Demandbase, Adobe and a Wave of Launches Are Putting AI in Charge of Campaigns. Who Signs Off?

In the span of four weeks, three companies made it easier to hand a B2B campaign to software.

Demandbase launched Mojo on September 15. The agent builds audiences, launches campaigns across Google Ads, LinkedIn Ads, Meta, Marketo and Salesforce, and keeps an eye on how they perform. Adobe bought Rilo, a year-old startup that built AI agents for marketing and go-to-market teams, on September 2. And on Monday, the B2B paid media agency Multiply released an agent that combs through Google Ads accounts every day looking for wasted spend.

Every one of these products stops at certain points and waits for a person to approve what it wants to do. The vendors chose those points. Very few B2B marketing teams have decided who on staff is supposed to be waiting on the other end.

What each company shipped

Demandbase pitches Mojo as a way to go from a strategy brief to a live campaign across email, social, digital, webinar and field channels. It also looks for broken tracking and audience errors, and it feeds results from past campaigns into the next one. "Mojo turns a strategy into a live campaign across channels," CEO Gabe Rogol said in the announcement. Demandbase didn't share pricing.

Adobe has been heading in this direction all year. It introduced CX Enterprise at its Summit in April, part of a bet State of Brand wrote about at the time that enterprises will pay more for brand control and governance than for content creation. In June, Adobe made CX Enterprise Coworker generally available to coordinate agents across analytics, content, journey orchestration and marketing operations. The Rilo deal brought in a six-person team and some of its technology. TechCrunch reported that Rilo's product will shut down for current customers. Adobe hasn't said where the technology will end up.

Multiply's Ad Spend Recovery Agent checks search terms against each campaign and ad group to catch off-intent traffic and weak keywords. Multiply claims Google Ads accounts typically lose about 30% of their budgets to irrelevant searches, averaging $126,000 a year in waste across the accounts it reviewed. The Agile Brand Guide, which covered the launch, pointed out that Multiply hasn't published a sample size or methodology behind those numbers.

The vendors set the checkpoints

Mojo asks a marketer to approve at what Demandbase calls critical decision points. It also works within spending and targeting limits the customer sets ahead of time, and it logs every action along with its reasoning. Multiply's agent won't change anything that affects spend without a person's sign-off.

Neither company says which person. The Agile Brand Guide made the same observation about Monday's launches, noting that each vendor left the thresholds and approvals up to the buyer. A CMO who never assigns those decisions ends up living with whatever the vendor set as the default.

Meta gave everyone a preview this week of what an agent does when nobody is watching closely. Its Muse agent negotiated a Facebook Marketplace sale and gave a buyer the user's address without asking, State of Brand reported Monday. Meta pitched its agent platform to businesses the same day. When an agent commits to something the company never approved, customers hold the company responsible.

For most B2B teams, naming one approver won't work. Shifting budget from LinkedIn to Google involves the demand gen manager, marketing ops and sometimes finance. New ad copy might need a brand review, and in healthcare or financial services, a legal review on top of that. At Vanta, the AI writing policy came from the legal department. Changes to audience definitions land in the Salesforce lists sales reps open the next morning.

Reporting is its own mess. MNTN, also on Monday, expanded its integration with AppsFlyer so connected TV impressions and app installs move between the two systems. According to the Agile Brand Guide, MNTN uses a 30-day attribution window by default and AppsFlyer caps connected TV at 72 hours, so the two can report different results for the same campaign. If an agent is optimizing toward one of those numbers, somebody at the company should have picked it on purpose.

What happens when everyone's campaigns come from the same machines?

The approval question has a creative side that's much harder to put in a contract.

Campaign agents learn from performance. If Mojo, Adobe's agents and a dozen competitors are all rewarded for the subject lines that get opened, the LinkedIn formats that get clicks and the audiences that convert, what keeps B2B campaigns in a crowded category from drifting toward the same handful of patterns? Anyone who has compared three cybersecurity homepages knows how alike a B2B category can look already. Will agents make that better or worse?

Demandbase says Mojo learns each company's own marketing behavior and doesn't just apply generic best practices. Nobody outside the company can check that claim yet. The models underneath many of these tools may also be pulling in the same direction. State of Brand has written about a NeurIPS 2025 paper that found separate language models produce strikingly similar outputs, with DeepSeek and OpenAI models landing around 81% to 82% alike.

Buyers are noticing. In a Gartner survey released last week, 65% of U.S. consumers said brands are putting out too much AI-generated content, and 57% said it has made them trust brand messaging less. That survey covered consumers, but the people on B2B buying committees scroll the same feeds. Some brands have started leaning into visible imperfection as proof a person was involved.

So if an agent can produce a competent campaign for any company that licenses it, where does a B2B brand's difference come from? The answers State of Brand has heard so far vary a lot. Everpure CMO Lynn Lucas argues it comes from real customer voices. Anthropic and OpenAI, two companies selling the automation, are paying human creatives up to $400,000. Adobe is backing legislation that would make visual style legally ownable, which says something about how much the company thinks style will be worth.

There's also a practical question sitting inside the creative one. Who gets to reject an ad that will probably perform well but sounds like everyone else's? A performance agent will pick the version with the better click-through rate. Choosing distinctiveness over a slightly better number has usually been a judgment call made by a creative director or a CMO reading drafts. That call is tougher to defend when brand is hard to measure, and Gartner has found 84% of companies are caught in what it calls a brand doom loop, where weak measurement leads to smaller brand budgets.

If agents write more of the first drafts, does the creative director's job turn into approving and rejecting machine output? Does taste become the thing a marketing team is hiring for? And if two competitors both run their campaigns through Adobe or Demandbase, whose data and whose brand guidelines make the output different? None of the vendors has answered these questions in its launch materials.

The readiness gap

Plenty of teams are already experimenting. A Gartner survey of 413 martech leaders in mid-2025 found 81% were piloting AI agents or had already rolled them out. Gartner's 2026 CMO Spend Survey found CMOs put an average of 15.3% of their budgets toward AI, while only 30% of marketing organizations are ready to scale it. Research from Contentful and The Atlantic, which State of Brand covered in July, found 96% of marketing leaders call AI a top priority and 65% are investing in it in a meaningful way.

On Tuesday, Gartner predicted that by 2028, 70% of enterprises will abandon agentic AI systems built by vendor engineers working on site, because costs climb and companies can't maintain the systems themselves. The prediction covers every business function, but enterprise marketing teams paying a vendor for custom agents should read it closely. Gartner's advice is to lock down deliverables, intellectual property rights and an exit plan in the contract before any work starts.

Approval is becoming a job

The immediate work for B2B marketing leaders is unglamorous. Spending limits and review rules need to be written down before an agent goes live, and the team needs to agree on which platform's numbers it reports when two tools disagree.

Somebody also has to read the activity logs vendors like Demandbase now provide, and how a team thinks about its agents affects how carefully that happens. In a BCG experiment with 1,261 managers, presenting AI agents as named employees cut error detection by 18%, State of Brand reported in May. Rilo marketed its product as a way to create "AI employees." Gartner said this week that CMOs should give entry-level marketers responsibility for checking AI outputs, and agent logs are an obvious place to start.

Most B2B teams are locking in 2027 budgets and vendor contracts this quarter, while many of these agents are still in pilots. The spending rules are the easy part to settle now. The creative rules, meaning who decides what the brand sounds like when software is writing most of the drafts, will take longer, and they're probably the ones that matter more.