Braze Conversational Agents sell over WhatsApp and SMS
Braze put Conversational Agents for WhatsApp, SMS, RCS and web into beta, plus an MCP layer for Claude and ChatGPT. Here is what to feed a sales agent.
Braze put Conversational Agents into beta on September 29. The agents answer shopper questions, clarify intent and recommend products across WhatsApp, SMS, RCS and web chat, grounded in a brand's product catalog and FAQs. In the same release, Braze opened its platform to Claude, ChatGPT and other MCP clients. Messaging stops being a broadcast channel and becomes a sales conversation. The agent is only as good as the catalog you feed it.
What actually happened
Per Braze's announcement, four BrazeAI products shipped in beta:
Conversational Agents. Brands configure them with brand guidelines, guardrails and knowledge sources. Braze positions them to move a customer "from discovery to conversion" inside the message thread.
BrazeAI Operator Connect. This is headless Braze over the Model Context Protocol. Marketers and developers can work in Braze from Claude, ChatGPT, Snowflake Cortex Code and Microsoft Copilot. Each action runs under the individual user's identity and that user's existing Braze role permissions.
Agentic Standards. This is automated campaign QA. It checks setup, copy, links, personalization, compliance and brand rules, then returns pass, warning or fail flags with an audit log. Braze expects general availability in October.
Decisioning Studio Go. This is self-serve 1:1 optimization of message variant, timing and frequency. Braze expects general availability on October 14.
One partner, Stitch, claims Operator Connect cuts speed-to-market "by up to 90%." That is a partner estimate with no baseline, so we do not rely on it.
Why it matters for your business
You probably do not pay for Braze. The pattern still applies to you, because Klaviyo, Attentive and Shopify Inbox are on the same path. Three lessons:
Your catalog is the script. A chat agent that recommends products reads your product data. It reads titles, variants, stock and return policy. Messy data gives wrong answers in a channel where the customer expects a person. Clean the catalog before you turn anything on.
Guardrails are a config file, not a vibe. Write down what the agent must never do: promise a ship date it cannot see, invent a discount, or discuss a refund above a set amount. Then test those rules before launch.
Permission-scoped MCP is the right model. Braze runs every agent action under the user's own role. If you connect Claude to your marketing stack, insist on the same thing. Do not use a shared admin key that any prompt can use.
WhatsApp and SMS also carry consent rules. An agent that texts a customer is still your business texting that customer.
Key takeaways
- Braze launched Conversational Agents in beta on September 29 for WhatsApp, SMS, RCS and web
- Agents run on brand guidelines, guardrails and knowledge sources such as product catalogs and FAQs
- Operator Connect exposes Braze to Claude, ChatGPT, Copilot and other MCP clients under each user's own permissions
- Agentic Standards (campaign QA) and Decisioning Studio Go target GA in October
- Before any chat agent sells for you, fix your catalog data and write explicit guardrails
A sales agent in the text thread is only as honest as your product data. We build catalog pipelines and scoped chat agents on channels you already use, with guardrails you can read and test. See what we build for commerce, or tell us where your customers ask questions today.
Sources: Braze investor relations.
- #braze
- #conversational-commerce
- #mcp
- #ai-agents
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Manus Cue gives every agent an email, phone, and wallet
Manus 2.0 shipped Cue: personal AI agents with their own phone number, email address, and budgeted wallet. Your phone line is about to take bot calls.
Read itAtomic raises $12.5M; DoorDash buys 90% autonomously
Atomic's AI runs about 90% of purchasing for DoorDash's DashMart. Autonomous inventory purchasing works when your decision rules are written down.
Read it