A new "Agent" campaign type surfaced in ChatGPT Ads Manager that swaps the destination URL for a live, advertiser-configured AI conversation wired to product feeds, MCP tool access and lead capture; e-commerce analyst Juozas Kaziukenas spotted it in his own account. OpenAI hasn't announced it, priced it, or explained how you measure a conversion that closes inside a chat thread.
Why it matters: If the post-click experience becomes a configured agent instead of a page, the craft shifts from landing-page testing to agent configuration. Worth learning before the platforms document it.
Urban Company's CEO said on the earnings call that creative production, campaign deployment, optimization, learning and redeployment are all AI-run in-house, alongside 90–95% of the company's code. The signal isn't the tooling, it's a client absorbing the execution layer agencies used to bill for.
Why it matters: Brands are insourcing execution, not just experimenting with it. The defensible work moves to designing the system and its guardrails rather than running the campaigns.
Netherlands shop Draft Digital collapsed five specialist buyers hitting twelve platform endpoints into a single Buyer Agent, arguing the legacy briefing chain multiplies token spend and dilutes strategy while keeping channel specialists in advisory roles. No published cost comparison yet, but an IAB Europe demo burned roughly €50 in tokens across three agents on one task.
Why it matters: Token cost per decision is the unit economic nobody quotes in agentic pitches. Put it in the build spec and the architecture argument gets a lot more concrete.
Magnite's CEO collected forecasts at Cannes and landed on $600–700M as the optimistic case, a rounding error against programmatic, with conventional buyers still holding a 13.4% CPM advantage over agents. Near-term justification for rebuilding buying workflows is efficiency and access, not spend migration.
Why it matters: A useful ceiling when someone asks whether to wait. Build for operating leverage now, because transaction volume isn't the reason yet.
GBRAID and gad_ parameters get stripped by redirect chains, Safari, CDNs and consent denials, landing paid sessions in GA4 as organic or (not set) and inflating the channel you don't pay for. There's now a campaign-data-accuracy diagnostic plus a View URLs control to find the affected paths.
Why it matters: Every agentic optimization loop inherits the attribution it's fed. Run the diagnostic before trusting an agent to reallocate budget.
Wholly synthetic uploads are now excluded from Spotlight's recommendation engine while human footage enhanced with Snap's own AI tools keeps full reach and gets a transparency label. Snap published no detection method, no threshold for where enhancement ends and generation begins, and gave advertisers no way to filter synthetic inventory.
Why it matters: Platform-by-platform rules on synthetic creative are becoming a distribution constraint. AI creative pipelines need a provenance record and a human-footage input, not just volume.