The AI agents race is already here - and nobody asked you first

by Luís Rijo 10 min read
Advertising Platforms Deals
The AI agents race is already here - and nobody asked you first
Robot sits alone at boardroom head, empty chairs, tech giant logos on walls.

Every few years, the technology industry decides something is both inevitable and urgent at the same time. Usually there is money behind that feeling. In 2026, the thing is AI agents - software programs that do not just answer questions but take actions, make decisions, and complete multi-step tasks without a human approving each step. They book meetings, build ad campaigns, negotiate deals, and crawl the entire internet on your behalf whether you know about it or not.

By the numbers, the shift arrived fast. McKinsey tracked $1.1 billion in equity investment flowing into agentic AI in 2024 alone. Job postings related to the technology increased 985 percent from 2023 to 2024. Two-thirds of advertisers told the IAB - the trade association for digital advertising - that they were already focusing on agentic AI for campaign execution in their 2026 plans. And every major technology platform from Amazon to Salesforce to Google has either launched a live agent product, published a technical framework for building one, or quietly rewritten its legal contracts to govern how agents interact with its systems.

The race is not theoretical anymore. The question is who controls the infrastructure the agents run on - and what happens to everyone who built their business assuming a human was still in the loop.

The Background

To understand what is happening in 2026, it helps to understand what programmatic advertising is - the automated system that buys and sells digital ads in milliseconds through a series of auctions that happen faster than a blink. When you visit a website and an ad appears, that ad was likely purchased in real time through a chain of intermediaries: a DSP (demand-side platform, the software that helps advertisers buy ads), an SSP (supply-side platform, the software that helps publishers sell them), and a shared auction standard called OpenRTB that lets the whole system communicate.

That infrastructure was built to serve humans making decisions. A media buyer at an agency would set parameters, and the software would execute within them. Humans were always, nominally, in control.

Agentic AI changes that premise. An AI agent does not wait for instructions on each step. It receives a goal - "manage this campaign and improve return on ad spend" - and then continuously monitors, adjusts, creates, and optimizes on its own. Return on ad spend, or ROAS, is simply how many dollars an advertiser earns for every dollar spent on ads. An agent trying to improve it can theoretically do in seconds what a human analyst would take days to accomplish.

The appeal is obvious. The anxiety is equally obvious. If software agents are doing the buying, the selling, and the decision-making, what exactly are the agencies, platforms, and middlemen doing? Ari Paparo, founder of Marketecture Media, argued in July 2025 that autonomous AI systems could automate the campaign setup, targeting, and optimization functions that DSPs currently charge for - potentially eliminating the centralized role those platforms occupy entirely. That analysis, made less than a year ago, no longer reads as speculative.

The money moved first. The standards are now scrambling to catch up.

What Actually Happened

The inflection point came in a single week in November 2025. Amazon, Google, and IAB Tech Lab each made major agentic AI moves within days of one another - not by coincidence, but because the competitive pressure had become impossible to ignore.

Amazon launched its Ads Agent on November 11, 2025, at its annual unBoxed conference. The system processes natural language instructions to execute complex advertising workflows: it reads uploaded media plans, builds campaign structures automatically, reviews thousands of audience segments to recommend targeting options, and generates analytics queries without requiring anyone to know SQL - the database language typically needed to extract data from marketing platforms. The following day, Google made its Ads Advisor and Analytics Advisor available across all English-language accounts.

IAB Tech Lab, the nonprofit standards body for digital advertising, released its Agentic RTB Framework on November 13 - a specification defining how containerized AI agents could participate in real-time bidding infrastructure. OpenRTB is the existing standard language that all programmatic auctions speak; the new framework was an attempt to extend that language before competing, incompatible protocols could splinter the industry.

By January 2026, agentic AI infrastructure had dominated the industry's annual CES week. Yahoo's DSP integrated agents that could autonomously monitor campaigns and execute corrective actions. PubMatic launched AgenticOS on January 5, with live campaigns running on connected television inventory - meaning autonomous systems were actually buying real ads on real TV screens without a human approving each transaction. Magnite announced a seller agent integration in its SpringServe ad server the following day.

