Marketing Without Marketers? Julius Körfgen on Autonomous AI, Growth, and the End of the Marketing Stack

Julius Körfgen, co-founder and CEO of Uplane, went on Signal & Noise with hosts Brett House and Rio Longacre to make a plain claim: independent performance agencies are about to be automated out of existence. Uplane generates the creative, buys the media, shifts the budget, builds the landing page, and reports back in a loop that runs in hours instead of the three-to-four months a large enterprise takes today.

The most interesting structural point came from House, not Körfgen: AI is collapsing the cost gap between software and services. That is the whole pitch. Managed media at software-like margins is why Y Combinator, Play Ventures, 20VC, and Rebel Fund put $4.5M into what is functionally a trading desk. Whether that thesis holds depends on retention past the launch spike, not the funding announcement.

The evidence, though, is entirely self-reported. "Half of ad spend is wasted" is Wanamaker repackaged as a market-size stat. "Better in almost all cases" against incumbents carries no third-party attribution. The brand-compliance score was set up by the client and vendor together. The 10 to 15 percent dynamic creative optimization uplift is unaudited. None of that means the product fails. It means nothing here has been graded by anyone but the seller.

The operational claim worth taking seriously is ingesting a 200-to-500-page brand book and running real-time compliance checks. That is genuinely where agencies struggle with regulated clients like Deutsche Bank. But when the system learns from approvals to reduce human review, and shifts budget overnight before anyone looks, you have two autonomous optimization layers running on top of each other: Uplane's and the platform's own black-box system. The failure mode is not bad creative. It is no one able to explain what happened to ROAS by Tuesday morning.

The pricing model is a fixed retainer plus a percentage of managed spend. That is a trading-desk arrangement, and the transparency fight that comes with it is not new. When one vendor controls creative, channel selection, budget allocation, and takes a cut of spend, the incentive to spend more is structural. The industry already rejected this structure once. The pilot wins deals; the pricing model invites the backlash that cost trading desks their trust.

The three-month side-by-side pilot is a genuinely hard offer for an incumbent agency to counter. No traditional agency can credibly volunteer to compete against itself for free. But displacing an agency of record at an enterprise is a legal, compliance, and relationship decision, not a ROAS spreadsheet. Procurement and risk functions at regulated clients move on multi-year trust cycles, not seed-stage scorecards.

The category is forming. The winner is not decided. Meta and Google own the inventory Uplane is renting, and both are building this capability in-house. A startup layering on top of the platforms is racing against the landlord.

Our call: no top-six agency holding company will publicly lose a named enterprise B2C account to Uplane or a comparable autonomous media-buying startup before the end of Q2 2027. Enterprise procurement moves slower than a seed-stage pitch. Revisit by 2027-06-30.

Julius Körfgen, co-founder and CEO of Uplane, went on Signal & Noise with hosts Brett House and Rio Longacre to make a plain claim: independent performance agencies are about to be automated out of existence, and Uplane is the thing doing the automating. AI generates the creative, buys the media, shifts the budget, builds the landing page, and reports back, in a loop that runs in hours instead of the three-to-four months it takes a big enterprise today. The pitch is a full-stack replacement for the "non-AI-native" agency.

What's actually being decided here, for the reader: whether to treat autonomous media buying as a real competitive threat now or a demo that stays a demo. This is a Type 2 call. Reversible. You can watch the pilots run at named accounts before you rewire anything. No forcing function this quarter. The forcing function is when a client you serve runs one of these side-by-side pilots and doesn't renew you.

The Market Analyst. The structural point came from Brett House, not Körfgen: AI token and infrastructure costs are narrowing the margin gap between software and services. That's the whole game. For years the clean SaaS business and the messy managed-service business sat in different valuation buckets. If AI lets a vendor run the service at software-like margins, the hybrid model, retainer plus a cut of spend, gets a lot more attractive to build and to fund. That's why Y Combinator, Play Ventures, 20VC, and Rebel Fund put $4.5M into a company doing what looks like a trading desk. In plain terms: investors are betting managed media can finally be run like software. Watch whether Uplane's ARR curve past $1M holds, because the funding thesis lives or dies on retention, not the launch spike.

The Skeptic. Every number in this pitch is self-reported. "Half of ad spend is wasted" is a paraphrase of a Wanamaker line from the 1800s, repackaged as a market-size stat. "Better in almost all cases" against incumbents, no third-party attribution. "Scores higher on brand compliance than the client's own design team," a score the client and vendor set up together. The 10 to 15 percent DCO uplift is unaudited. For this to work, a buyer has to trust a vendor's own scoreboard for a vendor's own pilot. Körfgen has every incentive to make the wasted-spend problem sound enormous. None of that means the product is bad. It means nothing here has been graded by anyone but the seller.

