MGS 2026 Recap: Braze on AI Customer Engagement

Does it feel like the more messages and creative you produce with AI, the less your customers respond? As send volume climbs, customers learn to tune brands out, and the extra work doesn't turn into results.

At Modern Growth Stack (MGS) 2026 on July 21, 2026, Shahid Nizami of Braze took the main-hall stage with "The Next Era of AI Powered Customer Engagement." Drawing on 25 years in the industry, he laid out what comes next for customer communication in the AI era.

The daglo marketing team attended MGS 2026, and this post shares what we learned. We transcribed the session with daglo and reread it to pull out what CRM and growth teams can check right away.

AI customer engagement is about outcomes, not output

The next era of AI customer engagement isn't about sending more. It's about making better decisions that drive better outcomes, and that means combining AI agents and reinforcement learning on top of first-party data.

First-party data is the data a company collects directly from customers in its own app or website. Nizami began by noting that marketing's standing has already changed:

• The days of marketing as a cost center are over. Marketing now drives revenue.
• The days of doubted ROI are over too. You can prove ROI at every stage of the customer journey.

For scale, Braze supports more than 8.5 billion monthly active users worldwide, and by the end of MGS day one, Braze customers would send more than 4 billion messages to their end users.

AI slop: infinite content, finite attention

"AI slop," named a word of the year in 2025, describes output that AI churns out nonstop without creating lasting value. AI-generated content is effectively infinite, but customer attention is finite, and loyalty is even scarcer.

• Demands for more content, campaigns, and creative end up producing more noise.
• Every piece of slop that reaches a customer trains them to tune you out.
• The moment you focus on output, you lose sight of outcomes.

Not everything gets better just because you add AI.

Nizami offered two examples. In the 1970s, Grandmaster Flash and the Furious Five released only two albums, yet became the first hip-hop act inducted into the Rock and Roll Hall of Fame. And HotWired's banner ad in 1994 had a 78% click-through rate; by 2011 that figure had fallen nearly 1,500-fold. 2011, the year the banner ad effectively died, is the year Braze was founded.

Four principles for building AI agents well

Agents need craftsmanship too. "Let's make everything an agent" is not the answer, and handing customer communication to agents without context management and firm guardrails is risky.

Give an agent too much information and it slows down or loses focus, which the session called context rot and attention collapse. Agent output is fundamentally non-deterministic, yet we trust agents with high-stakes work: talking to millions of customers, managing subscribers, running purchase funnels, and hitting quarterly targets. So agents need four things:

1. Managed context: Give them only the information they need.
2. Deterministic guardrails: Even when outputs vary, lock the lines that must never be crossed into rules.
3. Smaller problems: Break big problems down instead of giving one agent everything.
4. The right goal: Don't get lost in the tool and forget what you're building and why.

From first-party data to AI decisioning and reinforcement learning

Outcomes start with context. Brands hold plenty of first-party data but rarely use it, and connecting that data is what keeps personalization from turning into slop.

Braze said it processed more than 25 trillion data points through its SDKs and APIs last year and made about 9 trillion profile updates through cloud data ingestion. On top of that context, Nizami introduced AI Operator, which helps write emails and manage journeys; Decisioning Studio, which matches the right message to the right person; and AI Agent Console, which builds agents and places them into campaigns.

The demo used a hypothetical marketer at a live-music app who wanted to give fans a personal concierge experience. AI Operator built an agent that drew on company data, brand guidelines, and the media library to write personalized pitches, then produced the email HTML. Going from idea to launch took minutes.

• The screen shows three customer attributes, but the system looks at more than 100 and makes true one-to-one decisions on frequency, channel, message, incentive, and timing.
• Reinforcement learning means autonomously testing to discover patterns unique to your customers. When the market shifts, the agent shifts too, and every customer action feeds back into the loop.

You don't have to choose between agents and reinforcement learning. The most powerful systems combine both.

Takeaways: a checklist for CRM and growth teams

The key lesson is to win on decision quality, not send volume. Use this checklist to review how your team talks to customers.

• Separate output and outcome metrics: Overlay send volume and response rates to see whether more sends are eroding engagement.
• Count your personalization axes: Check how many of frequency, channel, message, and timing you actually personalize.
• Find idle first-party data: List the data you collect but never use in messaging decisions.
• Set guardrails before agents: For high-stakes flows like purchases and subscriptions, design smaller context, smaller problems, and rollback criteria first.

For recordings where several voices mix, like a Q&A, and knowing who said what matters, daglo's Speaker ID is a useful feature. It remembers the voices of the people in your recordings and fills in their real names instead of "Speaker 1" from the next recording on, and daglo is the first service in Korea to offer it to every user. Learn more about Speaker ID

Catch up on the other sessions in our full MGS 2026 recap.

#MGS2026 #ModernGrowthStack #Braze #CustomerEngagement #CRM #AIAgents #Personalization #daglo

👉 Transcribe your next session