Iterable, Klaviyo and the rest spent 2026 putting agents in the campaign builder. They are genuinely good at execution. The scarce resource moved somewhere else entirely.
In April, Iterable announced Nova Agent, an AI agent built to act on customer behaviour in real time, alongside a broader suite it calls Nova Intelligence. Klaviyo has been moving Composer, its AI marketing agent, through beta and pairing it with a Customer Agent. Braze has been shipping in the same direction. By the middle of the year, every serious customer engagement platform had an agent in the product.
The marketing response to this has been a familiar one: a quiet panic about whether the team should switch, followed by a demo, followed by a pilot. That is the wrong sequence. The right question is not which agent to buy. It is which decisions you are asking an agent to make on your behalf — and whether you have ever made those decisions deliberately enough to write them down.
Look past the positioning and the capability list is consistent across vendors. Agents build campaigns, audit messages, generate and vary content, run experiments, and optimise toward stated goals. Iterable's launch paired the agent with a centralised view of campaigns and performance, audience activation for people who have not yet identified themselves, a paid-channel integration, and a governance toolkit for compliance and message retention. Klaviyo's framing is an agent that surfaces what needs attention and another that acts on it.
These are real capabilities and they are not marketing theatre. If your team spends its week assembling campaigns, auditing links, and resizing images, an agent will genuinely take that work. The evidence from the vendor side is that the work is being taken: the pitch everywhere is speed — more experiments shipped, faster from idea to optimised journey.
An agent can execute a decision. It cannot originate one. That is not a limitation of the current models so much as a definition of the job it was given, and it is where every rollout quietly stalls.
Production was the bottleneck for twenty years. In 2026 the bottleneck is deciding what is worth producing, and most teams have no written answer.
When output becomes cheap, the constraint moves to input. A team that could previously ship one well-considered campaign a month can now ship thirty — and if the judgement behind them is unchanged, they have thirty mediocre campaigns, a rising complaint rate, and a deliverability problem to solve.
This is already visible in the shape of the industry conversation. Every serious discussion of AI in email now ends up on the same topics: guardrails, eval, and who signs off. Those are all judgement topics wearing engineering clothes.
Write the decision layer first — in a document, not a platform. Which signals matter, what each one is allowed to trigger, what the goal metric is, what the guardrails are, and who owns a bad send.
Then point the agent at it. Teams that do this find the agent useful within a week. Teams that buy the agent first and look for decisions later spend a quarter discovering their own opinions, at production volume, in production.
We are hired for judgement: knowing what to build, in what order, and what to leave alone. Agents have made the building faster. They have not made the knowing easier, and in a market where every competitor now has the same execution capability, the knowing is the whole difference.
If your platform shipped an agent this year, the useful next step is not a migration. It is an afternoon spent writing down the decisions you currently make by instinct, so that something other than you can carry them out.
Written by the team at LMR — a senior lifecycle marketing studio working in Iterable, Braze, Klaviyo and the tools you already run.
Start a project ↗