The Operator. Ingesting a 200-to-500-page brand book and running real-time compliance checks is the interesting operational claim, because that's the part agencies actually struggle with at Deutsche Bank and pharma clients. But think about Tuesday morning at 90 days. The designers approve AI ads on a Pinterest-style board; campaign managers approve budget shifts in a ticketing tool. The system "learns from approvals to reduce human-in-the-loop friction." That's the risky seam. Who owns the outcome when the loop shifts budget into a channel overnight and ROAS craters before a human looks? On Meta and Google the platforms already run their own black-box optimization. Layering a second autonomous layer on top means two systems making bets you can't fully inspect. The failure mode isn't bad creative. It's nobody able to explain what happened.

The Customer / End User. Two customers here. The brand CMO gets a low-friction trial: three months, side-by-side, outcome-based. That's a genuinely hard offer for a traditional agency to counter, because the incumbent can't credibly say "let me compete against myself for free." The other customer is the agency itself, and Körfgen is blunt that operational roles get automated while survivors move to brand strategy, storytelling, and change management. For a plain-English reader: the person building the campaign gets replaced by software; the person deciding what the brand stands for keeps a job. Whether brands actually want a vendor holding both the creative and the media buying, with a cut of spend, is the open question. That's the principal-agent problem the industry spent a decade cleaning up on trading desks.

The CFO. The pricing model is a fixed retainer plus a percentage of managed ad spend. That is a trading-desk arrangement, and the transparency fight that comes with it is not new. When the same vendor makes the creative, picks the channel, moves the budget, and takes a percentage of the spend, the incentive to spend more is baked in. An operator evaluating this, or building a competing managed-service attachment, should assume renewed scrutiny of principal versus agent, and should assume clients will eventually ask for independently audited performance. The margin math Brett House flagged is real and it cuts both ways: cheaper AI service delivery is exactly what lets a competitor undercut your agency fee, too.

The tensions. First: the Market Analyst sees a genuine structural shift in service margins, while the Skeptic sees a pitch with zero audited evidence. Both can be true. The margin shift is real and the specific vendor's numbers are still unproven. Second: the Customer sees an offer agencies can't match, while the CFO sees the same principal-agent conflict the industry already rejected once on trading desks. The pilot wins deals; the pricing model invites the transparency backlash that eventually cost trading desks trust. Third, quieter one: Körfgen dismisses Meta's own end-to-end agentic ad system as incremental DCO and points at the metaverse spend as proof Meta overpromises. Maybe. But Meta owns the inventory, the delivery system, and the audience data. A startup layering on top of Meta is renting the exact capability its biggest competitor is building in-house.

What this hinges on. Three beliefs. One, that AI actually collapses service margins toward software margins at scale, not just in a seed-stage demo. Two, that autonomous budget-shifting produces outcomes a buyer can trust without a human gate, which is unproven and where the compliance-score claim matters most. Three, that brands will accept a single vendor owning creative plus buying plus a spend cut, after the industry already learned to distrust that structure. The council leans skeptical on near-term ecosystem impact and genuinely interested in the structural read. This is a real category forming, not a real winner yet.

For the broad reader: demand a pilot graded by a third party before treating autonomous media buying as a live threat, not by the vendor. And the platforms, Meta and Google, are the ones actually positioned to own this, because they hold the inventory the startups are renting.

Prediction: By Signal & Noise's coverage cycle over the next two agency budget-review seasons, no top-six agency holdco will lose a named enterprise B2C account publicly to an autonomous AI media-buying startup on Uplane's pilot model before the end of Q2 2027.

Confidence: Medium. Enterprise procurement and trust cycles move slower than seed-stage pitch decks.

Why: The signal in this episode is that all of Uplane's evidence is self-reported, and its entry point is unaudited three-month pilots at accounts like Deutsche Bank, exactly the regulated, procurement-heavy clients that move slowest and demand third-party validation before firing an incumbent. The mechanism is that displacing an agency of record is a legal, compliance, and relationship decision, not a ROAS spreadsheet, and enterprises rarely hand creative plus buying plus a spend cut to a seed-stage vendor without a multi-year trust build. The opposite outcome, a public named-account displacement inside 18 months, would require an enterprise to override its own procurement and risk functions on the strength of a vendor's own scoreboard, which almost never happens at that speed.

Revisit by 2027-06-30: We're right if no top-six holdco has publicly lost a named enterprise B2C account to Uplane or a comparable autonomous media-buying startup on the pilot-to-displacement model. We're wrong if any such displacement is publicly announced at a named account.

The category is real and worth watching. The winner isn't decided, and the smart money question is whether Meta and Google simply absorb this capability into inventory they already own before a startup gets to scale.