On February 26, IAB Tech Lab formalized its umbrella initiative under the name AAMP - Agentic Advertising Management Protocols - to address what its CEO Anthony Katsur described as "confusion in the market." The organization had released so many specifications in parallel - the Agentic Real-Time Framework, the Agentic Audiences Framework, Buyer and Seller Agent SDKs - that practitioners could not tell what was part of a coordinated effort and what was competing with itself. AAMP was the answer: one name, three pillars (execution, protocols, and a registry), and a formal commitment to maintain the organizing logic as the buildout continued.

The registry that AAMP introduced is notable on its own terms. By March 11, it had expanded to 10 active entries, including Amazon, Burt Intelligence, Optable, Dstillery, and HyperMindZ.ai, all operating under the Model Context Protocol - a standard, originally developed by Anthropic, that defines how AI systems connect to external tools and data sources. The registry's stated purpose is bringing trust and transparency to agentic transactions: when a buyer agent is negotiating with a seller agent autonomously, someone needs to know who each party actually is.

Meanwhile, Salesforce on April 15 unveiled Headless 360 - its most ambitious architectural change in 27 years - which exposes every capability in its platform as an API, MCP tool, or command-line interface, removing the browser as the mandatory way to interact with the system. More than 100 developer tools shipped the same day. Co-founder Parker Harris had framed the logic the previous month in a direct question: "Why should you ever log into Salesforce again?" In April 2026, the answer is: maybe you do not need to.

This same week, Amazon, Meta, Microsoft, Salesforce, and Stripe all joined the UCP Tech Council - the governance body for the Universal Commerce Protocol, an open standard covering the full shopping journey from product discovery to post-purchase interaction. Amazon's membership is particularly striking. The company had spent most of 2025 blocking AI bots from competitors and building its own proprietary agentic commerce tools. The protocol conversion is a meaningful signal.

The Money Trail

The companies winning in the agentic AI race are not necessarily the ones building the best agents. They are the ones building the infrastructure the agents must run on.

Amazon's position is instructive. The company blocked AI bots from Anthropic, Meta, OpenAI, Google, and Huawei in August 2025 while simultaneously launching its own agentic advertising tools and then quietly updating its seller contracts in February 2026 to establish formal legal rules for any automated system that touches its marketplace. The sequence is deliberate: build your own agents, shut out competitors' agents, then write the rulebook for what any future agents must do on your platform. Amazon's MCP Server - launched in closed beta November 13, 2025, and open beta on February 2, 2026 at the IAB Annual Leadership Meeting - transforms its advertising API into something that AI agents can operate through natural language. The company is building a door it controls, then inviting the ecosystem in through it.

Google's position follows similar logic. Advertising Advisor agents running across every English-language account means the company's own autonomous systems are influencing spending decisions on Google's own inventory. That is not a conflict of interest anyone is loudly flagging. Google is also the entity whose infrastructure, according to research reported by PPC Land, ChatGPT's paid version now runs search on - despite official documentation that claimed otherwise. The infrastructure layer is where the leverage lives, and Google still controls more of it than anyone else.

The platforms that stand to lose are the traditional intermediaries. Microsoft shut down Microsoft Invest - formerly Xandr, the DSP it acquired for $1 billion in 2021 - effective February 28, 2026. The stated reason was that the traditional demand-side platform model was incompatible with the company's vision for "conversational, personalized, and agentic" advertising. Translation: a software category that charges fees for doing things agents now do for less is a liability, not an asset. The companies running those platforms should be paying close attention.

The commerce side of the story carries equal financial weight. OpenAI launched Instant Checkout for ChatGPT with Stripe in September 2025. Microsoft introduced Copilot Checkout with Stripe, PayPal, and Shopify in January 2026. OpenAI's systems, per a March 2026 joint report from Retail Economics, AWS, Botify, and DataDome, generate 198 crawls for every single visit they deliver to a retail website. For retailers, that ratio means AI systems are consuming far more than they are sending. Structured product data and protocol compliance are becoming prerequisites for presence in AI-mediated discovery - not optional improvements for ambitious brands, but table stakes for existing ones.

