A Reddit thread has been circulating among performance marketers. It is, depending on your tolerance for industry self-pity, either the most honest document of agency life in 2026 or simply a long exhale held since 2023. The list is familiar: clients comparing five-figure deliverables to five-minute prompts, attribution windows that no longer narrate cause and effect, CPMs that climb regardless of how clever the copy is, top talent leaving for in-house seats, and a quiet drift of execution work back inside the client's walls.

The temptation, reading it, is to treat the thread as a lament about artificial intelligence. That would be a mistake. AI is a participant in the story, not its author. What the thread actually documents is the unwinding of a particular bargain, one that agencies and their clients made twenty years ago, when digital channels were new and the work of running them was genuinely scarce. The bargain assumed that doing the work, which meant building the campaigns, writing the copy, and pulling the levers inside Meta and Google, was the agency's product. Strategy was the soft-tissue add-on; execution was the load-bearing wall.

That bargain is now visibly failing. And the question worth asking is not "how do agencies survive AI?" but the older, harder one underneath: what was the agency actually selling, and what part of it survives a world where execution is cheap?

I.The slow commoditization of execution

To understand what is happening to agencies, it helps to remember that the digital marketing industry was built on a temporary inefficiency. In 2008, you could not run a Facebook campaign without specialized labor. In 2014, you could not stand up a programmatic pipeline without a team. As recently as 2020, integrating attribution across paid social, paid search, email, and CRM was a real engineering problem. Agencies sold themselves as the solution to this thicket, and they were right to. The work was hard, the talent was scarce, and the platforms were unforgiving.

What has changed since 2023 is not that any single one of those tasks has become easy, but that the marginal cost of competent execution has collapsed. A mid-level marketer with ChatGPT, Claude, and a stack of point tools can now produce, in an afternoon, deliverables that once required a four-person pod. The work is not as good as the best agency work. But it is good enough to make a CFO pause when the renewal invoice arrives.

This is, importantly, not a story unique to marketing. It is the same story that has played out in legal research, in junior software engineering, in financial analysis, in graphic design. Whenever a profession's value proposition rests primarily on the doing of repetitive cognitive work, AI compresses the price of that work toward the price of the model call. The middle layer collapses. What remains is judgment at the top and commodity output at the bottom.

The Reddit thread captures this with unusual clarity:

The agencies feeling this most acutely are the ones that built their model on the middle: not boutique strategy shops, not lowest-cost execution arms, but the broad mass of full-service performance agencies whose pitch deck has, for the better part of a decade, said some version of "we run your channels better than you can." That sentence is increasingly hard to defend.

II.The attribution argument that no longer ends

There is a second pressure on agencies that gets less attention than AI but may matter more in the medium term. For most of the digital era, agencies defended their fees with a particular kind of artifact: the attribution report. The report told a story. This many impressions led to this many clicks led to this many conversions led to this much revenue, and the story justified the retainer. When the story was convincing, the renewal was easy.

That artifact has been quietly hollowed out. Apple's privacy changes in 2021 were the first incision; the gradual deprecation of third-party signals has continued the cut. What clients now receive is not a story but a model. It is a probabilistic estimate, hedged with disclaimers, that says something like "we believe these channels contributed to this lift, with these confidence intervals." A CFO reading that sentence is in a different conversation than a CFO reading a clean attribution waterfall in 2017.

This matters because it changes the texture of every renewal conversation. The agency used to be able to point at the dashboard. Now it has to argue. And in a year where the same client is also testing ChatGPT against the agency's deliverables, the argument arrives already weakened.

It is worth being honest about the asymmetry here. AI did not create the attribution problem. The platforms did, and they did it deliberately, to consolidate measurement inside their own walled gardens. But AI is a useful villain because it is more legible than ad-tech policy, and because it sits in the room with the client every day. The agency gets blamed for both at once.

III.The thing nobody on the call wants to say

The hardest sentence in the agency conversation, the one that does not get spoken in QBRs because saying it ends the meeting, is this: if your value can be replicated by a prompt, you don't have defensible value.

That sentence is uncomfortable in proportion to how much of the agency's deliverable schedule it describes. For a great many performance agencies in 2026, an honest audit of what the retainer actually buys would land somewhere in the neighborhood of: media buying labor, creative variants, weekly reporting, channel hygiene, and cadenced strategy conversations. Each of those line items has a credible AI-shaped substitute now. Not a perfect one. A good-enough one. And good-enough is what the renewal is fighting against.

So the question agencies have to answer, the one the Reddit thread is circling without naming, is not "how do we adopt AI faster than our competitors?" Most agencies have already adopted AI. The internal margin lift is real. The problem is that the same tools are available to the client, and the client is using them to shrink the scope of work, not to expand it.

The actual question is: what does an agency sell that the client cannot replicate, that AI cannot produce, and that improves with the agency's involvement rather than degrading toward parity?

There are a few honest answers. Genuine creative leadership is one, but it is rare, hard to scale, and concentrated in a small number of shops. Senior strategic judgment in a specific vertical is another, but it lives in individual heads, and those heads keep going in-house. The third answer, and the one we want to spend the rest of this piece on, is less discussed and more durable. It is the audience layer.

