The next internet will not be built by humans clicking buttons; it will be built by agents acting on behalf of people, brands, publishers, platforms and enterprises.
Some of these agents will be simple workflow assistants. Others will manage budgets, audiences, inventory and outcomes with increasing autonomy across systems that were never designed to work together.
This raises a timely question: If large language models can understand context, infer meaning and reason across messy data, do we still need standards and protocols to guide AI solutions?
Many of today’s standards exist because software historically struggled with ambiguity. We built schemas because machines could not understand intent. We built taxonomies because systems could not reconcile different definitions. We built APIs because applications could not communicate without rigid interfaces.
But today, an LLM can increasingly bridge those gaps on its own. It can recognize that “campaign start date,” “flight begin” and “launch timestamp” describe the same concept. It can translate schemas, map taxonomies, generate integration logic and infer meaning from incomplete information.
This has led some to argue that standards become less important as AI becomes more capable. This argument is directionally right. But it misunderstands why standards exist in the first place.
Standards are about trust
The real purpose of standards was never machine comprehension; it was coordination between partners.
LLMs are remarkably effective at interpretation and can act as semantic middleware between systems that were never designed to communicate.
But while AI can infer meaning, it cannot create trust. That distinction becomes increasingly important as agents move from helping humans make decisions to making decisions themselves.
A model may infer what a field means, but it cannot independently prove whether data was authorized, whether a signal originated from a legitimate source, whether an action remained within delegated authority or whether accountability exists for the outcome. Those are not language problems; they are trust problems.
Most of today’s trust infrastructure was designed to answer relatively narrow questions. Identity systems verify access. Consent systems verify permission. Fraud systems attempt to identify invalid activity. Measurement systems confirm that an event occurred.
The agentic internet introduces a much harder challenge: determining whether an action remains connected to the authority that authorized it in the first place.
That distinction matters because AI can already generate behavior that appears entirely legitimate. The audience looks real. Customer journeys look plausible. Optimizations look rational. Decisions look justified. Yet none of those characteristics prove that an action remains connected to an accountable principal, a valid permission boundary or a delegated authority structure.
As agents move from assisting decisions to making them, that gap becomes increasingly important.
The future challenge is determining whether actions remain connected to legitimate authority over time. Who delegated the action? What authority was granted? Did the action remain within scope? Can responsibility be assigned when something goes wrong?
These questions sit at the center of the next generation of standards.
The evolution of trust
The most important standards of the next decade may not be focused on data formats at all.
AI will likely absorb much of the complexity associated with schema mapping, taxonomy translation and integration logic. What remains are the questions inference alone cannot solve: provenance, permission, delegation, authority, continuity, auditability and economic accountability.
Historically, standards helped machines exchange information. Later, they helped markets exchange value. In the agentic era, they will increasingly help autonomous systems exchange trust.
As agents begin negotiating deals, configuring campaigns, selecting audiences, allocating budgets, and making optimization decisions, trust can no longer be evaluated solely through post-campaign reporting and data governance reviews. It increasingly needs to be evaluated at the exact moment an action occurs.
Trust moves from an audit function to an execution function. From governance review to runtime infrastructure. From post-campaign analysis to real-time control. In other words, the standards that promote trust must move up in the programmatic tech stack.
Media markets already trade on signals that claim to represent attention, intent, audience quality and commercial value. AI will make it easier to generate signals that appear valid and trusted while remaining disconnected from accountable origins. Hence why standards that promote trust need to move into the decisioning layer.
A new era for standardization
The common assumption is that better AI reduces the need for standards. The opposite may prove true. As models become better at creating coherence from incomplete information, markets become more dependent on mechanisms that establish provenance, authority, accountability and trust. The stronger the inference layer becomes, the more important the trust layer becomes.
The future is not a choice between protocols and reasoning. Protocols without AI become bureaucracy. AI without protocols becomes persuasive chaos. The winners will combine both: systems where models handle interpretation and adaptation while standards govern authority, accountability and trust.
The first era of the internet standardized information. The second standardized transaction. The third will standardize trust.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow Evgeny Popov and AdExchanger on LinkedIn.
For more articles featuring Evgeny Popov, click here.
