The industry conversation around AI agents has largely focused on the “front-end”, with LLMs that can draft media plans or negotiate deals. However, for those of us operating in the high-stakes, high-speed world of programmatic execution, a glaring “Latency Tax” remains. Standard API-based agents, communicating over the open internet, simply cannot survive the 300-500ms round-trip requirement of a real-time auction.
To move from agentic experimentation to agentic infrastructure, we must look toward the Agentic Real-Time Framework (ARTF), the first pillar of the IAB Tech Lab’s AAMP initiative. Specifically, we must embrace the “Agentic Sidecar” architecture.
The Problem: Intelligence vs. Infrastructure
Historically, adding specialized intelligence to the bidstream,whether for identity resolution, fraud detection, or environmental triggers in DOOH,required deep, linear integrations. This “hard-coded” approach is the enemy of agility.
If an agent has to “call home” to a cloud-based LLM during a live auction, the network latency alone creates a bottleneck that results in missed opportunities and timed-out bids.
The Solution: The Agentic Sidecar
The ARTF v1.0 specification introduces a paradigm shift: Containerization. Instead of an agent sitting across the globe, it sits as a “sidecar” within the host platform’s (DSP or SSP) own virtual environment.
This architecture offers three critical advantages for the future of the ecosystem:
1. Deterministic Execution at Machine Speed
By using gRPC and Protocol Buffers (Protobuf) instead of legacy JSON, the sidecar architecture allows for binary, high-performance communication. This reduces bid request processing time by up to 80%. In this environment, an agent isn’t just “chatting”; it is executing Bidstream Mutations,surgical, atomic changes to a bid request,in sub-millisecond timeframes.
2. The “Propose-and-Apply” Governance Model
A common fear of autonomous agents is the “black box” effect. The sidecar model solves this through a structured Intent system. An agent doesn’t simply change a price; it proposes a mutation with a declared intent (e.g. ADJUST_DEAL_FLOOR or ACTIVATE_SEGMENT). This gives the host platform the control to programmatically audit and approve actions, ensuring that “Agentic” does not mean “Unregulated.”
3. A Universal Adapter for Emerging Channels
For specialized sectors such as Digital Out-of-Home (DOOH) and CTV, the sidecar serves as a universal adapter. Instead of asking every SSP to build custom support for real-world environmental triggers (like foot traffic or weather), a specialist agent can be “dropped in” as a sidecar. It ingests complex, non-standardized signals and translates them into standard OpenRTB Patches that the host already understands.
The Path Forward: 2026 and the Standardized Container
As we look toward finalizing ARTF and expanding the Agentic Audiences protocol, the industry’s task is clear. We do not need to rebuild the “track” of programmatic advertising; we need to modernize the “engines” that run on it.
By adopting a containerized, sidecar-first approach, we ensure that the next generation of advertising is not just smarter, but faster, more secure, and,most importantly,fully interoperable.
Gulab Patil
Founder and CEO
Lemma