A company that builds AI tools for Google Ads has an obvious incentive to answer this question loudly in one direction. Here is the answer anyway, with the incentive declared: no, AI will not replace PPC managers — and the managers pretending nothing is changing are at more risk than the ones who put an agent to work on the account.
The answer first: what 60/40 means
Write down everything a PPC manager actually did last month and the list splits cleanly in two. One part is mechanical: pulling reports, scanning search terms, checking pacing, drafting ad copy, applying negatives, building campaign scaffolds, reconciling what changed. The other part is judgment: setting budgets against business economics, choosing what to measure, deciding risk, and being the person a client holds accountable.
The mechanical part is the majority of the hours — call it 60/40 as a working estimate, not a measured statistic; your own split depends on your accounts. The claim of this page is narrower and harder to dodge: the mechanical majority is already better done by machines, and the judgment minority is not automatable with anything shipping today.
The 60: tasks already better done by machines
A connected agent — an AI client with live account access through an MCP connector — now runs these in seconds, faster and usually more thoroughly than a human grinding through a dashboard:
- Querying. Any question the account can answer, answered in seconds instead of a pivot table: wasted spend, pacing, device splits, n-gram patterns in search terms.
- Monitoring and reconciliation. What changed, when, and by whom — read straight from change history instead of memory.
- Drafting. Responsive search ad variants, keyword lists, campaign scaffolds — grounded in the account's own data, created paused for review.
- Sweeping. The weekly negative-keyword pass, the disapproval check, the asset-coverage audit: repetitive, rule-shaped, tireless.
- Reporting. The Monday summary and the client-ready narrative, assembled from live numbers rather than screenshots.
None of that is speculative. It is what a first week with an agent looks like now.
The 40: what stays human, and why
Four things resist automation for structural reasons, not sentimental ones.
Economics. The account optimises toward whatever target a human chose. Whether a lead is worth pursuing at this cost, whether margin supports the bid, whether growth or efficiency wins this quarter — those are business decisions expressed in ads, and no agent holds the context.
Strategy. Which markets, which offers, which channel mix, when to deliberately overspend. Strategy is choosing among futures; models interpolate from pasts.
Accountability. Someone signs off the spend, owns the miss, and explains it. An approval-gated agent — the design argued for in autonomous agent levels — makes the accountable human faster; it cannot become them.
Trust. Clients and CFOs hand budgets to people. The relationship through which advice becomes action is not a workflow step that software can occupy.
What the industry's moves actually signal
Watch actions, not thinkpieces. Google shipped Ads Advisor — an agent inside Google Ads that analyses, troubleshoots and builds — and trade coverage such as Marketing Brew's reporting frames it plainly as further automation of the ad platforms. Note what Google automated: execution and advice inside the product. Note what it did not and cannot automate: whose money it is, and who answers for the results. The platform's own agent still needs a counterparty with judgment — arguably more than before, since advice from the seller's agent deserves scrutiny, not deference.
Meanwhile in-house teams are quietly restructuring around the new split — fewer hours on production, more on measurement and strategy — a pattern described in how in-house teams use agents.
What PPC managers should do about it this quarter
Four moves, all boring, all compounding:
- Connect an agent to a real account and give it your mechanical layer — reads first, then approval-gated writes. Learn what it does well and where it is confidently wrong.
- Reprice your time. Hours freed from production go into the 40: measurement design, budget strategy, and the conversations where trust is built.
- Make judgment legible. Write down the decisions you make that no tool made — targets chosen, risks declined, recommendations overruled. That document is your role's case, for a client or an employer.
- Keep the fundamentals sharp. Auction mechanics, attribution, incrementality. The operator who can tell a convincing agent answer from a correct one is the scarce input now.
Why a tools vendor is telling you this
Because the alternative pitch is false, and false pitches cost more than they earn. AdCopilot's own design assumes a human in the loop: writes surface for approval, deletion is impossible by architecture, and the audit trail exists so a person can answer for every change. That is a bet, made in the product, that the judgment layer stays human. The longer argument — task by task, with the honest uncertainties — continues in what remains when AI takes over PPC.