The Pause Between Promptnand Publish.

For about eighteen months, the only question anyone was asking about AI was: can it do this?

Can it write the brief? Can it answer the ticket? Can it generate the campaign copy, the report, the forecast, the landing page variant? The answer kept coming back yes, and the yesses kept getting more impressive, and the calendars of every founder and CMO we know started filling up with demos.

Somewhere in that rush, we stopped asking the second question.

Should it do this, alone, with no one watching?

That’s the question that matters now. Not because AI has gotten worse, it’s gotten dramatically better, but because the stakes of letting it run unsupervised have finally caught up with the hype.

We’ve watched, over the past year, AI-generated content tank brand trust in weeks. AI-drafted contracts surface clauses no one approved. AI-pulled reports quote numbers that don’t exist. And in April 2026, we watched an entire internal environment at a billion-dollar infrastructure company get compromised because one employee gave a productivity AI “Allow all” access to their Google Workspace.

None of this is an argument against AI. It’s an argument against AI without a second pair of eyes.

The companies quietly winning right now aren’t the ones using AI the most. They’re the ones who figured out earlier than everyone else that the second question, should it, alone, unsupervised?, is the one worth building a process around.

The term “human in the loop” came from AI safety research. It meant: at some point in the system, a person reviews what the machine has done before it’s acted on. Dry. Technical. Easy to dismiss as something the legal team worries about.

But if you sit with it for a second, it’s actually a beautiful idea, and a surprisingly human one.

It says: this thing is powerful, but it is not a substitute for your attention. It says: move fast, but not past the point where you can still see what’s happening. It says: the loop is where the judgment lives.

Every founder we talk to right now is trying to figure out where to draw the line. How much to automate. How much to oversee. Where AI earns trust and where it hasn’t yet. There is no universal answer. But there is a principle, and it’s this:

The fastest businesses of the next decade will not be the ones that remove humans from the loop. They will be the ones that figure out, with unusual clarity, which loops the humans absolutely must stay inside.

This is what an honest AI adoption strategy looks like in practice, not a slide deck, but a series of small, sharp choices about where the humans stay.

That’s a harder strategic question than “how do we adopt AI.” It’s also a more interesting one. And it’s the one this hub is built around.

Here’s a simple exercise we run with founders and marketing leads who are working through AI adoption in real time, and trying to work out where their human attention actually earns its keep.

Take the next AI tool your team is about to adopt. Before you roll it out, ask four questions.

01

If this is wrong, who finds out first, us, or our customer?

If the answer is “the customer,” a human needs to stay in the loop. This is the email that goes out with the wrong tone. The proposal that quotes the wrong number. The support reply that sounds nothing like your brand. Anything customer-facing needs a pair of eyes before it ships, at least for now. The cost of being wrong in public is still higher than the time saved by automating.

02

If this is wrong, is it reversible?

Sending an email is reversible-ish. Publishing a blog is reversible. Deleting data, committing code, issuing a refund, sending a contract, these are much less reversible, and the loop needs to be tighter. “Allow all” permissions to core systems is the opposite of a reversible action; it creates doors you often can’t close without a full audit.

03

Does this tool need read-only access, or is it asking for more?

This is the one almost nobody asks. Most AI tools will work perfectly well with read-only, limited-scope permissions. They ask for “Allow all” because it’s easier for their engineering team, not because they need it. When a tool asks for more than read access, that’s the moment to pause and read the scopes.

04

If a new hire started tomorrow, would we give them this level of access on day one?

If the answer is no, don’t give it to the AI either. AI tools should earn permissions the way people do, slowly, as trust is established, with the ability to be revoked.

These four questions won’t make you slow. They’ll take five minutes per tool. But they will stop you from becoming the cautionary tale in someone else’s LinkedIn post.

There’s a pattern we keep noticing with the founders and marketing leaders we work with. The ones who are using AI most effectively are not the ones who adopted it fastest or loudest. They are the ones who are oddly calm about it.

They haven’t cancelled their AI subscriptions. They’re not skeptics. Most of them use AI every single day. But they’ve made one small mental shift: they treat AI like a brilliant new hire, capable, fast, worth having around, but not yet someone they’d hand the keys to the office and walk away from.

They read the permission screens. They audit the integrations quarterly. They keep one person on the team whose job includes asking “wait, should we actually be doing this?” They are, in the most literal sense, staying in the loop.

And the quiet edge this gives them is enormous. While their competitors are cleaning up AI-generated disasters, losing data, rebuilding trust with customers, or explaining to boards why the campaign went sideways, they’re shipping. Faster than before AI, actually. Just with someone still watching the door.

This is what this hub is for.

We’re going to write about AI a lot here. The tools worth using. The mistakes worth avoiding. The frameworks worth borrowing. The moments worth pausing on. Some of it will be practical. Some of it will be opinionated. All of it will be written with the same conviction we started with:

AI is the most powerful accelerator our industry has ever had. The humans are what make it safe to use it at full speed.

Stay in the loop. Literally.

If you’re reading this and wondering what staying in the loop actually looks like on a Monday morning.. here’s the smallest version of the answer.

You don’t need a thirty-page document. Most of the small and mid-sized companies we work with are better served by something closer to a single page, written in plain English, that three people on the team have actually read.

It usually answers five questions. Which AI tools are approved and why. What level of data each tool is allowed to see. Who reviews AI-generated work before it ships to a customer. What happens when someone wants to try a new tool. And who in the business owns the list and keeps it current.

That’s it. That’s the document.

If you’d rather not start from a blank page, we turned this exact thing into a free one-page AI policy template you can fill in with the team in an afternoon.

The reason this matters isn’t compliance, it’s confidence. When the team knows what’s approved and what’s not, AI adoption actually speeds up, because nobody is second-guessing whether they’re about to get the business in trouble. The document isn’t a brake pedal. It’s a map.

If you’re rethinking how the rest of your marketing stack works alongside AI, we find the same principle applies to how marketing data and analytics get handled, clarity at the start saves rework later.

How can AI actually help my business right now?

The honest answer: probably in smaller ways than the noise suggests, and in more valuable ways than most teams give it credit for. The best use we see is unglamorous, drafting first versions, summarising long documents, spotting patterns in data your team doesn’t have time to trawl through. The teams getting the most out of AI aren’t using it for big dramatic moves. They’re using it to shorten the twenty small frictions that eat their week.

How do SMEs use AI in marketing without losing their voice?

The shortcut nobody wants to hear: you brief the AI the way you’d brief a junior copywriter. Give it your tone, your past work, your actual customers’ words. Then review everything it produces with the same eye you’d bring to a freelancer’s first draft. Tools don’t protect your voice, process does. The brands keeping their voice intact through AI adoption are the ones who treat AI outputs as starting points, not finished work.

Where do we actually start with AI adoption?

Pick one painful, repetitive task your team does weekly. Automate or assist that one thing first. Live with it for a month. Then pick the next one. AI adoption that starts with a grand strategy usually stalls. AI adoption that starts with one irritation fixed tends to compound.

Written by the team at NowComms, a smart, creative B2B marketing agency. We’re not AI specialists. We’re marketers with field experience, working with growing brands, using AI where it earns its place, and watching closely as the rest of the industry figures this out too. If you want a marketing partner who thinks this way, we’d love to have a proper conversation.