Hiring Gigi, Your AI Media Manager

Full analysis

Adam Epstein, the co-founder and CEO of Gigi and the guy who sold Perpetua to Ascential for a little over $100 million in 2021, went on Aperiam to make a specific bet. His company builds an AI agent that runs Amazon DSP campaigns end to end, and he thinks the way to win the agentic ad-tech land war is to go deep in one channel before going wide. That is the opposite of the "unified media platform" pitch everyone else is running.

Here is what actually gets decided by listening to this: whether agencies staff up on Amazon DSP with scarce humans, build their own tooling, or rent an agent. That choice is easy to undo at first (you can cancel a contract) but hard to undo once the agent has been running your campaigns for a year and holds the accumulated know-how. So the real question is who owns the intelligence the agent builds. Nothing here forces a deadline except the pace of Amazon's ad business, which Epstein pegs at almost $80 billion growing 20-plus percent a year.

The Market Analyst. The interesting structural claim is Epstein's, that the UMP category winner will be worth more than any current third-party ad-tech company that sits on the walled gardens. Read that against who is on the board: The Trade Desk, Criteo, and every independent DSP built to be channel-neutral. Epstein is saying neutrality is a weakness, because you can only sell outcomes if you control the full data plane in a channel, and you only get that by going all-in on one walled garden at a time. For a smart generalist: he thinks the money is in becoming Amazon's best contractor, not in staying above the fray. If he is right, the "we buy everywhere equally" pitch loses its premium.

The Skeptic. The 1,000-experts number is Epstein talking his own book. A smaller talent pool makes Gigi look like the only option, and there is no methodology behind it. Same with the 80% cost cut, self-reported, no audit, no case study. Steelman the case against Gigi: if Amazon DSP expertise is genuinely scarce and valuable, Amazon has every reason to build the agent itself and hand it to agencies free, the way it gives away tools to buy more data. Epstein's answer is the Snowflake analogy, a multi-billion-dollar company built on AWS while Amazon ran Redshift. Fine. But Snowflake owned the data layer. An agent that only pushes buttons inside Amazon's DSP owns nothing Amazon cannot replicate.

The Customer / End User. The customer here is the enterprise agency, and the thing that will actually move a procurement team is "learned context." Epstein put a defined term in the contract: the know-how the agent builds by running your campaigns under your playbook is yours, containerized, and exportable if you leave. That solves the real fear, which is renting back your own institutional knowledge from a vendor. Every agency that has been burned by a platform holding its data hostage understands this immediately. It is the most durable thing in the whole conversation, because it is a legal commitment, not a performance claim.

The Operator. Tuesday-morning reality: an agent that claims 100% task coverage inside Amazon DSP still breaks the first time Amazon ships a UI change or a new beta. Epstein knows this, which is why he argues you must cover every task in one channel before expanding. The second-order effect at 90 days is that the agencies pulling Gigi into "please add more channels" are the ones who will strain the depth-first thesis. Customer pull toward breadth is exactly the pressure that turns a deep product into a shallow one. That same tension is what makes most of the UMP map, in Epstein's words, legacy SaaS with an AI wrapper.

The CFO. The economics only get interesting at the endgame Epstein describes, outcome-based pricing, charging for incremental sales instead of software seats. He says that is two years out. Until then this is a labor-substitution buy: replace or augment a scarce, expensive Amazon DSP seat with a subscription. The payback math is clean if the 80% number is anywhere near real, and dangerous if it is not, because you have shed the humans who knew how to catch the agent's mistakes. The bigger cost nobody models is Amazon competing with you, which would compress the whole category's pricing overnight.

Where the council splits. Two real disagreements. First, is depth-first a moat or a trap? The Market Analyst buys the thesis that channel depth is where outcome-based value lives. The Operator sees customer demand for breadth pulling every deep player wide before the depth ever pays off. Second, does Gigi own anything Amazon cannot take? The Skeptic says an agent that only operates inside Amazon's own platform is a barnacle Amazon can scrape off. The Customer says "learned context" is the one thing that is genuinely the buyer's, not Amazon's and not Gigi's, and that is what makes the product sticky.

What this hinges on. Three beliefs. One, that Amazon DSP expertise stays scarce enough that renting an agent beats hiring. Two, that Amazon chooses to let a third-party agent thrive on its platform rather than building its own. Three, that "learned context" as a contract term actually survives an enterprise legal review and gets copied by competitors. The council leans toward the "learned context" idea mattering more than any single product. That framing, the buyer owns the intelligence the agent accumulates, is the part of this that outlives Gigi.

The thing to verify before betting on any depth-first agentic vendor: does the contract actually make the accumulated know-how exportable, and has anyone left with it intact? A claim of ownership is worthless until someone walks out the door holding it.

Prediction: At least one channel-neutral independent DSP or ad-tech platform (The Trade Desk, Criteo, or Viant) will publicly announce a purpose-built Amazon-DSP or retail-media agentic buying feature by the Q3 2027 earnings season in October 2027.

Confidence: Medium. The pull toward Amazon's channel is real, but timing depends on each vendor's roadmap.

Why: Amazon's ad business at almost $80 billion growing 20-plus percent a year is the fastest-scaling programmatic channel, and Epstein's whole pitch rests on agencies being unable to staff it with humans. That staffing gap is not unique to Gigi's customers; it hits every buyer, which means the channel-neutral platforms that sell to those same agencies will get the same "help me run Amazon" request. The independents already build channel-specific tooling to stay relevant, so adding an Amazon-DSP agent is the obvious defensive move once a well-funded specialist like Gigi shows demand exists. The opposite outcome, everyone staying purely horizontal, loses because horizontal breadth without depth in the channel where the money is growing fastest is exactly the weakness Epstein identified, and buyers will vote with their spend.

Revisit by 2027-10-31: We're right if The Trade Desk, Criteo, or Viant announces a named Amazon-DSP or retail-media agentic buying tool by the Q3 2027 earnings season. We're wrong if none of the three ships such a feature by then.