It's easy to believe that "AI" can do whatever you want and completely replace humans. In CRM/MA as well - "it can do everything" says the chorus of self-interest. Rather, Braze Agents promise to be able to generate content, make real-time decisions and enrich data directly in Braze. As a CRM manager, the question is therefore 'what are my use cases and do they continue to work even when we scale up?
I will tell you about the capabilities of Braze Agents and it is up to you to discover how well it works for you.
TL;DR
-
What they are: AI helpers you create in Braze that can generate content, make "smart" decisions about paths in Canvas and enrich data.
-
Where they are used: as Agent steps in Canvas and as fields in catalogs.
-
This results in: potentially less manual effort, fewer static rules and more scalable personalization.
What is BrazeAI and Braze Agents?
BrazeAI is a Braze collection of AI features. Agents are part of this portfolio and can be thought of as "custom" AI components that you can adapt to your needs within the categories: copy, data in Catalogs and decisions in Canvas.
As with any AI solution, it's important to make sure your business has enough well-structured data. Make sure to have it in place first before you even think about it.
If this is already the case - great!
What then is the value said to be? The value of agents is that they can reduce three bottlenecks:
-
Production of copy variations
-
Personalization in Canvas
-
Completing data in Catalogs
What can Braze Agents do?
Generate copy and content
Braze Agents, like most language models , are good at writing - headlines, subject lines, product text etc. But what is useful is the ability to write around a context. With access to attributes and the behaviour of individual users, you can potentially create individual personalization on a much larger scale than you have been able to do before.
For example, an online shop could personalize its newsletter with news so that one person (based on gender and an attribute like "latest product category") gets suggestions that include men's shoes in the style he likes, while another gets suggestions for women's bags.
It can thus both write and suggest and then, in this scenario, act as a "recommendation engine light".
Decisioning in Canvas
The more powerful capability is decisioning: using agents to guide choices in Canvas. In itself not so exciting, because it is best done by experienced experts BUT the feature comes into its own when this can be combined with unstructured data such as free text responses in forms.
Customizing paths through journeys and also creating personalized content on similar data is something new. Something that has not been possible for your team no matter how many and how competent you have been.
Braze, 2026.
Imagine a scenario where dissatisfied and loyal customers quickly receive a customized response and are contacted by support (who have info why they are dissatisfied). Properly implemented, it can do wonders for retention.
Data management in Catalogs
Braze, 2026.
Agents in Catalogs open up for use cases such as writing product descriptions from product images, creating tags from the same or product names, or filling in missing attributes. Useful in scenarios where data needs to be completed.
It will always be best to structure the data and then push it into Braze, instead of relying on a language model without errors to create everything from product description to color categorization and style. But where there are gaps in your data, I can see this feature being really useful!
Braze-powered vs "bring your own"
Braze also allows you to choose which model you want to power your agents. By default, Gemini's models are used but you can use e.g. OpenAI or Anthropic and also choose which versions to use for which agents. It is worth noting that if you use your own model, the use is billed via the supplier, not Braze.
Braze, 2026.
As always, just test what works best in your particular situation.
You can then see what your agents do in the Logs. Review runs and see input/output per call. An absolutely essential feature for people like me who get chills at the thought of connecting an agent and hoping the rest works out.
Braze, 2026
Summary
Braze Agents is not magic, but with clear cases that fit the capabilities of the language model, it can be a great tool to help your team scale your personalization to new levels. It doesn't replace an expert, and it certainly doesn't replace the need for tacticians linked to a clear strategy - it's simply another tool in your toolbox.
Do you have questions about Braze and AI? Do not hesitate to contact us.
