Most changes to a marketing account are never checked afterwards. Someone raises a budget, launches a campaign or rewrites an ad, and a week later nobody remembers whether it worked.

This gets worse when an AI agent makes the change. Your agent can launch a campaign in two minutes, and then the session ends, along with the only record of why the change was made and what it was supposed to achieve.

From today, when your own AI (Claude, ChatGPT, Cursor, Codex or the Cogny CLI) changes something through Cogny, Cogny keeps the record and follows up.

Launches become tickets on their own

On Free, Solo, Cogny AI and Cogny Pro, when your agent ships something through Cogny (a campaign, an ad, a post, a page, a published tag), Cogny opens a ticket for it without being asked. Everything one agent ships in the same stretch of work lands on one ticket, not one per tool call.

Cogny reads the impact three to seven days later, depending on what shipped, compares it with the numbers from before and writes the result on the ticket. You see it in the app, and your agent can read it with get_ticket.

If your agent then describes the work with create_ticket, its description is added to the same ticket instead of opening a second one.

Other changes: your agent asks once

Not every change is a launch. A budget increase, a paused keyword or a new bid is still worth measuring, but Cogny can't tell on its own which of those belong together. For those, Cogny prompts your agent to ask you once, at the end of the task. In Claude Code it reads like this:

Should I register this work with Cogny so it measures the results and follows up?

Say yes, and the agent registers one ticket for the whole task with:

What changed: the campaigns, budgets, keywords or posts, with their IDs.
The expected outcome: what the change should achieve, in numbers where possible. For example: "click-through rate of 4% or more, at least 10 signups under 500 SEK each in four weeks".

Say no, or ignore the question, and nothing is created. How closely an agent follows the prompt is up to the agent, so if yours doesn't ask, tell it to register the work.

Nothing gets lost if you skip it. cogny_status tells your agent how many of your changes from the last 14 days aren't on a ticket, list_unregistered_changes lists them, and create_ticket with action_log_ids attaches them to a ticket later.

On Cloud: work a ticket from your own AI

On Cloud, a team member can pick up a ticket in their own assistant with start_ticket, make the changes, and close it with complete_ticket. Every change their agent makes through Cogny in between is logged against that ticket, so the analysis knows exactly what changed. Changes made outside a ticket show up in list_unregistered_changes like on the other plans.

Why the agent asks first

Registering a ticket writes into your Cogny workspace, so for anything that isn't a launch the agent asks before it does it. You decide which changes are worth measuring: a one-day bid tweak might not be, a new always-on budget probably is.

The question comes at the end, alongside the summary of what the agent did, so nothing changes in how your agent works while the task is running.

Try it

Connect your agent: run npx @cogny/cli init in a coding agent, or add https://app.cogny.com/mcp as a connector in Claude or ChatGPT. The install guide covers both.
Ask it to make a real change in a connected platform.
Open Tickets in the app, or say yes when your agent offers to register the work.

What each plan can do over MCP is in the buying guide.