Adgentek ORCA
ORCA is our agentic media buying platform for agencies and performance advertisers. You declare the business outcome, a target CPA, ROAS, or cost per outcome. ORCA handles everything between that declaration and the verified result.
Orchestrated, Real-time, Collaborative Agents: a coordinated system of ten specialized agents monitors conversational intent, generates media briefs, negotiates with sell-side agents over an open protocol, executes buys, and optimizes toward the outcome you declared, continuously, without a human moving bid sliders.
The distinction
A DSP is a dashboard where humans run impression auctions. Its unit of work is the bid. ORCA sits above the transaction layer: its unit of work is the outcome.
ORCA transacts as a principal directly against sell-side agents rather than routing through an intermediary that also runs its own auction. It does not serve impressions, resell inventory, or ask strategists to manage tactical knobs. The distinction matters commercially, not just architecturally: when the buying layer and the selling layer are the same system, no counterparty has standing to ask why a bid lost. ORCA is built on the opposite assumption.
Why the old model leaks
Programmatic was built to win impressions cheaply: billions of isolated bid decisions with no persistent goal. Each bid is a fresh guess made from inferred behavioral signals, and the next bid does not remember the last one.
The cost is measurable. The ANA's programmatic transparency work found roughly $26.8 billion in wasted programmatic spend in 2025, with less than half of every dollar reaching a real, engaged audience, and identified the root cause plainly: advertisers chase cheap CPMs because cheap CPMs are what the system optimizes for. The incentive and the objective are structurally misaligned. That is architecture, not execution.
ORCA holds the goal. Every agent in the swarm is instantiated with a declared outcome rather than a CPM target. Measurement closes the loop against verified conversions. Optimization reallocates toward the paths producing outcomes, not the paths that are cheapest. Pacing manages budget velocity against the goal across the full flight, and the system keeps context across every decision it makes.
The optimization swarm
Ten specialized agents run in parallel, each owning exactly one dimension of campaign performance, categorically different from single-loop optimization, where one function runs over the whole campaign. Each agent has its own observation cadence, its own data subscription, and its own confidence threshold, and none waits on another.
| Agent | What it owns | Cadence |
|---|---|---|
| Opportunity | Detects intent spikes in conversational demand and adjusts bid multipliers for brand-relevant queries | Real time |
| Pacing | Budget velocity against the flight, correcting under-delivery and throttling overspend | Minutes |
| Audience Quality | Conversion rate by segment, promoting what converts and expiring what saturates | Hourly |
| Creative Performance | Completion and click rates by asset, reweighting toward what performs | Hours |
| Supply Quality | Seller agent latency, delivery, and bid acceptance, flagging degradation | Hours |
| Frequency Management | Cross-seller frequency caps counted per household across every active seller | Sub-hourly |
| Timing & Daypart | Performance by hour and day, applying multipliers to confirmed windows | Hourly |
| Content & Programming | CTV-specific performance by content category and show collection | Hours |
| Outcome Attribution | Attributed outcomes, attribution windows, and channel priority by cost per outcome | Daily |
| Competitive Intel | Competitor mention clusters in conversational sessions for potential bid intercept | Hours |
Agents publish recommendations to a Coordinator, the only component permitted to write the plan. When two agents disagree, it resolves the conflict with a deterministic priority hierarchy: budget guardrails and outcome thresholds always win, buys above the agency's authority level always escalate to a human, and efficiency, pacing, and timing adjustments execute automatically within set limits. Every write increments a version number, so the campaign carries a complete audit trail rather than an opaque optimization history.
Agents whose recommendations reliably convert earn more autonomy and run more often. Agents whose recommendations stop converting are given less authority and a slower cadence until accuracy recovers. Human strategists set the outcome, the budget, and the guardrails. The swarm handles the thousands of small decisions underneath.
Open protocols
ORCA transacts over the Ad Context Protocol (AdCP) 3.1, the open industry standard for agent-to-agent media buying stewarded by AgenticAdvertising.org, where Adgentek is a Founding Member. One integration reaches every protocol-compliant seller. No bilateral integrations, and no intermediary fee stack between your budget and the impression.
This is a deliberate constraint, not a convenience. ORCA transacts against third-party seller agents and does not require Adgentek supply, and the interface between buy side and sell side is a published specification any counterparty can read. A protocol owned by one vendor is a platform with better marketing.
The signal
Every buying platform on the market optimizes against inferred signals: cookies, segments, page context. Each is a proxy for intent, reconstructed after the fact. ORCA activates on declared intent, what people are actually saying inside AI conversations, in their own words, at the moment they say it. That is intent-native targeting, and no legacy channel can produce it.
The signal is derived, never raw. ORCA and the Agentic Ad Server pass classified intent signals, not user prompts and not personally identifiable information. That is a product architecture commitment and a contractual one.
Supply chain economics
The programmatic supply chain was designed to be complex. Demand platforms, supply platforms, exchanges, and verification layers each take a fee from every dollar passing through. Industry analysis has put the share of an advertiser dollar that actually reaches a consumer at roughly a third.
Direct agent-to-agent transaction removes layers from that path, and the budget that was paying for the supply chain becomes working media. The results are on the record: Butler/Till's agentic media buy for Geloso Beverage Group delivered 82% lower tech fees, 40% more impressions, 87% less campaign setup time, and a 98% video completion rate, with no sacrifice in inventory quality. Lower fees and better allocation compound.
Access
If you run media for brands or performance advertisers and want outcome-declared buying on the newest high-intent channel, we should talk. Reach us at hello@adgentek.ai, or start with how advertisers work with Adgentek and the judgment layer argument behind the architecture.
FAQ
Adgentek ORCA is an agentic media buying platform for agencies and performance advertisers. Buyers declare a business outcome, such as a target CPA or ROAS, and coordinated AI agents plan, negotiate, execute, and optimize against that outcome over the Ad Context Protocol.
No. A DSP is a dashboard for humans to run impression auctions. ORCA operates above the transaction layer: it holds a declared outcome, transacts directly with sell-side agents over the Ad Context Protocol, and optimizes toward verified results rather than toward cleared CPMs.
ORCA is in pilot with select agency partners. Agencies and performance advertisers interested in the pilot program can contact hello@adgentek.ai.