A 2025 MIT study found that only 5% of AI investments deliver positive ROI. At the same time, more than 99% of marketing leaders say their brands are already using AI for customer engagement.

Braze’s own research makes the main reason clear.

​Although 93% of marketing leaders think AI helps them understand customer needs, no more than 53% of consumers feel that brands are actually predicting what they want.

​There is a huge gap between what brands hope to do with AI and what they actually achieve. The time to close this gap is running out. Right now, only 19% of consumers use AI agents when interacting with brands, but that is expected to rise to 46% by the end of 2026. Gen Z is leading the way, with 64% already saying that brands are keeping a human touch with AI. 

This is the context in which Braze launched Agent Console. (Not to be confused with Salesforce’s Agent Console which is a dashboard for humans answering help requests.)

Agent Console became available to everyone in early Q1 FY2027, ahead of schedule. But what exactly is Agent Console in Braze? How does it help you? And, is it even worth the hype? Let’s find out. 

What problem does Agent Console address?

Traditionally, customer engagement in Braze is chiefly rules-based. You create segments, set conditions, assemble components in Canvas, and deploy.  

But as the customer journey becomes more complex and the data signals you’re working with multiply, a purely rule-based system starts to hit a ceiling. 

Suppose you want to route a new user into the right onboarding cohort based on their behavioral signals. Writing fixed rules for every combination of those signals is not feasible. 

And the rules you do write often default to the lowest-common-denominator path. 

Agent Console is Braze’s way of addressing these challenges. 

Instead of a fixed rule, you give an AI agent a set of behavioral signals, a goal, and the authority to make a decision. The agent evaluates everything holistically and makes a call. 

What Agent Console is 

Agent Console in Braze is a feature that helps you build, set up, and manage AI agents. You can then add these agents as steps in a Canvas workflow.

It’s a Canvas step. Instead of adding a filter or a delay, you add an Agent Step.

This step uses an agent you set up in the Agent Console. The agent’s output becomes a context variable that other parts of your Canvas can use.

You define Braze’s AI agents using these settings:

• Instructions: Describe in natural language what you want the agent to do.
• Model settings: Adjust options like the temperature dial to control how creative or precise the agent’s output will be.
• Catalog access: You can connect agents to Braze Catalogs. This way, they use real, curated data you’ve added, instead of making up product names or program titles.
• Brand guidelines: If you set these up at the workspace level, the agent can use them.
• Output format: Choose what you want the agent to return, such as a string, a boolean, an HTML block, or a structured recommendation.

The output from the agent step will be saved as a context variable in Canvas. You can use this variable anywhere later in the journey.

How context gets to the agent 

Where does the agent pull its data?

​It gets its data from the Canvas. ​When an Agent Step runs, it can use all the context variables set in the Canvas. This includes entry properties, event properties, data from connected Catalogs, and any variables set by earlier steps. This is the “include all canvas context” option, usually the best place to begin.

​This setup is deliberate. Braze processes and compresses user context before sending it to the agent. As a result, the agent gets a focused, structured snapshot of what matters for that user right then, without wasting tokens on sorting through extra data. By handling the context layer, Braze lets the AI model focus on making the right decision instead of sorting raw data.

​This approach also helps control costs. Since the context is already shaped and compressed, you don’t need to use the biggest or most expensive models. Faster, smaller models work well here, which is important when running agents at B2C scale. Even small costs per use can add up quickly across many users. 

For use cases that require precision, you can also pass specific key-value pairs directly into the agent step as inputs. There’s no need for opening up the full canvas context. 

Braze Agent Console is built for teams, not just individual workflows 

Canvas has always functioned as a shared control plane, a place where cross-functional teams coordinate on the customer experience. Agent Console extends that into the AI layer. The agents you build are shared resources in your workspace. The decisions they make are embedded in collaborative Canvas flows that your whole team can see, adjust, and build on top of. 

If your team is looking to scale AI across the org, understanding this multiplayer architecture is one of the most important things to know about Agent Console in Braze. 

Before you get started with Braze Agent Console 

A few practical considerations can come in handy down the road: 

• If your agent is making recommendations from a defined set of options, such as learning paths, product categories, content pieces, cohort names, etc. put those options in a Catalog and connect it to the agent. Without a catalog, the agent will crank out fictional outputs. 
• The more context you provide in the Canvas when the agent runs, the better its decisions will be. Consider which data points the agent should access, and make sure to pass them as context using entry properties, event triggers, or upstream context variables.
• Agent Console tracks usage by counting each time an agent runs for a user as one invocation. This is important for high-volume situations. If your Canvas has 500,000 monthly active users who go through an Agent Step, that means 500,000 invocations each run. Plan accordingly. 
• To improve testing results, provide specific output format instructions. Vague instructions often produce lengthy outputs, while clear guidance yields more useful context variables.

The last word on Braze Agent Console 

At Braze’s London City by City event, automated QA entered beta as a direct offshoot of the Agent Console product. It was also, notably, the top-demanded feature at the customer advisory board that same day. It was the first time QA had ever risen to that level in a Braze CAB! 

The implication is significant. 

Agent Console is beginning to support the internal workflows marketers use to create and test campaigns. While this is still in the early stages, it’s something to keep an eye on. 

Agent Console is worth getting hands-on with. Begin with a basic use case, connect your catalog, adjust your output format instructions, and make sure to test everything before you launch it in a live Canvas. It’s easier to start seeing value from Braze’s AI agents than you might think. 

And there’s a lot of potential to achieve even more as you get comfortable with it.