Marina Saroka — Ideas

AI is not a channel. It is the operating system.

Fri, 07 Aug 2026 00:00:00 GMT

Every few years, marketing absorbs a new channel. Search. Social. Mobile. Influencers. Retail media. The playbook is always the same: open a budget line, hire a specialist, run pilots, report incremental results.

Most companies are running that exact playbook on AI right now. A tool budget. An "AI content" workstream. A pilot with a chatbot. A slide in the quarterly review.

This is the mistake.

Channels sit inside the company. Operating systems reorganize it.

A channel is a place where your message meets an audience. It changes where you spend. An operating system changes how everything runs — how decisions get made, how work gets produced, how fast the whole organization can move.

AI is the second thing pretending to be the first.

When intelligence becomes cheap — when analysis, production, iteration and personalization stop being scarce — the constraint on marketing is no longer capacity. It is judgment. The bottleneck moves from "can we make it?" to "do we know what is worth making, and can we decide fast enough?"

That shift does not fit inside a budget line. It reorganizes the org chart.

What the channel framing gets wrong

Treating AI as a channel produces predictable behavior:

Pilots that never compound. Each experiment lives in a silo, proves a narrow point, and dies in a deck.
Efficiency theater. The company celebrates producing the same output 30% cheaper, while a competitor redesigns the workflow to produce something the old process could never make.
Specialist bottlenecks. One "AI person" or one "AI agency" becomes the gatekeeper for a capability that should be ambient, like spreadsheets or email.

The companies that win will not be the ones with the best AI pilots. They will be the ones that rebuilt their default way of working while everyone else was piloting.

What AI-first actually means

AI-first is not "we use AI a lot." It is a design principle: when you design a process, a team or a product, you start from what intelligence-on-demand makes possible, instead of retrofitting it onto how things were done before.

In practice that looks like:

Strategy cycles measured in days, because research, synthesis and scenario work no longer take weeks.
Creative teams that produce systems — prompts, pipelines, taste — rather than individual assets.
Brands designed to be legible to machines as well as humans, because machines increasingly do the reading.
Fewer handoffs, because the handoff was always the tax, and AI removes the reason most of them existed.

None of this is about tools. Tools change monthly. It is about where the company assumes scarcity lives. The old assumption — production is expensive, so plan carefully and produce little — is dead. The new scarcity is judgment, taste and speed of decision.

The uncomfortable part

An operating system migration is not something a marketing department can do alone, and that is precisely the point. If your AI conversation still lives entirely inside marketing, it has not really started.

The question to ask is not "what is our AI strategy?" It is: "which of our current assumptions about cost, speed and team structure stop being true this year — and what do we do about it before someone else does?"

That question does not belong in a channel plan. It belongs on the CEO's desk.

Marketing to machines

Wed, 15 Jul 2026 00:00:00 GMT

For a hundred years, marketing has optimized for one reader: a human with limited attention, strong emotions and a short memory.

A second reader has arrived. It has unlimited attention, no emotions, and perfect memory.

When someone asks an AI assistant "what's the best project management tool for a small agency?" or "find me a good hotel in Lisbon under €200", the assistant reads dozens of sources, compares options and returns a shortlist. The human chooses from three, not three hundred. Increasingly, the human doesn't even see the long list.

That means a growing share of purchase decisions now includes a step where your brand is evaluated by software — and most brands have no idea how they perform in that evaluation.

The new funnel has a machine in the middle

The classic funnel assumed your message reached a person. The emerging funnel often looks like this:

A person delegates a task to an agent.
The agent searches, reads and compares.
The agent produces a recommendation.
The person picks from the shortlist — or approves the agent's default.

Everything brands spend on step 1 visibility is wasted if they fail at step 2. And step 2 rewards different things than human attention does:

Structured, verifiable claims beat emotional taglines. An agent can't feel your brand promise, but it can check your pricing page, your reviews, your documentation, your return policy.
Consistency across sources beats a great campaign. Agents cross-reference. If your site, your reviews and third-party mentions disagree, you read as unreliable.
Machine legibility beats production value. Clean structure, schema markup, clear language and honest specifications are now conversion assets.

