Building an agent in Cora is a conversation, not a one-shot exercise. You describe what you want, Cora builds it, you test it together, and then you refine it until it works the way you need. Most agents take two to five iterations to get right, and that is completely normal.

Think of it like pair programming with an AI. You bring the domain knowledge (what the workflow should do, who it is for, what "good" looks like) and Cora handles the implementation. You stay in the conversation the entire time, iterating until you are happy with the result.

If you are new to agents, start with Understanding Agents for foundational concepts.

There is no form to fill out or wizard to click through. You build agents by chatting with Cora in the regular chat interface. Just describe the workflow you have in mind, and Cora will ask clarifying questions to fill in the details.

Cora is context-aware. It knows which integrations you have connected, what data is available, and what lifecycle events are active in your workspace. It uses this context to make sure the agent it builds will actually work.

Let us walk through a concrete example. Say you want a weekly summary of account health sent to your team every Monday morning.

Open the Cora chat and describe your goal in plain language. For example:

"I want a weekly pipeline review sent to Slack every Monday at 9am."

Cora will walk you through a short series of questions to nail down the details. Expect questions like:

Cora offers multiple-choice options for most questions, so you can pick by number rather than typing long answers. If none of the options fit, you can always describe what you want in your own words.

Once Cora has enough information, it creates the agent. You will see a summary of what the agent does, when it runs, and any important notes (like if an integration needs to be connected first).

Every agent is created paused. Nothing runs until you explicitly turn it on, which gives you space to test and refine first.

This is where most of the magic happens. Once Cora builds the agent, do not just skim the summary and move on. Ask Cora to test it right away:

"Can you test this agent?"

Cora runs the agent in a safe sandbox where no real messages are sent and no real actions are taken. It shows you what the agent would produce: the Slack message it would send, the data it would pull, the decisions it would make.

Now look at the results. Is the summary too long? Missing a key metric? Using the wrong tone? Tell Cora what to change, right in the same conversation:

Cora updates the agent and you can test again immediately. You do not need to leave the chat, navigate to a settings page, or manually edit anything. Just keep talking.

The typical loop looks like this:

Most agents take two to five rounds of this loop to get right. That is not a sign something is wrong; it is how the process works. Each iteration gets you closer to exactly what you need, and it is much faster than trying to specify everything perfectly upfront.

When the test results look right and you are happy with the output, turn the agent on. You can do this in two ways:

Once live, the agent runs according to its trigger: on its schedule, when a signal fires, or when you manually run it.

Here are some tips for building agents with Cora:

Be specific about the output. Instead of "keep me updated on accounts," try "send me a Slack message every Monday with a summary of accounts that have declining health scores."

Name the channel or recipient. Tell Cora where the output should go: a Slack channel name, your email, or "each account owner."

Include the schedule or trigger. Saying "every weekday at 9am Eastern" or "whenever a renewal risk is detected" gives Cora what it needs to set the trigger.

Start with Crawl or Walk. If you are not sure about giving an agent full autonomy, start with internal-only alerts (Crawl) or drafts with approval (Walk). You can always increase autonomy later.

Test early, test often. Ask Cora to test the agent as soon as it is built. It is much easier to refine based on real output than to try to get the description perfect on the first try.

Give feedback in plain language. You do not need to know how agents work under the hood. Just tell Cora what you want to change ("make it shorter," "add this field," "only run on weekdays") and Cora handles the rest.

You can always refine later. Even after an agent is live, you can come back and iterate on it. See Managing and Refining Agents for how to adjust, refine, and manage your agents over time.