Agentic Media Buying Is Real Now — Which Indian Startups Are Actually Letting AI Run Campaigns

Ask a media planner what changed this year and most won’t reach for the word “agentic” at all. They’ll describe something smaller and stranger: a campaign that reallocated its own budget at two in the morning, moved spend off an underperforming placement, and had already recovered its CPA by the time anyone logged in to check. Nobody approved that decision. Nobody was awake to. That’s the actual texture of the shift happening inside Indian AdTech right now — less a keynote-stage revolution than a quiet redrawing of who, or what, gets to decide.

The global trigger point was almost comically concentrated. In the nine days before Cannes Lions this year, eight of the industry’s biggest names — DoubleVerify, LiveRamp, Pixalate, Mediaocean, Magnite, Yahoo, Stagwell and Fox — each shipped some version of autonomous buying or coordination infrastructure, as if every roadmap in the business had quietly synced to the same deadline. That kind of clustering rarely happens by accident. It happens when a category tips from “worth exploring” to “too risky to be seen without,” and everyone scrambles to not be the one caught explaining why they weren’t ready.

India’s version of that scramble looks different, mostly because the word “agentic” is doing more work here than it can honestly bear. It gets attached to anything with a recommendation engine and a friendly interface. The useful test is narrower: does the software merely suggest an action for a human to click “approve” on, or does it read the situation, form a plan, act on it, watch what happens, and revise — on its own, inside guardrails a person set once and mostly forgot about? Most of what gets called agentic in Indian marketing decks today is the former. A handful of platforms are genuinely building toward the latter, and they’re not all trying to solve the same slice of the problem.

Take the budget layer first, because it’s the one marketers actually lose sleep over. Navi Mumbai’s Adsnex has built what it calls an advertising operating system — less a tool bolted onto existing workflows than an attempt to replace the workflow itself, so pacing, bid adjustments and cross-network budget shifts happen inside one system instead of across five disconnected dashboards a human used to reconcile by hand every morning. It’s the unglamorous, high-stakes end of the stack: nobody writes a case study about pacing discipline, but it’s the thing that quietly determines whether a campaign’s money survives the month.

Creative is a different fight entirely, and it’s being fought at two very different scales. Gurugram’s Marx AI Technology keeps a human hand on strategy deliberately — its workflow AI handles competitor analysis and creative ideation fast enough that a brand team can test far more variants than they’d ever produce manually, but the calls about what the campaign is actually trying to say, and to whom, stay with people. It’s a useful corrective to the all-or-nothing framing that autonomy debates tend to fall into: agentic in execution, assistive in judgment, and arguably closer to what most Indian marketing teams will actually be comfortable running in the next year than a fully hands-off system. Bengaluru’s Lumiad goes further down the automation curve on a narrower problem — the flood of short, UGC-style video that e-commerce feeds now demand. Scripts, AI-avatar production and performance-informed iteration happen inside one loop, collapsing what used to be a creator, a shoot day and a week in the edit suite into something closer to same-day turnaround. Neither company is trying to run the whole campaign. Both are betting that owning one link in the chain, done well enough, is worth more than owning all of it badly.

Then there’s the harder-won kind of advantage: incumbency. DeltaX, which has been optimising cross-channel bidding since 2012, isn’t chasing this trend so much as it’s sitting on the thing every agentic system actually needs and can’t fake — years of proprietary campaign data to train against. A model is only as good as what it’s learned from, and a platform serving well over a thousand advertisers across search, social, display and video has learned from a lot more real budget decisions than any startup demoing a slick interface this quarter. Mumbai’s HockeyCurve is making a related bet from a different angle, treating creative and placement as one continuous optimisation problem rather than two departments handing work back and forth — dynamic creative and media delivery adjusting to each other in real time, instead of one waiting on the other’s report.

What’s easy to miss, looking at five companies solving five different pieces of the puzzle, is that none of them has actually built the thing the Cannes headlines implied already existed: a single system that autonomously plans, buys and optimises a campaign end to end, across every network, with no human in the loop at all. That system doesn’t exist yet, in India or anywhere else. What exists is a stack being assembled one layer at a time, by companies that have each decided a different layer is worth owning first.

The capital sloshing around the category suggests everyone expects that stack to matter. Industry estimates put more than $100 billion in global programmatic spend on a path toward flowing through AI agents by 2028, and India’s broader AI market is tracking toward $17 billion by 2027, buoyed by government backing through the IndiaAI Mission and a sharp rise in venture inflows. None of that money is earmarked “agentic media buying” on a line item — it’s flowing into the GPU access, infrastructure and model-building that sits underneath products like these. But it means the current generation of Indian AdTech founders isn’t building against the capital scarcity that shaped the last one, which is its own kind of structural advantage.

The trust gap is the more honest story, though, and it’s worth resisting the urge to skip past it. The World Federation of Advertisers found that the overwhelming majority of media buyers are highly or extremely concerned about data privacy when using generative AI for media — concerned enough that an entire sub-industry now exists purely to build the governance layer autonomous buying needs before anyone will hand it real money. An agent that can technically execute a buy across a dozen networks is not the same thing as an agent a CFO is comfortable leaving unsupervised with the quarter’s budget, and Indian marketing teams, often working with tighter budgets and closer scrutiny than their global counterparts, are unlikely to close that gap faster than anyone else.

There’s a second, quieter cost to all of this that the industry hasn’t fully sat with yet: the media planner whose job was, until very recently, exactly the sequence of decisions these systems now make on their own. The honest read is that the role doesn’t disappear so much as it moves — from executing inside the loop to designing the guardrails the loop runs inside, from making individual bid calls to defining what “good” should even mean for a system to optimise toward. That’s a real change in what the job demands, whether or not the job title survives it, and agencies treating this purely as a headcount conversation are likely to be the ones caught flat-footed by it.

It’s also worth being honest about which layer is likely to win that fight, because the incentives point one direction more clearly than the industry likes to admit. Budget orchestration and incumbent data advantages compound — the more decisions a system makes correctly, the more data it has to make the next one better, and the harder that lead becomes to close. Creative automation, by contrast, is closer to a commodity race: a faster script-to-video pipeline is a real edge today and a baseline expectation in eighteen months, the way stock-photo-quality banner generation went from differentiator to given inside a couple of ad cycles a decade ago. That doesn’t make Marx AI’s or Lumiad’s bets wrong. It makes them time-limited in a way Adsnex’s and DeltaX’s aren’t, which is precisely why the founders building on the execution and data layer talk less about speed and more about defensibility.

There’s a version of this story that agencies would rather not hear, which is that the trust gap isn’t purely a technology problem waiting on better guardrails — it’s also an incentive problem. A platform that autonomously reallocates budget performs better, on average, when it’s allowed to move fast and occasionally wrong, the same way any optimisation system does. But “occasionally wrong” is a hard sell to a client who signs off on the media plan and answers for it in the quarterly review. Until the industry works out who actually absorbs that risk — the platform, the agency, or the brand — a meaningful chunk of the caution around full autonomy isn’t going to be solved by better models. It’s a contractual and cultural problem wearing a technical costume.

None of that makes the shift any less real. It just means the interesting question for the next eighteen months isn’t whether Indian AdTech is building toward agentic media buying — it clearly is, on five fronts at once — but which of these bets on budget, creative, incumbency or risk tolerance ends up being the layer everyone else has to build on top of. For a category that’s spent most of its history importing playbooks from elsewhere, that’s a genuinely different position to be arguing from.