Answer engine optimization is not SEO with a new name

SEO optimized for ranking on a results page a human would scan. This is different: there is no page. There is an answer. You are either in it or you don't exist for that query.

Being in the answer depends on whether AI systems have a clear, consistent, corroborated model of who you are, what you do and what you're good at. That model is built from everything public about you — which means it can be deliberately shaped, or left to chance.

Most brands are leaving it to chance.

What to do about it

Three moves matter now:

Become a clean source of truth about yourself. Structured data, unambiguous language, one canonical version of the facts. If a machine reads everything you publish, would it come away with a precise answer to "what is this company and when should I recommend it?"

Earn corroboration. Agents weigh independent confirmation. Reviews, citations, third-party coverage and genuine expertise published in public are worth more than another paid placement.

Instrument the new funnel. Ask the major assistants the questions your customers ask. Look at what they say about you and about your competitors. That transcript is the new brand tracker.

The brands that treated Google's crawler as an audience in 2005 owned a decade of demand. The brands that treat AI agents as an audience now are making the same bet, at the same moment, with the same asymmetry.

The second reader is already reading. Write for it.

The agency model isn't dying. It's inverting.

Sat, 20 Jun 2026 00:00:00 GMT

Every conversation about AI and agencies eventually arrives at the same question: "will clients still pay us when they can generate this themselves?"

It's the wrong question, because it assumes the agency's product was ever the deliverable.

What agencies actually sold

The billable unit was the hour, and the visible output was the asset — but what good agencies actually sold was a chain of decisions: what to say, to whom, in what tone, with what idea, and why this and not that. Production was how those decisions became tangible, and because production was expensive, it was also how they were priced.

AI collapses the cost of the tangible part. It does not collapse the cost of the decisions. If anything, it raises it: when anyone can produce infinite competent options, knowing which one and why is the entire game.

That's the inversion. The thing that was priced (production) becomes nearly free. The thing that was given away in the pitch (judgment) becomes the product.

Why "AI makes us more efficient" is a trap

The default agency response to AI has been margin defense: same deliverables, faster, cheaper, quietly pocketing the difference. It works for about two budget cycles — until procurement notices what production actually costs now, and reprices the relationship accordingly.

An agency that uses AI to produce the old outputs faster is competing on a curve where the price trends toward zero. An agency that uses AI to change what it can take responsibility for is competing on a different curve entirely:

Not "we'll make your campaign" but "we'll run the system that makes, tests and evolves your campaigns continuously."
Not "here are three routes" but "here are three routes, each already validated against your audience, your data and the agents that will read it."
Not a deck about the consumer, but a working prototype in the client's hands during the first meeting.

The new shape

The inverted agency looks structurally different:

Smaller, senior, faster. The pyramid existed to make production scale. When production scales by itself, the layers whose job was coordination lose their reason to exist. What remains is the expensive part: people with taste, strategic judgment and the ability to direct machines.

Systems as deliverables. The most valuable thing an agency can now leave behind is not a campaign but a capability — pipelines, prompts, evaluation criteria, a way of working the client didn't have before.

Proof over promise. When a prototype costs a day instead of a quarter, the pitch deck is a confession that you didn't bother. Showing beats telling, and AI has made showing almost free.

The uncomfortable symmetry

Clients face the same inversion. In-housing production was a cost decision; it now costs almost nothing, so it will happen by default. What clients cannot in-house by buying software is judgment that has seen a hundred categories, taste that has been tested under pressure, and the outside view.

Agencies that sold hands are competing with a model that costs cents per hour. Agencies that sell heads have never had less competition — because the noise of infinite competent content makes distinctive thinking easier to recognize, not harder.

The model isn't dying. It's being turned inside out. The question for any agency is simply which side of the inversion its revenue is on.