Warner Bros. Discovery's ad revenue dropped 7% year-over-year to $1.85 billion in Q1 2026. Most companies in that situation would run optimization sprints — A/B test the sales pitch, tighten targeting, squeeze more yield from existing infrastructure. WBD did something fundamentally different. They announced they were scrapping their legacy advertising technology stack entirely and rebuilding it from the ground up around autonomous AI agents running on AWS.
This isn't a pilot. It's not an experiment bolted onto existing systems. It's a complete architectural reset — every workflow in their ad operations, from media planning to order management to campaign stewardship, handed to AI agents with humans retained for strategic oversight. The rollout is already underway, with unified media planning launching in Q3 2026 and composable order management following in Q4.
For enterprise leaders watching from the sidelines, this is the case study worth dissecting. Not because WBD is a media company, but because every enterprise has a version of the same problem: siloed legacy workflows that AI can't effectively augment — they can only be replaced.
What They Built (and Why It Had to Be a Rebuild)
WBD's old advertising operations looked like most large enterprises: separate systems for linear TV and digital channels, disconnected planning tools, manual order management, fragmented measurement. Each business unit had its own stack. Buyers had to navigate multiple interfaces to run campaigns across WBD's portfolio.
The problem with legacy ad tech isn't that it lacks AI features. Vendors have been bolting AI onto these systems for years. The problem is architectural. When your data lives in silos and your workflows were designed for human handoffs, adding AI to individual steps only creates faster local optima. You still have the bottlenecks between systems. You still have the latency of sequential handoffs. You still have the measurement gaps that prevent closed-loop optimization.
WBD's leadership recognized this and made the harder call. Rather than optimizing the existing system, they rebuilt around a unified data model — all inventory, all channels, all audience signals — with AI agents operating across the entire workflow rather than within isolated steps.
The business pressure helped. With ad revenue under pressure and a planned merger with Paramount on the horizon, there wasn't time for incremental improvement. The transformation needed to be structural.
The Technical Architecture
For CTOs evaluating similar initiatives, the WBD stack provides a concrete reference architecture built on AWS:
Amazon Bedrock AgentCore serves as the orchestration layer — the platform where AI agents are built, deployed, and optimized. This is where the agent logic lives: the decision rules, the tool calls, the memory management that allows agents to learn from prior campaign performance.
Amazon Bedrock hosts the foundation models themselves within WBD's closed environment, ensuring that proprietary audience data and campaign intelligence doesn't leave the controlled perimeter. This is a critical requirement for any enterprise managing sensitive commercial relationships.
Amazon SageMaker handles the custom ML models trained on WBD-specific data. Foundation models are general-purpose; the real competitive advantage comes from models fine-tuned on your own historical performance data, your audience segments, your inventory patterns. SageMaker is where WBD builds the differentiated intelligence layer.
Amazon S3 with Apache Iceberg provides the unified data lake — the single source of truth that enables agents to operate with consistent, real-time data rather than navigating between disconnected systems. The Iceberg format is particularly relevant for enterprises managing large-scale data that needs time-travel queries and schema evolution support.
Amazon ECS handles application hosting, and Amazon Q surfaces AI-driven insights to the Ad Sales team through natural language — allowing sellers to query campaign performance, inventory availability, and audience forecasts without writing SQL or navigating dashboards.
What's notable about this architecture is its intentional separation of concerns. The agents (AgentCore) are distinct from the models (Bedrock) are distinct from the training infrastructure (SageMaker) are distinct from the data layer (S3/Iceberg). This modularity matters: you can swap foundation models, retrain domain-specific models, or adjust agent behavior without rebuilding the entire system. That's enterprise-grade design, not a prototype.
What the Agents Actually Do
The autonomous capabilities WBD has already deployed in 2026 cover three domains:
Intelligent Planning and Dynamic Forecasting: AI agents handle media planning across linear and digital inventory simultaneously. Rather than planners manually building schedules and running forecasts, agents analyze demand signals, audience patterns, and inventory availability in real-time to generate and adjust plans dynamically. Demand forecasting updates continuously as conditions change, rather than running as a periodic batch process.
Real-Time Campaign Optimization: Agents monitor live campaign performance and make optimization decisions within defined parameters — adjusting targeting, pacing, and creative delivery without human intervention on each individual action. The system supports closed-loop measurement, meaning optimization decisions feed back into forecasting models to improve future performance.
Audience Targeting and Measurement: Flexible targeting across brands and audience segments, with attribution that spans linear and digital touchpoints. The unified data layer makes cross-channel measurement tractable in a way that siloed systems simply can't support.
Coming in Q3 and Q4, WBD will launch unified media planning (a single interface for buyers across all WBD channels) and composable order management, pricing, and stewardship (the operational backbone of ad operations, fully automated).
The Business Case for Enterprise Leaders
For CFOs and business leaders asking the obvious question — what's the ROI? — the WBD case frames it this way:
The old model required sequential human handoffs across every stage of the ad operations workflow. Proposals required planners. Orders required trafficking. Optimization required analysts. Measurement required reporting teams. Each handoff introduced latency, error risk, and cost.
