Braze AI in 2026: What’s Live, What’s Coming, and What to Do About It

Every engagement platform is shipping AI right now, and every marketing team we talk to is trying to figure out what’s real. Which of this can I turn on Monday? Which needs a call with my account team? Which is genuinely worth my team’s time this quarter, and which can wait until next year?

We spend our days in these platforms, so we’re going to work through them one at a time. Braze first — it’s moving the fastest, and Forge lands in Las Vegas September 28–30, so another wave of announcements is a few weeks out.

Here’s where things actually stand.

What’s live today

Braze puts its AI under the BrazeAI™ banner. Most of it you can use today — but availability, beta status and credits vary by feature, and that’s where teams get surprised.

The assistant sitting in your dashboard

Operator is the AI assistant built into Braze, opened from any page next to your profile. It’ll answer questions, help you get unstuck (“why isn’t my Liquid rendering?”), and draft campaigns, Canvases, segments and content you can review and drop straight into your work. It knows what page you’re on, and it remembers the conversation until you clear it.

Two things most teams never turn on: you can give Operator your brand guidelines as context so its copy sounds like you, and there’s a prompt library of examples worth stealing. Load your guidelines once instead of re-explaining your tone every time you open the panel.

Agents you build yourself

Braze Agents are AI helpers you create inside Braze and manage in Agent Console. They write content, make routing calls and enrich data — push copy shaped around one person’s actual preferences, or catalog fields calculated on the fly.

They run inside a journey through the Agent step: you decide what context goes in, the agent responds, and the response lands in a variable you can use three ways. Route people on it — a lead-scoring agent returning “Sales Ready,” “Marketing Qualified” or “Disqualified,” with the first triggering a Slack alert and the last dropping out of the journey. Personalize with it — drop the response into the message. Or store it — a sentiment score or product affinity written back to the profile or pushed out by webhook.

One thing to sort out before you design anything: agents require Message or Action Credits. If you don’t have Action Credits today, that’s a conversation with your account manager first.

And a limitation Braze is refreshingly honest about: an agent personalizes beautifully in the moment, but it doesn’t learn from what happened next. It has no way to see whether its copy actually converted and feed that back into the next decision. Braze’s own phrase for this is “vibe” decisioning — which is exactly why the next thing exists.

Decisioning, for when you want the machine to optimize toward money

Decisioning Studio — the former OfferFit, now folded into Braze — swaps A/B testing for agents that optimize toward a number you care about: revenue, profit, CLV, retention, renewal, winback. Rather than crowning a winner for a whole segment, it picks the best message, offer, channel, timing and frequency for each individual out of your first-party data. It’s a “talk to Braze” motion rather than a toggle, and the Pro tier comes with their AI Decisioning Services team helping tune your agents.

If that’s a bigger swing than you’re ready for, Decisioning Studio Go is the on-ramp: individual-level decisioning for your recurring email programs. You supply the options — subject lines, CTAs, images, send days, send times — and the agent picks the best combination for each recipient based on how they’ve engaged before, honoring quiet hours and local time zones.

The detail we love: it holds back a random control group of at least 5% that gets random combinations of the same content, forever. So you always have a like-for-like read on what the AI is adding. If you’ve ever been asked to prove AI’s value to a CFO, that control group is the answer.

The quiet workhorses

Easy to skip past because they aren’t new. Still the fastest wins on the list:

• Intelligence Suite — Intelligent Selection quietly shifts traffic to your best-performing variant, Intelligent Timing sends at each person’s best hour with a fallback for thin data, and Intelligent Channel reaches people where they actually respond.
• Predictive Suite — churn risk and predicted events you can use as filters in segments and journeys.
• Winning Path — test entire journey paths against each other, then send everyone who follows down the one that converted best. Watch one trap here: if you’re measuring opens or clicks, the message has to be the first step in the path or it won’t count.

Content and creative

Generative copy and creative work inline while you build. Content Optimizer goes further, running as a journey step that generates a pile of variants — subject lines, preheaders, sender names, headers, body copy, CTAs and images for email; titles and messages for push; hooks, bodies and CTAs for SMS, MMS and RCS — then quietly retires the losers and scales the winners. It’s still in beta, limited to those channels, and you’ll need your CSM to switch it on. Item Recommendations pulls the right catalog items into each person’s message.

