The future of the email team
An AI reads most email before a person does, and increasingly a second AI acts on that email from outside the mailbox entirely. Both shifts are already in production across the roughly three billion Gmail users who entered what Google calls the Gemini era in January 2026, where a thread summary can become a reader's only contact with a message. The reading problem is already stark: 85% of commercial messages are never read, and nearly 90% of deletes happen without the message ever being opened. So the interesting question isn't whether AI sits in the inbox. It is what the people who send email for a living do once it does.
What follows is a position, not a survey. The email team isn't shrinking, it is moving up a layer, and the teams that see the move early will still be here in five years. The reflex read is that AI eats the email team: generate the copy, generate the subject line, generate the layout, ship it, done. That read is wrong, and wrong in a specific way. Generation was never the hard part of enterprise email. Producing a message was cheap even before AI made it cheaper. The hard part was always coordination, getting a hundred marketers across a dozen regions to produce on-brand, launch-ready assets that render correctly, pass legal, respect the approval chain, and sync cleanly to the marketing automation platform. AI lowers the cost of the easy half and does very little to the hard half. So the work that survives is the coordination work, and the people who do it move from making messages to governing the content machines read.
Humans stay in the loop, on purpose
There is a comfortable story where agentic AI takes over the send, the human steps back, and the inbox runs itself. Enterprise marketing will not run that way, and not out of technophobia. The failure modes of an AI acting on a mailbox are not cosmetic. An email body can carry an indirect prompt injection that quietly moves data out, a hosted connector can be poisoned after you have granted it trust, and an over-broad permission scope turns a helpful agent into a confused deputy holding the keys to the whole mailbox. Those are governance problems, and governance problems get solved by people who own the decision, not by removing the person from the loop.
The posture Knak has argued across this series is the workable one: the AI recommends, the human approves, and the audit log records both. That isn't a hedge against progress. It is what production-grade means when the output goes to millions of recipients under a brand's name. A useful test for any vendor selling autonomy is to ask what happens when the input data is incomplete. If the answer is that the agent proceeds, that is a liability. If the answer is that it flags a human and waits, that is a system you can put in front of an enterprise. Tiered governance is the shape this takes in practice: routine work flows freely, exceptions get review, and the team's job becomes deciding where that line sits and holding it.
The MAP settles into a system of record
The marketing automation platform is quietly changing jobs, and this part is a directional read rather than a vendor announcement. For twenty years the MAP was where marketers did the work: built the email, built the landing page, built the nurture. That was always an awkward fit, because MAP builders are painful at scale and the creation experience inside them was never really the point. As AI-native production layers take over the making of assets and structured connectors take over the moving of data, the MAP settles into what it was always best at: sending, storing the record, holding the data of who received what and what happened next.
That shift matters for the team because it moves where the craft lives. When the MAP is a system of record rather than a system of creation, value moves to the layer that produces the structured content feeding it, and to the people who govern that layer. The relationship stops being "we replace your MAP" and becomes plainer: your MAP sends, and the production layer produces. The team's center of gravity moves with it, away from wrestling assets into a MAP editor and toward defining the templates, principles, and controls that let a machine assemble launch-ready output correctly the first time.
Structured content becomes the job
If an AI is the first reader on the way in and an agent is a possible actor on the way out, the content itself has to be legible to a machine, and that is now a core competency rather than an accessibility footnote. The series made the mechanism concrete: providers parse email for structure, and the 47% of email images that ship with empty alt text are content a machine simply cannot read. Google's own research found that reading the text inside images lifted offer detection by 9.12%, a direct measurement of what machine-readability buys a sender. The code is the data. The decorative HTML is not. A team that internalizes that stops optimizing for how an email looks in a screenshot and starts producing for how it parses.
This is the through-line of the whole series. Machine-readable, structured content isn't one tactic among many, it is the durable advantage in an AI-mediated inbox, because every downstream behavior, the summary, the ranking, the classification, the agent's action, depends on a machine reading the content correctly. That is why the email team's new job description is less designer of messages and more producer and governor of structured content at scale. The craft doesn't disappear. It moves from pixels to structure, from how the asset looks to the integrity of the blueprint underneath it.
This is where the production layer earns its place. A team that has to hand-govern structure across hundreds of assets and dozens of regions needs a system that bakes the structure in, so brand principles, accessibility, and machine-readability are properties of the template rather than things a reviewer catches by hand. That is the argument for a platform like Knak: it lets a distributed team produce structured, on-brand, launch-ready content without a developer queue in the middle, which is exactly the coordination work that survives the AI shift. The claim holds without the product name. The product is just the cleanest way to do the thing the argument says now matters most.
Stop treating the inbox as the whole channel
The last part of the position will land oddly for an email team, so here it is straight. If a growing share of value reaches people through summaries they never open and previews the provider generates, then opens and clicks are decaying signals, and building an entire strategy on a channel measured by decaying signals is a bet against your own instruments. Click-to-open is already down more than 8% year over year in some datasets, while the provider's ranking now moves the numbers directly: after Gmail's 2025 Most relevant Promotions sort, top above-the-fold slots saw 5 to 15% higher opens and 13 to 17% fewer unsubscribes. The inbox isn't dying, but it is no longer safe to treat it as the whole channel. A team that hedges by producing structured content that travels, across email, landing pages, and whatever surface the next AI assistant renders, outlasts any single provider's ranking algorithm.
That hedge isn't a retreat from email. The same structured content the inbox now rewards is exactly what makes content portable to other surfaces. Produce for the machine once, and the output is legible everywhere a machine reads: the summary, the landing page, the agent's answer. The team that owns the structured production layer isn't betting on email or against it. It is betting on being readable wherever the reader, human or machine, actually shows up.
None of this asks the email team to become smaller or less central. It asks the team to move up a layer, from making messages to governing the structured content machines read, rank, and act on, and to treat the inbox as one important surface rather than the entire game. The teams that make that move early will spend the next five years doing the coordination work AI can't touch, on a production layer built for it. To see what producing for that world looks like in practice, book a demo of how Knak builds structure in from the first draft.