The agent model replaces sequential handoffs with autonomous execution across the full workflow. Agents don't need handoffs. They operate asynchronously, continuously, across all channels simultaneously. The human role shifts from execution to oversight and strategic decision-making.
For WBD specifically, the business rationale is competitive differentiation. Fox has separately announced what it calls the "first end-to-end agentic advertising platform." The largest media companies are converging on the same conclusion: traditional ad tech is a competitive disadvantage in a world where buyers expect unified, real-time campaign management. The companies that maintain fragmented legacy stacks will lose share to platforms that can offer more efficient, data-driven buying.
For enterprises outside media, the parallel business cases vary by industry — but the structural logic is the same. Any enterprise running complex operational workflows with multiple hand-offs, high data volume, and continuous optimization requirements is a candidate for the same architectural pattern.
The Competitive Pressure Is Industry-Wide
WBD's announcement didn't land in a vacuum. Fox's competing claim to an "end-to-end agentic advertising platform" signals that this is an arms race, not a one-off experiment. When two of the largest U.S. media companies are simultaneously rebuilding core business infrastructure around agentic AI, it establishes a new competitive baseline.
The technology supply side supports this. Amazon Bedrock AgentCore only became widely available in 2025. The tooling to build reliable, scalable AI agents at enterprise scale didn't exist two years ago. The infrastructure is now mature enough that major enterprises can bet their core workflows on it — and the ones that move first gain the learning curve advantage. WBD's agents will have trained on real campaign data by the time competitors are still building their orchestration layers.
This dynamic is playing out across industries simultaneously. The enterprises that recognize the structural nature of the shift — and commit to rebuilding rather than optimizing — will exit 2026 with compounding advantages.
The Human Oversight Architecture
One of the critical enterprise governance questions around agentic AI is where human oversight sits. WBD has been explicit: "Humans will still be in charge of important decision-making." The agents execute; humans make the strategic calls.
This framing is more important than it might appear. The failure mode of poorly designed agentic systems is autonomous action without appropriate guardrails — agents making consequential decisions without the context to make them well. WBD's architecture assigns agents to execution tasks with measurable, bounded outcomes (adjust this campaign's pacing, reforecast this inventory block) while retaining human authority over strategic allocations (this advertiser gets premium placement, this deal structure is approved).
For enterprise leaders defining governance frameworks, this separation is the right model. Autonomy should scale with confidence and reversibility. High-frequency, low-stakes execution decisions are ideal for agents. Low-frequency, high-stakes strategic decisions need human judgment.
The Amazon Q integration for the Ad Sales team reflects the same principle on the business side — sellers aren't replaced, they're augmented with instant access to insights that previously required analyst support. The human value shifts from data retrieval to interpretation and relationship management.
5 Enterprise Lessons from the WBD Blueprint
Conversations with technology leaders running similar transformation programs surface consistent patterns in what separates successful AI infrastructure rebuilds from failed pilots:
1. Unified data before agents. Agents operating on fragmented data produce fragmented results. WBD's investment in a unified S3/Iceberg data lake precedes the agent layer. The sequencing matters. You can't build effective agentic workflows on inconsistent, siloed data.
2. Rebuild, don't retrofit. Adding AI features to legacy systems creates local optima. The step-change improvements come from redesigning workflows around agent capabilities from the start — eliminating the handoff latency and integration overhead that legacy architectures impose.
3. Separate training data from inference infrastructure. WBD's SageMaker layer for custom model training is distinct from Bedrock for inference. This lets you continuously improve domain-specific models without disrupting production operations. Enterprise AI systems need this separation to scale.
4. Define the human-agent boundary explicitly. Ambiguous governance is the fastest path to either under-utilized agents (humans override everything) or liability exposure (agents act without appropriate authority). Define the boundary by decision type, not job function.
5. Phase deployment to build trust. WBD didn't flip a switch — they deployed direct response automation first, then added forecasting, then unified planning, then order management. Each phase demonstrates value and builds organizational confidence in agent reliability before expanding scope.
The Bottom Line
WBD's decision to rebuild its $1.85 billion advertising operation around agentic AI on AWS is significant not because it's surprising — it's because it's inevitable. The economics of agent-native workflows versus legacy operational processes favor the former decisively at scale. The companies moving now are building the data assets, the organizational competency, and the compounding performance advantages that will define market position three years from now.
The enterprise AI conversation has spent too long focused on copilots and assistants — AI that helps humans do what they already do. The WBD case represents the next phase: AI that runs the operations, with humans setting strategy and maintaining oversight. That's not the future. For WBD, it's Q3 2026.
Rajesh Beri is the founder of THE DAILY BRIEF, a newsletter covering enterprise AI for technical and business leaders. He writes about AI implementation strategy, technology leadership, and the decisions that separate enterprise AI winners from casualties.
