Skip to main content
A conversation with Concord carries everything that happened in it: your messages, the agent’s replies, and the full result of every platform read, report, and file it worked with. That is what lets you say “extend it by two weeks” without naming the campaign again. It also means a conversation has a size limit. When a thread approaches it, Concord condenses what came before so the conversation can keep going instead of stopping.

​
What condensing does

Concord writes a summary of the thread so far and continues from that summary rather than from the full transcript. Some things are carried forward deliberately, because they are the parts that must not drift:
  • What you are trying to accomplish — the goal of the conversation
  • What has already been done — the decisions taken, the entities created or edited, the findings
  • What is still open — unresolved questions and work in progress
  • What comes next — the concrete remaining steps
  • The most recent exchanges, verbatim — the last part of the thread is not summarized at all
  • Resolved identifiers — the ids Concord looked up for you, such as a geography or an audience, are carried across exactly as they were resolved, never re-described
  • References to large results and files — so Concord still knows a report or a document exists rather than fetching it again
The Skills that are still relevant to the open work stay loaded. The ones that no longer apply are dropped. The running record of the conversation is not condensed. What Concord changed on your platforms, the rules and decisions you gave it, and your answers to its questions stay exactly as recorded. See Best practices.

​
What can drift

A summary is a rewrite. Every time a conversation is condensed, detail that was in the transcript is expressed more briefly — and a detail expressed more briefly can come back slightly reformulated, or not come back at all. One condensation in a long thread is usually invisible. Ten is not. The more times a single conversation has been condensed, the more likely it is that a detail of the discussion, such as a figure you quoted or a nuance in your reasoning, comes back imprecise. Explicit rules and decisions are protected by the record; detail around them is not.
If an answer late in a very long conversation contradicts something established early in it, condensing is the first thing to suspect. Start a new conversation and re-state the constraint rather than arguing with the thread.

​
How to keep answers sharp

One task, one conversation. This is the single most effective habit. When you move from building a campaign to pulling a report, or from one advertiser to another, start a new conversation. It costs nothing and it keeps every thread short. Keep research and execution apart. Long exploratory work — comparing options, digging through a spreadsheet, resolving targeting — fills a conversation quickly. Once you know what you want to build, start a fresh conversation and state it. Put durable rules in a Skill, not in the thread. A constraint you stated in this conversation applies to this conversation only. The same constraint as a Skill is loaded whenever it is relevant, in every conversation. Attach files rather than pasting long content. An attached brief is read when it is needed and referenced afterwards. The same brief pasted as text sits in the transcript and counts against the conversation’s size.

​
Finding your way back

Nothing is deleted. Condensing changes what Concord carries forward in the conversation, not what is stored: the full thread stays readable above, and past conversations stay in the sidebar when you need what was said in them. Work that produced something durable is unaffected either way — a submitted change is in Action history, and a change applied on a platform stays on the platform. Neither depends on the conversation that produced it. Working in conversations · Creating Skills