The regulatory response is arriving too, though at a different pace. UK watchdogs - the CMA, FCA, ICO, and Ofcom - published a joint assessment on March 31, 2026, concluding that AI agents do not fall outside existing UK law and that existing obligations around transparency, fairness, safety, and competition continue to apply. The more technically alarming section of that document concerned algorithmic collusion: experiments had already shown LLM-based agents colluding in price-setting and bidding scenarios without being explicitly told to. In those experiments, agents "repeatedly converged to supra-competitive prices and maintained them, even when the environment was noisy." Nobody programmed that behavior. It emerged.

What It Means

The OECD published a 34-page working paper in February 2026 attempting to define precisely what agentic AI is - a sign that the gap between deployment speed and conceptual clarity has become impossible to ignore at a policy level. The most popular tools in actual use remain general-purpose large language models: ChatGPT, GitHub Copilot, Google Gemini, Claude Code, and Microsoft Copilot. Full autonomy remains rare - only 4 percent of enterprise teams allow agents to act without any human approval at all.

But "rare" and "inconsequential" are not the same thing. PubMatic already has live campaigns running. Amazon's agents are already active on its platform. The agent registry has real companies with real deployments. The 4 percent who let agents act fully autonomously are making real purchases with real budgets.

For anyone who watches television, the connection is direct: connected TV is where some of the first fully autonomous ad buys happened. For anyone running an e-commerce business, the consequence is concrete - if structured product data and protocol compliance determine whether AI agents can find you, then the technical decisions a developer makes are now marketing decisions. For anyone working in media buying or campaign management, the Microsoft Invest shutdown is the most honest signal available: a technology company looked at what AI agents could automate and decided the human-operated platform could not survive the comparison.

Gartner predicted in June 2025 that over 40 percent of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, and inadequate risk controls. That prediction may well be right. Many of the protocols proliferating in 2025 and early 2026 will not survive. But canceling projects is not the same as reversing direction. The infrastructure being built - the registries, the protocol layers, the agent SDKs, the legal frameworks in platform contracts - is designed to outlast any individual deployment. The foundations are going in whether the early projects succeed or not.

The Bottom Line

The AI agents race in 2026 is not a contest between chatbots. It is a contest over who controls the infrastructure that autonomous software must pass through to do anything - buy an ad, place an order, manage a campaign, or crawl a website. Amazon, Google, Salesforce, and Microsoft are each building that infrastructure with different bets and different moats. The standards bodies are trying to prevent fragmentation before competing protocols make the whole ecosystem unworkable. Regulators are issuing warnings while acknowledging they are mostly watching. The businesses most at risk are the ones whose entire value proposition is doing the thing the agents are being trained to do instead - and that category is larger than anyone in those businesses wants to admit.

Timeline

Summary

Who: Amazon, Google, Microsoft, Salesforce, OpenAI, IAB Tech Lab, PubMatic, Swivel, Stripe, Meta, and a growing list of ad-tech companies deploying or governing agentic AI systems - alongside UK, EU, and OECD regulators beginning to respond.

What: A race to deploy and govern agentic AI - autonomous software that takes actions and makes decisions without human approval at each step - across advertising, commerce, and enterprise software, with $1.1 billion in 2024 equity investment and 985 percent growth in related job postings.

When: The infrastructure buildout accelerated sharply in November 2025, with a coordinated wave of launches from Amazon, Google, and IAB Tech Lab. The race has intensified through Q1 and Q2 2026, with live campaigns running and major platform contracts being rewritten to govern agent behavior.

Where: Global digital advertising infrastructure, e-commerce, enterprise software, and connected television - the systems governing where money flows in the media and platform economy.

Why: Every major platform sees autonomous agents as a way to lock in infrastructure advantage before competitors establish alternative standards. The businesses most threatened are those whose value proposition is performing tasks agents are designed to automate.

Luís Rijo

Luís Rijo

Luís Rijo is a seasoned marketing professional with over 10 years of experience in Digital Marketing, Search, Social, Display, Video, and DOOH. Based in Europe. Also writing in PPC Land. Reach out via [email protected]