Paid-search automation (the delegation arc)
The decade-long transfer of paid-search execution from the advertiser to the ad platform's own models, and now to a second automation layer the advertiser builds on top. groas and google-ads-ai-agent-roadmap both describe the newest step; this page holds the sequence they assume.
The staging comes from google-ads-ai-agent-roadmap: "We've already spent the last decade handing more execution to algorithms via Smart Bidding, broad match, and Performance Max. Agentic AI is simply the next stage of that evolution."
What was handed over, in order
Smart Bidding — the bid. The advertiser stops setting a price per click and sets an objective (target CPA, target ROAS, maximize conversions); Google's model picks the bid per auction using signals the advertiser cannot see. What moves is the decision, and with it the ability to inspect why any given click cost what it cost.
Broad match — the query. The advertiser stops choosing which searches to appear on and hands matching to the platform's understanding of intent. What moves is the targeting boundary: the account's reach becomes an output of the model rather than a list the advertiser wrote.
Performance Max — the placement and the creative. One campaign spans Search, Shopping, YouTube, Display, Discover, Gmail and Maps; the advertiser supplies assets and a goal, and the platform decides channel mix, placement and asset combination. What moves is where the money goes, and the campaign becomes a single reporting object rather than a set of levers.
Each step trades control for the platform's data advantage, and each is reported at a coarser grain than the one before. That is the through-line: the delegation and the loss of visibility are the same move, not two separate complaints.
Why the agent layer is a different kind of step
The first three steps handed decisions to Google. An agent is built by the advertiser and sits outside the platform, which makes it a reaction to the delegation rather than a continuation of it. Read that way, google-ads-ai-agent-roadmap's step 3 — joining ad performance against stock, margin, pricing and CRM data — is an attempt to rebuild the visibility Performance Max removed, using data the advertiser owns and the platform does not.
That reframes the ceiling argument. The roadmap holds that agents should audit, monitor and surface opportunities while humans keep judgment; groas sells an agent that runs the account outright. But the account it would run has already delegated bid, match and placement, so what the aggressive position actually claims is the supervision of Google's automation, not the execution of the campaign. The disagreement is narrower than it looks, and the two positions are recorded in synthesis as a live contradiction.
What the delegation collides with
The measurement caveat in b2b-ppc-metrics gets heavier at each step. Optimizing toward a misreported conversion or an average-rather-than-marginal CPA is a mistake a human makes slowly; a model tuned to that objective makes it at auction speed, and the layers above have progressively less reporting granularity for anyone to catch it in.
The opacity has a documented precedent. search-ad-auction traces Google's 2002 quality score as a published formula — bid × estimated click-through rate — while google-ad-rank today names six inputs, publishes no weights, and turns quality into a gate that can keep an ad out of the auction at any bid. The auction mechanism got less legible over the same twenty years that execution moved to the platform's models. The spoke's standing complaint about google-guidance-seo-authority — the party being optimized against writes the rules and meters the result — is the paid half's oldest pattern, not a new one imported from AI search.
Related
google-ads-ai-agent-roadmap · groas · agentic-adoption-ladder · search-ad-auction · google-ad-rank · b2b-ppc-metrics · seo-operating-model-shift · search-marketing · synthesis