Six platforms, four months, one word: what actually shipped in marketing automation this year

TL;DR

• Six platforms, HubSpot, Marketo, Dynamics 365, Salesforce, Klaviyo and Braze, shipped AI agents between April and July 2026.
• Every one of them requires human approval before anything goes live.
• Every one of them only works well once a company’s customer data is already clean and organized.

Quick definition first, because the rest of this only makes sense with it: an “AI agent,” the way these companies use the term, is software that doesn’t just suggest what to do next. It goes and does the task itself, then checks with a person before anything goes live. That’s the whole idea behind the word “agentic.” Not smarter software. Software that acts on its own, inside limits a person sets.

Between April and July 2026, six marketing platforms all announced one of these. HubSpot, Adobe’s Marketo Engage, Microsoft’s Dynamics 365, Salesforce Marketing Cloud, Klaviyo and Braze each shipped something they call an AI agent, and all six used the word “agentic” to describe it. That’s not really a coincidence. It’s what happens when a whole industry runs out of manual steps to automate by hand, and starts trying to automate the decision about what to do next instead.

Here’s what each one actually shipped, because the word “agentic” is doing more work than the feature lists back up, and the differences between the six tell you more than the shared buzzword does.

HubSpot: reacting to a traffic problem

HubSpot’s April announcement is really a response to a number they didn’t like. Traffic to HubSpot customers’ websites from search engines like Google was down 27% over the past year. At the same time, 42% of buyers now do their research inside an AI chat tool like ChatGPT instead of Google. HubSpot says leads that come from AI search convert into customers three times more often than leads from a regular Google search.

So the main new product isn’t really an agent at all. It’s a $50-a-month tool called Answer Engine Optimization, which shows a company how it’s being described when someone asks ChatGPT, Perplexity or Gemini about them.

The agent features sit underneath that. One watches for signals like a company raising funding or hiring a lot of people, then drafts an outreach message on its own. HubSpot says early users get replies at twice the normal rate, and one customer booked 28% more meetings. Another agent now handles about 65% of email customer support questions without a person touching them. A third listens to sales call recordings and writes up notes and follow-up emails. All of these still get checked by a person before anything is sent, which tells you something about how HubSpot expects people to actually use this.

Marketo: the quiet one, and the honest one

Adobe’s update to Marketo Engage, on April 20, is the only one of the six with no customer results attached to it. Everyone else led with a number. Adobe led with something simpler: you can now type a request into a chat box instead of clicking through Marketo’s notoriously complicated menus.

What that chat box actually does is mostly cleanup work, not creative work. One agent finds duplicate contacts and fixes inconsistent data when you import a big list. Another checks a new campaign against the company’s naming rules before it goes live. Another turns a plain-English request like “show me everyone who opened an email but didn’t click” into the actual filters Marketo needs. Another catches broken links before a campaign sends. None of this is glamorous. It’s the tedious quality-checking that eats a marketing operations person’s whole Tuesday, now handed to something that doesn’t get bored doing it. Adobe also added a feature that watches campaign performance and flags on its own when something looks off, like a sudden drop in sign-ups.

Dynamics 365: betting on the phone call

Microsoft’s April 27 update for Dynamics 365 barely mentions email at all. The new feature, called Conversational Journeys, is built around back-and-forth AI conversation over phone calls and text messages. Microsoft points to research showing 82% of customer interactions still involve a phone call at some point, either right away or after something else didn’t work. That’s a different starting point from the other five, which are mostly still fighting over email and chat.

The phone version of this is already available inside Microsoft’s customer service tools. Text messaging is being added next. Microsoft is also careful to say the AI only steps in “when it truly matters,” which sounds like an attempt to avoid the backlash chatbots have gotten for years for being annoying and useless.

Salesforce: the biggest launch, with a data problem underneath it

Salesforce’s announcement is the largest of the six. Two agents are fully available now. One, called Piper, talks to people browsing a company’s website, figures out if they’re a real potential buyer, and connects them to a salesperson without making them fill out a form. Another, called Hunter, finds potential customers based on buying signals and runs email sequences to reach them. Two more agents are still in early testing: one that writes and translates marketing content for different countries and channels, and one where a marketer just states a goal and some rules, and the AI builds and runs the whole campaign against that goal on its own.

The results Salesforce is sharing are real customer numbers, not projections. One customer, Rawlings, builds campaigns 75% faster now. Another, Emplifi, cut the number of salespeople needed to screen leads by about 20%, while the number of actual deals created went up more than 22%. Salesforce also announced it’s buying a content management company called Contentful, which tells you what gap they still see: the AI can plan a campaign, but there isn’t enough good content ready to fill it.

The most useful thing said about this launch didn’t come from Salesforce itself. A trade publication covering the announcement pointed out that the agents “have limited context” without Salesforce’s separate data product feeding them clean, well-organized customer data first. In plain terms: the automation is real, but only after a company has already done the unglamorous work of getting its customer data in order. Most companies haven’t.

