Live on its own and never waits for an approval on any step of its own runs.
An autonomous agent needs at least one connected tool to act with, so Autonomous is only available once you’ve ticked an integration. When you set it, Growth Method asks you to confirm you understand the agent can launch campaigns and change or delete data in those tools without review. You won’t be able to save the agent until you tick the box to accept. Every later change to an autonomous agent asks for the same acceptance again, so whoever makes a change takes responsibility for it. Growth Method records who last accepted and when, and shows that on the agent. If the agent’s last connected tool is later disconnected, it stops at Planning on a schedule until a tool is connected again.
Whichever you choose, an agent can’t run away. A single turn is capped at a fixed number of steps, it can only move a campaign one stage forward, and it can’t repeat a stage, so it does the work in front of it and stops. On a schedule, an autonomous agent goes live only when every step can be done without you; if a step needs a person or a tool action fails, it stops at Planning and tells you what is left.
Start with Human in the loop while you build trust in an agent, then move to Autonomous once you’re comfortable.
Warning: an autonomous agent uses your connected tools, and can launch campaigns, without asking first. Keep it on Human in the loop until you’ve watched a few runs in chat and you’re happy with its work.
Switch an agent off
To stand an agent down without losing its setup, switch it off. Select Edit on the agent and use the switch at the top of the panel, then select Save changes.
While an agent is off:
- It doesn’t run on its schedule, and Run now is unavailable.
- It stops acting in your connected tools without review. When a campaign it launched earlier finishes, the results review waits for a person instead of running on its own.
- Its row in the agents list is dimmed, reads Off under Runs, and drops below the agents that are on.
- You can still chat with it, where you stay in control of every step.
Switch it back on at any time: its schedule, autonomy, and recorded acceptance are all kept, so it carries on exactly as before.
This is stronger than setting Runs to On demand. That only stops the schedule: an autonomous agent on On demand still reviews its live campaigns on its own when they finish. Switching the agent off pauses that too.
See what your agent did
The Agents page shows each agent’s latest run outcome next to its last run time: the campaign it created, No campaign when it decided not to create one, or Failed, in amber, when the run hit a problem. A run cut off before it could finish shows Stopped early; open its thread and ask the agent to carry on where it left off. Each outcome links through to the campaign or the run’s conversation, so you can see what happened overnight in one click. A run still working through its steps shows No reply yet.
Every run happens in a conversation, and that conversation is the record. Each phase of a campaign gets its own: one for creating it, one for planning, one for launching, one for reviewing the results, each named for the campaign and the phase. Within each one, the agent’s actions and the results it got back are written into the thread, in order. Nothing it does is hidden, and every detail in these records comes from what the system observed as the agent worked, never from the agent describing its own work. Even a run that is interrupted and never finishes keeps its record: the thread shows what the agent was asked to do and notes that the run stopped early.
When you open a conversation linked to a campaign, the campaign’s earlier conversations appear above it, read only, so the whole campaign still reads as one story from idea to verdict. A short note marks where the history ends and the live conversation begins. The agent in the live conversation can’t see those earlier messages, so if you need it to build on a past phase, say so in your message.
The campaign page itself carries the same record at a glance. Its Agent activity list shows one line per conversation on the campaign: the phase, the date, whether it acted without review, how many queries it made and which data sources it consulted, with a per-source count for recent runs. Each line links to the full conversation. The same detail appears on the dividers when a campaign’s history is shown above a chat. Shared campaign pages carry a quieter version of this record: a note under each stage shows whether the AI acted without review, how many queries it made and which data sources it consulted, without links into the conversations.
The thread also labels work the agent did on its own. When an autonomous agent acts in a connected tool without review, the thread shows the same detail an approval card would have shown, marked Ran without review. For every action, you can see whether a person approved it or the agent acted alone. If you share a conversation, the shared page notes the work was done autonomously by an AI agent, while the detail of individual tool changes stays private to your team.
Anyone on your team can open a run’s conversation, even when the run created nothing. A run that declines to create a campaign, for example because everything it found was too close to a campaign you already have, explains its decision in that thread. You can reopen it at any time to see exactly what the agent did and why, long after the run has finished. It’s the same thread you watch when you test an agent in chat.
Who is Ada Lovelace?
Ada Lovelace is the automation account that runs your campaigns when no one is at the keyboard. You’ll first meet her in your team’s member list, in Settings under Users, even though no one invited her. She belongs to every team by default.
A few things to know about her:
- She isn’t a person, and she never signs in.
- When you set up an agent with a schedule, Ada creates new campaign ideas for you automatically on that cadence.
- When an agent runs on a schedule, it runs as Ada, and she’s credited as the author of any campaign it creates.
- She owns the campaigns she creates only until you take over. Move one of her campaigns to a new stage and it becomes yours. A campaign you already own stays yours unless you reassign it.
- She doesn’t direct the other agents. Every role runs either as you, when you work in chat, or as Ada, when it runs on a schedule. The steps and their order are the same both ways.
The name is a nod to Ada Lovelace, the 19th-century mathematician often regarded as the first computer programmer. Working with Charles Babbage on his Analytical Engine, an early mechanical computer, she wrote what many consider the first algorithm meant to run on a machine, and she saw what others missed: that such a machine could do far more than arithmetic. We named our automation account after her as a small tribute to that mix of method and imagination, the same qualities we built Growth Method around.
Try an agent out
You don’t have to wait for a scheduled run to see what an agent does. Open Agents from the sidebar and select Edit on your agent. From there you have two ways to test it.
Start a chat opens a conversation with the agent so you can watch it work through a campaign, step by step. You see each step as it happens, and you set the pace: even an Autonomous agent waits for you to move the campaign to its next stage, and with a Human in the loop agent you approve each action that touches a connected tool before it runs.
Run now starts a run immediately, exactly as a scheduled run would, so you can see what the schedule will do without waiting. Watch the agent’s row in the list: its latest run outcome appears next to its last run time, linking to the campaign it created or to the conversation showing its reasoning. A Human in the loop agent drafts a campaign and leaves it in your Backlog; an Autonomous agent you’ve accepted responsibility for runs the full cycle and can take the campaign live in your connected tools, just as it would overnight.
Next to these buttons, the link icon copies a direct link to the agent’s panel. Send it to a teammate and the panel opens for them straight away.
Testing an agent, whether in chat or with Run now, doesn’t change its schedule. It still runs at its next scheduled time.
Next steps
- Connect more tools for your agent to use in Settings, under Integrations.
- Give the AI access to your data