How InMobi Is Reinventing Advertising with AI Agents

Full analysis

At Cannes, InMobi CEO Abhay Singhal laid out a bet that AI agents will hollow out the middle of the marketing funnel — and, more provocatively, that the traditional demand-side platform (DSP, the software buyers use to purchase ads) is on its way out, replaced by AI agents talking directly to an ad exchange. He's backing the vision with a new "Agentic Selling Platform," a Samsung TV shopping deal, and an Apple Search Ads acquisition.

What's actually being decided by operators here: not "should we believe InMobi," but how fast the agent-to-exchange plumbing and funnel-compression story becomes something you have to staff for. This is a Type 2, reversible question for most operators — you can watch and move — but the underlying architecture shift is Type 1 if it takes hold. No hard forcing function beyond the competitive clock.

The Market Analyst

Follow the margin. InMobi is a supply-side player arguing the buy-side software layer is obsolete — of course it is. If buyers wire their own agents straight into exchanges via an MCP server (a standard way for AI agents to plug into external tools), the fee that sits with The Trade Desk and Google's DV360 is the thing under threat. In plain terms: the toll booth in the middle of the ad-buying road might get bypassed. But note the counterprogramming at the same event — Google's Brian Jankovsky pitched more infrastructure, not less. The incumbents aren't conceding the interface. Watch whether any real buy-side budget flows through agent integrations in 2026, or whether this stays a keynote.

The Skeptic

The load-bearing assumption is "search is going away," and the episode's own fact-check dismantles it — Google search revenue still grew last quarter and remains the largest ad format on earth. Everything downstream (Glance as the new top-of-funnel, funnel collapse, DSP death) rests on that overstatement. The 30% lift on conversational CTV comes from the CEO selling the product, with no baseline, no methodology, no third party. The "50 million Samsung TVs" is a fleet size, not a rollout. And "one of three Apple Search Ads partners" overstates exclusivity. This is vision-selling at Cannes. Strip the hyperbole and you have three early, unproven products.

The Operator

Tuesday morning, the interesting thing here isn't the vision — it's the plumbing. An SSP exposing deal creation, audience segmentation, and RTB filtering through an agent interface is a real integration project, and the meeting notes show exactly what it looks like in practice: securing test accounts, webhook endpoints, red-teaming an agent that dispatches long-running tasks. That's the unglamorous reality behind "the era of pressing 10 buttons is over." The other operator signal buried in those notes: enabling multi-format auctions caused Magnite, TripleLift, and Sharethrough to drop display revenue hard. Translation — supply-side experiments break demand-partner integrations first. Agentic buying will hit the same wall: your agent is only as good as the partners who honor its calls.

The Customer / End User

Two customers here, and they want different things. The advertiser doesn't want to write code to buy media — they want fewer buttons, yes, but they also want auditability, brand safety, and someone to blame when spend goes sideways. An agent making bidding decisions in a black box is a trust problem, not a convenience win. The consumer on a Samsung TV is the bigger question: are people actually going to converse with an ad, or shop from their lock screen? Glance is monetized on conversion, not impressions — a healthy sign that InMobi is being honest that this only pays if people actually buy. That's the whole ballgame, and it's unproven.

The CFO

The MobileAction deal is the tell. No terms disclosed, ~30 people, buying into Apple Search Ads — a fast-growing but walled channel. That's a cheap, sensible bolt-on: own a managed channel, bundle it, capture app-install budget. Real economics, near-term payback. Contrast that with the Agentic Selling Platform and Glance, which are R&D bets funded by story. The affiliate model on Glance means revenue only shows up on conversion — great for InMobi's risk, brutal on near-term revenue predictability. For any operator tempted to copy this: the ASA-style acquisition pays back; building an agent interface ahead of buyer demand is a cost center until buyers show up.

The tensions

1. Is the DSP dying or just changing clothes? The Market Analyst sees a genuine margin threat if agents bypass the middle layer. Google, same week, argued the opposite — the future needs more infrastructure, not less. Both can't be right about who owns the interface.

2. Funnel compression: structural or self-serving? Best Buy's Patrick Albano makes the same funnel-collapse argument from the retail-media side, which lends it credibility. But the Skeptic notes it conveniently sells Glance. Corroboration from a second Cannes stage isn't independence when everyone's pitching the same trend.

3. Vision vs. plumbing. The Operator's read is that the boring integration work — webhooks, partner compatibility, red-teaming — decides whether any of this ships, and demand partners breaking on format changes is the leading indicator of how agent adoption will actually go.

Synthesis

This hinges on two beliefs: (1) buyers will genuinely route budget through agent-to-exchange integrations rather than DSP UIs, and (2) consumers will transact on new conversational/visual surfaces. Belief 1 has real architectural momentum and the internal meeting notes show the plumbing is buildable. Belief 2 is pure hope right now. The council leans toward: the agent plumbing is real and worth staffing to understand; the funnel-collapse and consumer-behavior claims are marketing until proven. Before committing anything, operators should demand a methodology on that 30% lift and watch whether any named buyer publicly routes real spend through an agent integration — not a pilot, real budget.

Prediction: By InMobi's next major update at CES in January 2027, no independent buy-side spender will have publicly disclosed material ad budget routed through InMobi's Agentic Selling Platform (agent-to-exchange, bypassing a DSP) — it will remain a demo-and-keynote story, not a booked-revenue channel.

Confidence: Medium — Cannes vision-selling with unverified metrics rarely converts to disclosed spend within two quarters.

Why: The episode is heavy on vision and light on data — the fact-check flags every headline claim as unverified or overstated, and the "30% lift" comes from the vendor with no methodology. Agent-to-exchange buying also fails the Operator's compatibility test: the same meeting notes show demand partners breaking when an SSP changes auction mechanics, and buyers won't route real money through a black-box agent without auditability and brand-safety guarantees. The opposite outcome — a named advertiser disclosing real agentic spend by January — would require both the technical plumbing and buyer trust to mature in under six months, which the evidence doesn't support; the safer path (the MobileAction ASA bolt-on) is where near-term revenue actually sits.

Revisit by 2027-01-31: We're right if no independent buyer has publicly disclosed material spend through the Agentic Selling Platform by CES 2027. We're wrong if a named advertiser or agency confirms real (non-pilot) budget flowing through it.