AI Agents Are Rewriting the Rules of Media Power

The upfront advertising marketplace isn’t dying--it’s mutating. What appears to be a slow automation of media buying is actually a silent power shift, where AI agents aren’t just tools but emerging negotiators with real agency. The non-obvious consequence? The middlemen--agencies and sales teams--are being forced to prove their value in a world where bots can forecast inventory, optimize spend, and even haggle without fatigue. This isn’t about efficiency; it’s about control. Executives in media buying, ad tech, and brand strategy need to see this not as a technical upgrade but as a systemic reordering. Those who wait for AI to mature will find their roles reduced to oversight, while those who map the feedback loops now--between automation, flexibility, and power--will shape the next phase of media. The real advantage lies in understanding not just how AI changes execution, but how it rewrites the rules of leverage in an $80 billion ecosystem.

Why the Obvious Fix--More Automation--Creates a Power Vacuum

The upfront marketplace has always been a dance of commitment and control. For decades, TV networks and streamers relied on the annual upfront cycle to lock in guaranteed revenue from advertisers, trading discounts and premium placements for financial certainty. The system worked because it was human-driven: relationships mattered, haggling was an art, and access was earned through trust and scale. But that control is eroding--not because of AI taking over negotiations, but because automation is dissolving the very need for long-term exclusivity.

Buyers no longer have to choose between upfront certainty and real-time flexibility. Programmatic guaranteed (PG) deals once promised the best of both worlds: automated execution with guaranteed spend. But as Michael Bergie notes, the pendulum has swung. Private marketplaces (PMPs) are now overtaking PGs, not because they offer better inventory, but because they offer escape hatches. Brands want the option to pivot--quarter to quarter, campaign to campaign--without being locked into rigid commitments. This isn’t just about agility; it’s about risk mitigation in uncertain economic times.

"Brands and their agencies want to be sure that although they're committing long-term dollars that there are exit ramps along the way."

-- Michael Bergie

That demand for flexibility has forced sellers to adapt. Networks like Warner Bros. Discovery and NBCUniversal are no longer just offering inventory--they’re offering integration. They’re building AI agents that can reason, negotiate, and adjust in real time. Ryan McConville from NBCUniversal described this as “premium automation,” where AI doesn’t just execute trades but handles the nuanced parts of deals that traditional programmatic pipes can’t touch--custom targeting, dynamic pricing, and even creative alignment. The system responds: as buyers demand flexibility, sellers respond with smarter, more adaptive automation.

But here’s the hidden consequence: every layer of automation reduces the friction that once protected human gatekeepers. The more seamless the process, the less value there is in the middleman who used to manage complexity. Agencies that once charged premium fees for negotiation and trafficking now face a world where AI can do both--faster and cheaper.

The 18-Month Payoff Nobody Wants to Wait For: Training AI with Real Deals

Most companies experiment with AI in low-stakes environments. Not here. The upfront cycle has become a live training ground. Sellers aren’t waiting for perfect models--they’re using real negotiations to teach their agents how buyers behave, what concessions work, and when to hold firm. John Kozak at Televisa Univision put it plainly: give us six months to a year, and we’ll use this cycle to understand buyer agents, refine our own, and see what deals actually emerge.

This is systems thinking in action. Each deal isn’t just a transaction--it’s a data point that trains the next negotiation. The feedback loop is already active: NBCUniversal tested AI agents during live NFL playoff games. Warner Bros. Discovery uses them for supply forecasting. Paramount is in closed beta with seller agents. These aren’t proofs of concept; they’re incremental investments in a future where the agent, not the salesperson, sets the terms.

And the buyers? They’re learning the hard way. A small agency let an AI agent manage a client’s investment--and it blew through ten times the intended budget in 24 hours, hyper-focusing on a single high-performing channel with no sense of balance or risk. The agent was smart. It just had no context.

"It was smart but had no context of limitation or well, don't put all your eggs in this basket."

-- Kimika McCoy

That failure is revealing. It’s not that AI can’t invest--it’s that unconstrained AI optimizes for immediate wins, not long-term strategy. The real competitive advantage isn’t in deploying AI first, but in building the guardrails that align its goals with business outcomes. The teams that win won’t be those with the most advanced models, but those who’ve spent the time--the painful, invisible time--defining constraints, ethics, and fallbacks.

This is where conventional wisdom fails. Most assume AI adoption is about speed: who can automate fastest wins. But the opposite may be true. The deeper the integration, the more dangerous unchecked optimization becomes. The six-month delay in full deployment? That’s not a lag--it’s a strategic window. The agencies and brands using this time to train, test, and constrain their agents are building moats that others won’t replicate because they’re too busy chasing short-term efficiency.

How the System Routes Around Human Gatekeepers

Amazon and YouTube aren’t playing by the old rules. They’re treating upfronts like enterprise software deals--volume commitments tied to platform lock-in. Want premium YouTube inventory? You’ll need DV360. Want Amazon’s Fire TV reach? Use their DSP. This isn’t just distribution control; it’s ecosystem dominance. And because these platforms already operate at scale with automated buying, they don’t need the upfront theater. Their “upfront” is baked into the platform terms.

The system responds: TV networks and streamers, desperate to compete, are adopting the same playbook. But they’re doing it with a twist--agentic AI. NBCUniversal’s “premium automation” isn’t just about efficiency; it’s about creating a proprietary layer of intelligence that can’t be replicated by third-party DSPs. If Amazon owns the pipe, networks want to own the agent inside the pipe.

And here’s the kicker: this could accelerate in-housing, but not in the way most expect. Small brands may never get log-level data or direct access--not because the tech isn’t there, but because scale still buys privilege. Publicis gets log-level data because they spend enough to make it worth the seller’s effort. The same will be true for AI access. The holding companies will negotiate not just for inventory, but for agent interoperability. They’ll demand that seller agents can speak to buyer agents, creating a new layer of technical and strategic complexity.

This shifts the game from media buying to system design. The winners won’t be the best negotiators, but the best architects--those who can design AI workflows that align with business goals, anticipate competitor moves, and build feedback loops that compound over time.

Key Action Items

• Over the next quarter: Audit your current upfront commitments and identify where flexibility is being traded for discounts. Map which portions could shift to PMPs without financial penalty.
• Within 6 months: Begin testing AI agents in non-critical planning workflows--audience modeling, inventory forecasting--but with strict budget and targeting guardrails. Treat failures as training data.
• This pays off in 12--18 months: Invest in agent interoperability. If you’re a buyer, demand APIs and protocols that allow your AI to negotiate directly with seller agents. If you’re a seller, build your agent to be the easiest to integrate with.
• Start now: Shift agency conversations from “Can you get us better rates?” to “How are you training your AI to represent our long-term strategy?” The value of agencies will increasingly hinge on their ability to manage this layer.
• Flag for discomfort: Allow AI to make small, reversible decisions autonomously--even if it feels risky. The teams that normalize AI-driven choices now will have a 6--12 month lead in confidence and capability.
• Over the next year: Prepare for the “forever upfront” model, where annual commitments are replaced by rolling, AI-negotiated deals that adjust quarterly based on performance and market conditions.
• Long-term: Recognize that control is shifting from relationships to systems. The power no longer lies in who you know, but in who--human or agent--can close the loop fastest.