IV.Why audience-building is the part AI cannot do

Here is a distinction worth holding onto. AI is extraordinarily good at generation — producing artifacts that resemble training-distribution outputs at very low marginal cost. It is much worse, sometimes embarrassingly worse, at identification — finding the specific real-world entities that match a complex, signal-dependent description, with verified accuracy, at scale.

That distinction matters because almost everything an agency produces on the creative side falls into the first category, and almost nothing falls into the second. A model can write an ad. A model cannot, on its own, tell you which 3,400 dental practices in three states are about to expand because they just hired a regional manager and signed a lease in a new market. A model can draft an email sequence. It cannot tell you which 1,200 mid-market companies just rolled out a specific HRIS platform and are therefore vulnerable to a competitive pitch in the next ninety days.

Those questions are not generation problems. They are identification problems — and they require, to answer well, three things AI alone does not provide: a continuously refreshed signal layer about real companies and people, a verification process to separate signal from noise, and a domain-specific understanding of which signals actually correlate with intent.

This is the part of the marketing stack that has been quietly underbuilt for a decade. Most agencies, and most clients, treat audience-building as a setup step — a thing you do in Ads Manager before the "real work" of campaign optimization begins. You pick a lookalike, you layer on some interests, you set a geo, you go. The audience itself is treated as a commodity input.

When audiences are commodity inputs, execution determines outcomes. And when execution gets commoditized too, the entire stack becomes commoditized. This is precisely the trap agencies are now sitting in.

The escape is structural. Agencies that survive the next two years will be the ones that move the locus of their value from what they do with the audience to which audience they can build in the first place. The campaign becomes the downstream consequence of a defensible upstream advantage, rather than the place where all the differentiation is supposed to happen.

The reframe, in five lines

V.What an audience-led agency actually does

To make this concrete: an agency operating on an audience-first thesis spends its early hours of a new engagement differently than a traditional performance shop. Instead of opening Ads Manager and inheriting whatever the client was already targeting, it asks a different opening question. Who, specifically, is the buyer we want — and what real-world signal would tell us they are about to be in market?

The answers are usually unglamorous and specific. A franchise services company wants to reach owner-operators of multi-unit franchises with at least three locations and recent expansion activity — not "small business owners interested in business." A cybersecurity vendor wants accounts running a particular EDR product whose contract is up for renewal in the next two quarters — not "IT decision-makers." A hospitality SaaS platform wants independent properties that just changed PMS, where the new system is a gateway to a competitive switch — not "hotel managers."

None of those audiences exist as a checkbox in any ad platform. They have to be built — assembled from firmographic data, technographic signals, hiring activity, web behavior, and a layer of human verification that distinguishes a real signal from a stale record. This is genuinely difficult work. It is also, importantly, work that compounds: the audience the agency builds for one client informs the methodology for the next, and the methodology becomes a moat the way a brand or a creative library used to.

Once that audience exists, the campaign work changes too. It is no longer a search for incremental optimization on top of a noisy input. It is the activation of a clean, high-intent signal across the right channels, with reporting that ties pipeline back to the specific audience cohort rather than to a generic platform-attributed conversion. The math gets clearer because the input got cleaner.

This is, not coincidentally, the model we have spent the last several years building toward at LeadGenius. Our role on most engagements is to be the audience layer for agencies and direct teams who have realized that the campaign is downstream of the targeting, and that the targeting is downstream of the data. AdScope sits inside that thesis as the competitive intelligence layer — telling you what your client's competitors are actually running, against whom — and AdGenius is how that audience gets activated and orchestrated across channels rather than left in a CSV.

VI.The agencies that make it through

The Reddit thread reads as a list of grievances, but it is really a forecast. The agencies that emerge from the next eighteen months in stronger shape will not be the ones with the cleverest AI workflow or the most aggressive cost cuts. They will be the ones that quietly rebuilt the answer to a single question — what do we sell that cannot be prompted? — and made the answer load-bearing in their pitch.

The honest version of that answer, for most performance agencies, is going to involve audience data. Not because audience-building is the only thing AI cannot do, but because it is the thing AI cannot do that sits closest to the agency's existing capability set, that maps cleanly onto the campaigns they are already running, and that produces a pipeline result the CFO can see. It is, in the language of the original Reddit complaint, the thing that turns "we run your channels" into "we change what your channels see."

None of this is a triumphant story. The compression is real. Agencies will get smaller before some of them get more durable. The middle will continue to hollow out. But the path through is visible, and it has the unusual virtue of being honest: stop selling the work AI is about to do for free, and start selling the input that makes the work actually pay off.

That is a harder sentence to put on a slide than "we leverage AI to scale your campaigns." It is also, for the first time in a long time, a defensible one.

LeadGenius is a B2B audience intelligence platform. We build the custom, signal-verified audiences agencies and in-house teams use when prebuilt databases stop performing — and activate them across channels through AdGenius. The Reddit thread referenced in this essay is a composite drawn from publicly shared discussion among agency operators in 2026.