Agents t ≈ 7 min

Ad Agents Now Launch Campaigns. Where Does Your Approval Gate Go?

InMobi Accelerate, Replit Agent, and Google AI Max all moved more ad decisions to software in November 2026. Here is how each one handles approval, what to audit, and how to decide what an agent may change without you.

yfx(m)

yfxmarketer

November 21, 2026

→

In one week of November 2026, two companies shipped agents that plan, write, and launch paid campaigns. Replit Agent started building and running Google Ads for its users on November 12. InMobi launched Accelerate on November 18, a set of seven AI agents that run campaigns across Google, Meta, Microsoft, TikTok, and ChatGPT ads.

Inside Google Ads, AI Max keeps taking over more of the decisions you used to make by hand. The question for paid teams isn’t whether agents will touch your ad accounts. It’s where you put the approval gate, and what you check after it.

TL;DR

Ad agents now cover the full loop: analysis, audiences, copy, creative, budgets, launch, and optimization. The vendors shipping them default to human approval before launch, and Replit enforces its spending approval in code rather than leaving it to the model. Google’s AI Max shows what happens after approval: the system can expand queries, rewrite copy, and change landing pages. Decide in advance which actions an agent may take alone, audit what it did weekly, and judge it on lead quality and margin, not conversion count.

Key Takeaways

  • InMobi Accelerate launched November 18, 2026 with seven agents and is free to start
  • InMobi presents every plan, asset, and budget for approval before launch; continuous optimization is opt-in
  • Replit Agent drafts copy, generates images, and recommends daily budgets for Google Search, YouTube, Display, and Discover
  • Replit’s purchase approvals are enforced in code, and the agent never gets open-ended spending permission
  • Google AI Max can expand search terms, generate copy, and swap landing pages; Dynamic Search Ads migrate in February 2027
  • One practitioner study of 250+ AI Max campaigns found a median 16% higher CPA and a median 0% ROAS change
  • Write down your autonomy rules before you connect any agent to a live ad account

What did InMobi and Replit actually ship?

InMobi Accelerate. A self-serve platform for small and mid-sized businesses and the agencies that serve them. The seven agents are a business analyst, audience strategist, media planner, copywriter, creative designer, launch engineer, and optimizer. You give it your website and a campaign objective. It returns analysis, audiences, a media plan, assets, and budgets. Channels at launch are Google Ads, Meta Ads, Microsoft Advertising, TikTok Ads, and ChatGPT Ads (PPC Land).

InMobi pitches it against a measurement problem. In a NewtonX survey InMobi commissioned of 800 US SMB advertisers and agencies, 93% said they can’t confidently measure performance across the platforms they use. Only 4% can get a campaign live the same day they decide to run it.

Replit Agent ads. Pro users can connect a Google Ads account or have Agent create one. Agent reads your site or app, drafts copy, generates images, and recommends daily budgets. You set the goal, targeting, budget, and landing page, then review, approve, and launch. Agent can add conversion tracking for sign-ups and sales and suggest changes from the latest performance stats (Replit).

Action item: If you run paid media for small clients, sign up for one of these on a sandbox account. Compare its media plan with what your team would have built for the same brief. The gaps tell you where it’s useful and where it isn’t.

How do these agents handle approval?

Both vendors put a human gate before money moves. The details are worth copying.

InMobi’s default is explicit: “Every plan, creative asset and budget is presented for approval before anything goes live.” Continuous optimization is something you turn on, “within parameters they set,” limited by rules, approvals, and role-based access. The announcement doesn’t list which parameters you can set or whether changes are logged. Ask before you turn it on.

Replit’s purchasing feature, shipped in the same release with Stripe’s Link and Visa, goes further. Agent stops at checkout and waits for the owner to approve the final total. It never gets open-ended spending permission. Shell access and general browsing are turned off during payment. And the approval checks are enforced in code, not left to the model.

That last point is the one to steal. A prompt that says “always ask before spending” is a request. A check in code is a control. If you build your own ad agents on Claude or another model, put the budget cap and the approval step in the tool layer, where the model can’t talk its way past them. We covered that split in deterministic orchestration for marketing ops.