Klaviyo: one system doing two jobs

Klaviyo’s launch, on June 30, pairs a marketing agent with a customer-service agent, and correctly points out that most competitors only built one of the two. The marketing one, called Composer, reviews everything a brand currently has running, points out gaps and opportunities, and builds a ready-to-review campaign in a few minutes. The service one, called Customer Agent, handles support conversations over web chat, email, text and WhatsApp, including actions like processing a refund, and reviews its own performance every day to find where it’s getting things wrong.

Both agents pull from the same central store of customer data, so information from one can inform the other. One early customer, a beauty brand, used Composer to find that it had 113 separate automated email sequences running, several of which were quietly working against each other. That’s a genuinely useful thing for an agent to do: not writing new copy, but finding a mess that was already there and nobody had noticed.

Braze: leading with the design team

Braze’s announcement, made on April 23 at an event in London, is the one aimed most directly at the people actually building campaigns day to day, rather than the operations team behind the scenes. One new tool lets a team build their own custom agents. Another connects directly to design tools like Figma and Canva, so a design doesn’t have to be exported and re-uploaded to go live. A separate update adds European data hosting for companies that need customer data to stay within the EU by law.

The results here are some of the strongest of the six. One customer, Cleo, cut people unsubscribing from its emails by 81% and got 284% more people opening its app. Another, Dayuse, increased the rate of people actually completing a booking by 90%. A staff member at the American Diabetes Association described working with Braze’s AI assistant like this: “I felt like I was talking to someone who worked for Braze. It’s my thought partner.”

The pattern across all six

Line all six launches up next to each other and a pattern shows up fast. Every single one of these agents works inside a human-approval step. None of them is fully autonomous, and none of the companies are pretending otherwise. Every one of them only works well once a company’s customer data is already clean, organized and in one place, which is the exact thing these same companies have been telling customers to fix for the last ten years, and most companies still haven’t done it. And every single result mentioned above came from the vendor’s own customer story, not from an independent study.

None of that makes these launches meaningless. It just means “agentic” in 2026 has a fairly specific, narrower meaning than the word suggests: an AI that can act on its own, but only within limits a person set, and only once the data behind it is in good shape.

A separate industry report on advertising backs this up. It expects regular AI writing and building tasks to become completely normal in 2026, but expects the AI making bigger decisions on its own to take longer to catch on. The report says the real holdup isn’t whether the AI can do the work. It’s trust: 60% of advertising professionals surveyed said they’re worried about accuracy and not being able to see how the AI made its decision, and that’s the main thing slowing them down, not the technology itself.

So here’s the actual story of 2026. It’s not that these platforms suddenly got smarter. It’s that six of the biggest companies in this space ran into the same wall at almost the same time: there was nothing left to automate by hand, so they started automating the decision-making instead, with a person still signing off before anything goes out. Whether that’s a real turning point or just an old idea with a new name depends entirely on one thing none of these companies can fix for you: whether your own customer data is clean enough for any of this to actually work. Most of the case studies above don’t say what it took to get there. That’s the question worth asking before you sit through any of these demos.

Sources

• HubSpot targets AI-driven buyer behavior shift with new tools and agents (SiliconANGLE)
• HubSpot Launches AEO & Expands AI Agents in Spring 2026 Spotlight (CMSWire)
• Agentic AI in Adobe Marketo Engage for modern marketing ops (Adobe)
• Turning customer experience into a growth engine (Microsoft Dynamics 365 Blog)
• Salesforce Puts an AI Marketing Team in Every Marketer’s Hands (Salesforce)
• How Salesforce Is Rebuilding Marketing Cloud for the Age of AI (Salesforce Ben)
• Klaviyo Launches AI Agents that Work Together to Drive Revenue for Consumer Brands (Klaviyo)
• Braze Delivers Powerful New Agentic AI Capabilities (BusinessWire)
• GenAI will take over programmatic advertising in 2026 and agentic AI isn’t far behind (eMarketer)

Frequently Asked Questions

What does “agentic AI” actually mean in these marketing platforms?

Software that carries out a task on its own instead of just suggesting it, but still checks with a person before anything goes live. It’s automation of the decision, not just the execution, within limits a person sets.

Which platforms shipped agentic AI features in 2026?

HubSpot, Adobe’s Marketo Engage, Microsoft Dynamics 365, Salesforce Marketing Cloud, Klaviyo and Braze all announced AI agents between April and July 2026.

Do any of these AI agents run without a human checking their work?

No. Every one of the six launches keeps a human-approval step before anything is sent, published or acted on.

What do companies need in place before these agents actually deliver results?

Clean, well-organized customer data. Every vendor’s own case studies point to this as the real precondition, and most companies still haven’t done that underlying data work.