Programmatic has always been automated. Machines have been buying and selling impressions in milliseconds for more than fifteen years. So when people say AI is transforming programmatic, it is fair to ask: what is actually new?

The honest answer is that three things are changing at the same time. AI is moving from optimizing campaigns to running them. It is moving from the buy side only to every step of the supply chain. And it is being used by fraudsters as much as by the people fighting them.

The buyers are ready

The IAB's 2026 Outlook study gives a clear signal. US ad spend is expected to grow 9.5% in 2026, and AI is at the top of buyers' agendas:

  • 78% of buyers plan to increase their focus on generative AI in their media campaigns.
  • 66% are increasing their focus on agentic AI for ad buying and campaign execution.
  • AI accounts for five of the six top technology, data and measurement priorities in the study.

In other words, AI is no longer an experiment. It is becoming part of the default way media is planned and bought.

From optimization to agents

For years, AI in programmatic meant optimization: bid shading, budget pacing, lookalike audiences, creative testing. Useful, but always inside a system designed by humans.

Agents are different. An AI agent can receive a goal ("reach this audience on CTV in these markets with this budget"), discover inventory, negotiate terms, launch a campaign and adjust it, with a human supervising rather than operating.

To make this possible, the industry needs common rules. Two initiatives stand out:

  • The Ad Context Protocol (AdCP), launched in October 2025 by a group including PubMatic, Scope3 and Swivel, is an open standard that lets buy-side and sell-side agents communicate and transact. It includes a media buying protocol, a creative protocol and a signals activation protocol, and it is designed to work with existing DSPs, SSPs and ad servers.
  • The IAB Tech Lab's Agentic RTB Framework (ARTF), released for public comment in November 2025, lets companies run their models as secure containers inside another platform's infrastructure. Instead of calling each other across the internet, a DSP's model can run directly inside an SSP, close to the data, with lower latency. Index Exchange, OpenX, The Trade Desk and Chalice developed the specification, and companies such as Amazon Ads, Netflix, Magnite, PubMatic and HUMAN took part.

Both point in the same direction: more intelligence, closer to each decision.

AI on the sell side

Most of the attention goes to buying agents, but some of the most practical gains are on the supply side.

An SSP sees every request before any buyer does. That is the ideal place to use machine learning to answer a few simple questions:

  • Is this request real? Fraud detection models can score a request before it is sent anywhere.
  • Who is likely to want it? A prediction model can estimate, for each buyer, the probability that it will bid on a specific request, based on the app, the device, the format, the time of day and past behavior.
  • What is it worth? Dynamic floors can adapt to demand instead of using fixed prices.
  • When will demand peak? Pacing forecasts help send the right volume at the right moment instead of flooding buyers.

This is not a theoretical exercise. Sending every request to every buyer is the default in programmatic, and it creates enormous waste: buyers process requests they will never bid on, and good requests get lost in the noise. Predicting before routing reduces that waste on both sides.

AI on the other side of the fight

AI is also making fraud more dangerous. DoubleVerify reported that CTV fraud schemes grew 140% in the first quarter of 2026 compared with a year earlier, and that fraudsters use AI to scale operations and build harder-to-detect schemes. Its Fraud Lab described a network of around 200 AI-generated websites built to capture programmatic spend.

The consequence is simple: verification also has to be AI-driven, and it has to happen before the bid. Rules written by hand cannot keep up with schemes generated by machines.

What it means for publishers and buyers

For publishers, AI raises the bar for supply quality. Agents and prediction models will favor inventory with clean signals: authorized sellers in ads.txt and app-ads.txt, transparent sellers.json entries, complete supply chains and independent verification. Inventory that cannot prove what it is will be routed less and paid less.

For buyers, AI promises less manual work and less waste, but it also creates a new risk: delegating decisions to systems you cannot see. Transparency standards matter more than ever, because an agent can only make good choices if the information it receives is accurate.

How we approach it at PeopleAds

We built PeopleAds around a simple idea: smart routing, powered by prediction. Every request goes through five steps:

  1. Verify: ads.txt, app-ads.txt, sellers.json and supply chain checks, plus independent scanning by Pixalate, DoubleVerify and HUMAN.
  2. Predict: a score of how likely the request is to convert for each buyer.
  3. Route: the request goes only to the buyers most likely to bid.
  4. Price: dynamic floors that follow demand.
  5. Pace: forecasts that match volume to each buyer's demand over the day.

The result is what we call sending less so buyers spend more. Fewer wasted requests, more relevant ones, and a supply path that both sides can trust.

AI will not replace programmatic. It will make it more selective. The winners will be the platforms that decide, before anything else, which request deserves to be sent and to whom.

Sources