Action item: For every agent with write access to an ad account, answer one question. If the model ignored its instructions, what would stop it from raising a budget? If the answer is “nothing,” move the cap into code or into the platform’s own spend limits.

What does Google AI Max decide after you approve?

Approval at launch isn’t the end of agent decisions. Google AI Max shows why.

AI Max makes three kinds of decisions in Search campaigns. Search term matching uses broad match and keywordless matching to reach queries beyond your list. Text customization generates extra headlines and descriptions from your ads, keywords, and landing pages. Final URL expansion picks a different landing page on your site based on the query (Search Engine Journal).

The rollout is mostly automatic. Automatically created assets and campaign-level broad match campaigns began moving to AI Max in September 2026. Dynamic Search Ads migrate in February 2027.

The data from practitioners is mixed. Mike Ryan of SMEC looked at 383 million impressions and found that more than 25% of impressions attributed to exact match keywords came through AI Max search term matching. In one account, competitor searches made up 69% of total Search impressions. Across 250+ campaigns, Ryan found a median 13% more conversion value, a median CPA 16% higher, and a median ROAS change of 0%. Individual campaigns ranged from 42% above to 35% below baseline.

A median of zero with that spread means AI Max can help or hurt any given campaign. You only know which one you got if you look.

Action item: Pull the AI Max search terms report for your top five campaigns this week. Flag competitor terms, brand terms, and any query whose landing page doesn’t match the intent.

What should you audit every week?

Use the same checklist for any ad agent, whether it’s AI Max, InMobi’s optimizer, or something you built.

Configuration. Which automated features are on, in which campaigns? Recently migrated campaigns are the most likely to have something on that you didn’t choose.

Queries and audiences. What new traffic did the agent buy? Check intent, competitor terms, and branded terms against cost and conversions.

Generated copy. Is every claim accurate and allowed? SEJ notes that pinned assets may not be respected when Final URL expansion or URL inclusions are on. If pinning matters for compliance, turn those off.

Landing pages. Which URLs got agent-routed traffic, and do they have a working conversion path? Use URL exclusions for pages that shouldn’t get paid traffic.

Conversion quality. Compare lead quality, revenue, ROAS, and margin with your existing traffic. Conversion volume alone will flatter the agent.

Change one setting at a time. If you flip three AI Max features at once, you won’t know which one moved the numbers.

Action item: Put these five checks into a recurring Monday task. If you already run Claude Code routines, a weekly routine that pulls the reports and flags exceptions is a good first build. See Claude Code routines for marketing teams.

How much autonomy should you give an ad agent?

Sort actions by two things: how much money they can move, and how easy they are to undo.

Agent acts alone. Reversible, low-cost actions inside a budget you already approved. Pausing a weak ad. Shifting spend between ad groups within a fixed campaign cap. Adding a negative keyword.

Agent proposes, human approves. New budgets or budget increases. New channels. New audiences. Any new claim in ad copy. Any new landing page.

Agent never does it. Billing changes, account access, and anything your legal team would want to see first.

This matches what the vendors ship. InMobi gates plans, assets, and budgets. Replit gates the final charge. For a broader view of how to scope agents by role, see hub-and-spoke agent architecture. For connecting agents to the rest of your stack, see MCP as the martech integration layer.

Action item: Write your three lists on one page and share them with whoever owns the ad budget. Get sign-off before any agent gets optimization rights.

Final Takeaways

Ad agents now cover the whole campaign loop. InMobi and Replit both shipped them in November 2026, across Google, Meta, Microsoft, TikTok, and ChatGPT ads.

The vendors default to approval before launch. The strongest version, from Replit, enforces that approval in code rather than in the prompt.

Google AI Max shows that agent decisions keep happening after launch. Practitioner data shows a median ROAS change of zero with a wide spread, so the result depends on your account.

Audit configuration, queries, copy, landing pages, and conversion quality every week. Judge agents on margin and lead quality, not conversion volume.

Decide what an agent may do alone before you connect it. Reversible and capped goes to the agent. New money and new claims stay with you.

yfx(m)

yfxmarketer

AI Growth Operator

Writing about AI marketing, growth, and the systems behind successful campaigns.

read_next(related)