On the production side, Creative Studio connects design to execution, with Figma and Canva integrations that bring designs and templates into Braze instead of making someone rebuild them.

One more thing worth having ready before security asks: for the features that use OpenAI, what goes over is your prompts plus the dashboard and workspace content relevant to the request. That data isn’t used to train OpenAI’s models, it may be held for 30 days for abuse monitoring, and the output is yours. Braze documents all of it, which makes the conversation with your legal team a short one.

The MCP question: what actually changed

If you’ve been paying attention you’re already asking it — hasn’t Braze had an MCP server for a while? It has, and the difference matters more than the headline.

The original, from August 2025, ran on your machine. You installed a package, kept a config file, minted an API key with the right permissions and hoped whoever set it up documented what they did. That version is now deprecated and won’t get further updates.

The current server is remote and hosted by Braze. Nothing to install — you connect a supported client in about a minute, authenticate with OAuth, and your access is tied to your Braze dashboard account. Your agents inherit the permissions you already have.

That’s the real shift. Not “Braze has MCP now,” but that MCP access is finally governed by who someone is rather than a key sitting on a laptop. It reads and writes — campaign and Canvas analytics, segments, custom attributes, and it can create email templates, content blocks and media library assets — and user-level PII is deliberately off the table, so nothing from a user profile comes back.

Practically: if your team lives in Claude, Cursor or an internal assistant, this is what lets someone ask “how did last week’s onboarding journey do?” and get an answer without a CSV export — with auth your IT team will actually sign off on.

What we know is on the roadmap

Braze has been unusually open about what’s coming, and most of it is in beta or already crossing into general availability rather than sitting on a slide:

• A headless Operator — the same capabilities available outside the dashboard UI, so you can wire them into your own workflows.
• A more capable Operator that builds new journey steps from a conversational prompt and lets you stand up pre-built agents, instructions already filled in, in minutes.
• A deeper, two-way Databricks CustomerLake connection — governed audiences activated into Braze, engagement signals flowing back out, with a lot less middleware between your warehouse and your journeys.
• EU hosting for Decisioning Studio, for teams with regional data requirements.
• More at Forge, September 28–30. We’ll be there, and we’ll write up what we hear.

What to do about it now

If none of this is switched on today, don’t start with agents. Start here.

1. Turn on the optimizers you’re already paying for. Intelligent Timing, Intelligent Channel and Intelligent Selection need no new data model, no prompt writing and no new governance — just a sensible fallback for people without much history and a holdout so you can see the lift. Best return for the effort in the whole stack, and it’s sitting switched off in most accounts we walk into.

2. Put your predictive scores into a journey, not a dashboard. A churn score that only lives in a report changes nothing. The same score triggering an earlier win-back, holding back a discount from someone who was going to buy anyway, or easing off the cadence for an at-risk customer — that’s worth something. One segment, one branch, measured.

3. Give Decisioning Studio Go one recurring email program. It’s the cleanest path to real individual-level decisioning in production, and that mandatory control group means you finish the quarter with a number instead of a story.

4. Pilot one agent on one narrow, low-stakes job, with a human in the loop. Subject-line variants. Catalog enrichment. Sort out credits, then the guardrails: what it can say, what it must never say, who reads it before it sends, how you back it out. And keep in mind it isn’t chasing your goal — if that’s what you need, you need decisioning, not an agent.

5. Decide how MCP access works before someone improvises it. OAuth and permission inheritance are what make this a decision you can actually govern: who gets a connected client, what those dashboard users can see, and whether the write tools are in scope.

6. Do the unglamorous foundation work. Consistent event and attribute names. Clean catalogs. Journeys that aren’t ten years of accumulated exceptions. Everything above reads your data and your structure — solid foundations turn AI into leverage, messy ones just make the mess faster.

Where we come in

We’re marketing geeks who help organizations modernize marketing through Strategy, Technology & Data. We work across the platforms that power modern engagement — Oracle, Adobe, Salesforce, Braze, Iterable and the data behind them — so marketing teams can put AI to work responsibly.

If you’re weighing which of these to take on first — or you want help getting your data, catalogs and journey architecture ready for them — let’s talk. And if you’d rather start with an honest read on where your team stands, take our Marketing AI Readiness Assessment.

One caveat: this space moves weekly. Availability, beta status and credit requirements change, so confirm the details with your account team before you build a plan on them.