From Puppies To Programmatic To AI Agents

Full analysis

The automation money has been pouring into the ad transaction itself for twenty years, while the manual grind between platforms (pacing checks, discrepancy reconciliation, QA) got nothing. That's the observation worth pulling from Tanja Mimica's conversation with Allison Schiff. Kovva AI is one of a crop of vendors now building agents that sit above the DSPs and do that grind. The question for operators is whether this is a real new layer of the stack or a feature that the DSPs and the holdcos absorb in eighteen months.

What's actually being decided: not "should I buy Kovva," but "is agentic middleware a durable category an agency should build its ops around, or a bridge product that gets eaten?" Easy to undo at the pilot level. Hard to undo once your institutional knowledge lives in someone else's agent. No hard deadline, but the adoption curve is moving.

The Market Analyst

Mimica's own resume is the most useful data point here. She co-founded Martin, an AI-powered DSP, in 2018. PubMatic bought it in 2022 and turned it into the backbone of Activate, their supply-path product. That is the pattern: smart independent tool gets built, proves the thesis, gets absorbed by a platform that wanted the capability but not the company. Kovva is the next iteration, and the exit logic is already visible. For a non-specialist: the plumbing vendors in ad-tech rarely stay independent, because the big platforms would rather buy the capability than let it commoditize their API. The buyers for Kovva-style middleware are the holdcos, Mediaocean, or a DSP wanting to own the workflow layer.

The Skeptic

The 500 hours a month and 4.5x ROI are self-reported by a vendor selling the tool, from one pilot of twelve users, using the agency's own time-per-task estimates multiplied by frequency. No third party audited it. No disclosed price, so the 4.5x is built on a number we can't see. Treat it as directional marketing, nothing more. The deeper doubt: Mimica admits autonomous execution for even low-risk tasks is not live yet. Every "right action" still needs a human click today. So the thing being sold is read-only briefings plus approval queues. That is useful. It is also a lot less than "agents run your media ops."

The Operator

The three-to-four week trust curve reveals exactly how steep the adoption friction is. Buyers won't grant write access until read-only briefings earn it, and that's correct. But here's what breaks at 90 days: the agent connects to ~50 tools, and every DSP and verification vendor controls its own API. The indie-agency gripe in that Digiday episode you saved is the same wall from the other side. Indies say DSPs don't want their dollars, charge fat fees, and gatekeep direct access. If DSPs are already stingy with access, they will not cheerfully expose the APIs that let a third-party agent automate around their dashboards. The platform owns the choke point. That is where this stalls.

The Customer / End User

Two customers here, and they want opposite things. The agency COO loves that institutional knowledge compounds in the agent instead of walking out the door when a senior buyer quits. Allison Schiff named the flip side directly: if the agent retains the know-how regardless of turnover, it makes buyers easier to replace. Mimica half-conceded. So the senior media buyer, the person who'd actually operate this, is being asked to pour their tacit knowledge into a system whose explicit pitch to their boss is reduced dependence on them. That tension doesn't kill adoption. It does mean the humans feeding the agent have every reason to feed it slowly.

The CFO

The build-vs-buy math is the whole decision. If an agency can replace three FTEs of ops work with a subscription, that reframes headcount planning for every media operations team. But the saved hours are the soft part of the estimate, and the hard part, the subscription price, wasn't disclosed. The durable CFO question isn't this year's ROI. It's lock-in. Once your playbooks and decision logs compound inside Kovva's agent, switching costs climb and so does the vendor's pricing power next renewal. You're trading a people cost you control for a vendor dependency you don't. That can still be a good trade. Price it with eyes open.

Where they disagree

The Market Analyst and the Operator split on whether this category survives. The Analyst sees a clean acquisition path, so the independent product is a feature of someone else's platform in waiting. The Operator says the DSPs that gatekeep access are the same ones who'd have to open it for the middleware to work, which caps how big any independent gets before a platform decides to own it.

The second split is the CFO versus the Customer. The CFO wants the knowledge locked into the agent because it compounds and defends against competitors. The senior buyer, reading the same fact, sees their own replaceability. The agency that wins is the one that resolves that tension in favor of the humans long enough to get them to cooperate.

What it hinges on

Two beliefs. First, whether DSPs and verification vendors keep API access open enough for independent agents to function, or close ranks to protect their own dashboards and workflow products. Second, whether the value sits in the agent doing work (replaceable, absorbable) or in the compounding knowledge store (sticky, defensible). The council leans toward this being a real and growing category of work, bought by operators who should negotiate data portability hard, but a shaky place to be an independent vendor. Verify the ROI with your own time-audit before you believe the hours. Nail down what happens to your playbooks and logs if you leave, before the knowledge compounds past the point of easy exit.

Prediction: At least one agentic media-ops middleware vendor sitting above the DSPs, either Kovva AI or a direct competitor in cross-platform buying automation, will be acquired by a DSP, SSP, agency holding company, or Mediaocean by the end of Q1 2027 earnings season (mid-February 2027).

Confidence: Medium — The acquisition logic is clear, but timing on any single deal is hard to pin.

Why: Mimica already lived this exact arc once: she built Martin, an AI DSP, and PubMatic bought it in 2022 to power its Activate product rather than let the capability sit outside its walls. The platforms and holdcos that gatekeep API access have both the incentive and the cash to own the workflow layer rather than let an independent commoditize it, and the agentic-ops category is hot enough right now that a cheap tuck-in looks smart. Keeping every one of these vendors independent through the whole buying season would require the platforms to ignore a capability they clearly value and have bought before, which cuts against the entire consolidation pattern of this industry.

Revisit by 2027-02-28: We're right if a DSP, SSP, holdco, or Mediaocean announces the acquisition of a cross-platform agentic media-buying-ops vendor by the end of Q1 2027 earnings season. We're wrong if no such acquisition is announced in that window.