Field notes, April 2026: agentic marketing moves from demos to workflows
AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.
Agentic marketing in April is mostly a story about narrowing ambition. The questions we keep returning to are less about what an agent could do in a demo and more about what a team could run every week.
These are our monthly notes: themes we are tracking, questions we are studying, and what we expect. They are working hypotheses from our research and prototypes, not measurements of the market, and we revisit them as evidence accumulates.
The demo gap is the theme of the season
A year ago, the public conversation about agents in marketing centered on demonstrations. An agent would plan a campaign end to end, write every asset, and launch it. Those demos showed what was possible, and they were a poor guide to what a team could rely on.
The questions we find most useful this spring are narrower. How do you connect an agent to a CRM without risking the records in it? How do you review output when volume grows faster than reviewers? How do you know whether an agent saved time at all?
We read the shift toward questions like these as a sign of a field moving from exploration toward operation, and we expect it to continue through the year.
The mechanism behind the gap is simple. A demo runs once, on chosen inputs, with someone watching. A workflow runs every week, on whatever inputs arrive, often with no one watching. Most of what separates the two is the dull part: error handling, review, logging, and a plan for what happens when the agent is wrong.
Protocols may turn integration into plumbing
Standard ways of connecting agents to tools, and agents to other agents, are getting more attention. Our working view is that these protocols, such as Model Context Protocol, could make integration a smaller barrier than it has been. We wrote about what that means for marketing teams in MCP for marketers.
If integration gets easier, we expect the main barrier to shift to governance: deciding what agents should be allowed to do once they can technically reach almost anything. A connector that lets an agent read a campaign list can often let it edit one too. The permission question moves from "can we connect this" to "should this agent ever take this action, and who decides."
A pattern worth testing is to treat every new tool connection as a permission decision with an owner, recorded alongside the connection itself. That keeps the record of what an agent can do as current as the agent.
Evaluation looks like the next bottleneck
The theme we expect to define the year is evaluation. Our hypothesis is that the teams that make steady progress will be the ones that can tell, quickly, whether a change made things better. Sophistication of the agents themselves may matter less.
Evaluation sounds technical, but most of it is editorial. A useful starting kit is small: a set of real examples, a rubric that says what good looks like, and a habit of recording how much reviewers change agent output. We sketched one version in a starter rubric for content agents.
Consider a hypothetical team that changes its content agent's instructions every week. Without a fixed set of test cases, each change is judged on whatever drafts happen to come out next, and a regression can hide for weeks. With 20 to 30 saved examples and a rubric, the same team can compare versions side by side in an afternoon.
Review volume is the constraint we keep running into
Our fourth theme follows from the third. Agents can produce far more drafts than a review team can read with care, and approval queues are where that gap shows up first. When the queue grows, reviewers face a choice between delay and shallower review.
We think the answer lies in how approval is designed rather than in reviewing faster. Batching similar items, routing by risk, and sampling low-risk repeat formats are all patterns we are testing. We describe the early version of this work in approval queues that don't become bottlenecks.
Questions we keep returning to
Several open questions came up repeatedly in our research this month. We list them so readers can test them alongside us.
- Approval at scale: how should approval workflows change when agent output grows faster than reviewer capacity?
- One agent or several: which marketing tasks benefit from multiple specialized agents, and which are better served by one well-designed agent?
- Presence in AI answers: how should a brand measure its presence in AI-generated answers when those answers vary by user and session?
- The unit of cost: should model spend be tracked per token, per task, or per outcome?
Our current leaning on the last question is per task. Tokens are too granular for a marketing budget, and outcomes are too slow and noisy to manage spend week to week. That is a starting view, not a finding.
What we are watching next month
May will tell us more about how these themes play out in practice. These are the signals we plan to watch.
- Built-in agents: how agents built into existing marketing platforms are introduced, and how teams decide what those agents may change.
- New roles: whether responsibility for operating and evaluating agents starts to settle into a recognizable role.
- Commercial queries: how answer engines handle product and buying questions, and what that means for content planning.
The open question we carry into May is whether evaluation will be treated as a shared team practice or as a specialist task. If it stays specialist, we expect it to become the bottleneck we described above. If it spreads, the pace of improvement could look quite different.
How to apply this: pick one agent you already run and write down how you would know if a change made it worse. If you cannot answer, build a small set of saved examples and a rubric before you change anything else.
Related research
How we run a research sprint on a new agent pattern
A look inside our two-week research sprint format: how we frame a question, build the smallest useful prototype, and decide what we learned.
Field notes, May 2026: protocols, evaluations, and the rise of the agent operations role
Our May notes on agentic marketing: why we expect a role to form around running agents day to day, and why permissions now matter more than connections.
From retainers to systems: replacing your agency with AI
Why the agency retainer model is breaking, and how to move to an AI-run marketing system you own.
Put this research to work
Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.
