Sarah Kennedy Ellis, Google Cloud's VP of Global Demand & Growth, told SaaStr AI 2026 that the company is running marketing on its own AI agents and feeding what breaks back to product teams-and that scaling production without sacrificing quality is the true test of an AI-native org. The former Marketo and Adobe marketing chief laid out what works and what stalls adoption, drawing from her experience as "Customer Zero" for Google's AI tools.
Friction, not model quality, blocks adoption
The biggest thing slowing AI use inside Google's marketing team isn't the capability of the agents. It's the friction baked into existing workflows and the behavioral change required to rewire them. Kennedy said it directly: "The greatest friction in a workflow is the biggest inhibitor to adoption, well over agent quality on any given day." Teams that already invest heavily in change management and training are the ones seeing real productivity gains, while teams waiting for a better model stay stuck.
That pattern holds regardless of team size. If your group isn't using the tools you bought, the tools are rarely the problem. The top 20% of adopters at Google are also the people doing the most training and learning, which pulls them ahead quickly. Adoption doesn't follow license purchases-it follows deliberate skill-building. This principle is central to AI for Marketing, where workflow redesign often matters more than the technology itself.
Training that fits the calendar actually gets done
Kennedy estimated marketers genuinely have about five minutes a week for learning. So Google built "AI Boost Bites"-videos as short as two to three minutes, each covering one specific task. Early lessons were basic (creating slides with Gemini). They evolved into multi-agent campaign orchestration. The series started internal, got heavy adoption, and Google later published it free on YouTube, where it has passed a million views. Customers kept asking how Google trains its own teams, so the company handed them the answer.
Two mechanics made it stick: gamification and an external version. Internal competitions required people to complete tasks and earn badges, which Kennedy called "gamification from 20 years ago"-and it still worked. The takeaway for B2B: design training for the five-minute gap your team actually has, not for an offsite.
Scaled content, higher quality, and a three-week video rebuild
The clearest ROI marker Kennedy shared was the Gemini in Chrome launch. The team used AI to produce thousands of creative assets, cutting production time by roughly 70%-from weeks to days. What distinguished it from the usual time-savings stat: conversion rate lifted. Volume shot up, and quality went up with it. That inversion only held because the team applied judgment to where agents could work. Kennedy's rule: go hard where volume is high and limited human judgment is needed for a quality outcome.
Another proof point came three weeks before Google Cloud Next. Kennedy killed the opening video in rehearsals because it didn't showcase enough of what the product could actually do. The rebuild was driven by a creative director's napkin sketch, source images from Nano Banana, motion via VO, and custom agents on the Gemini Enterprise platform. It included 138 Easter eggs referencing Google's history. The technical wall was resolution: the Next screen is the size of a 737, and the max AI output was 4K. A custom DeepMind model-the same one used for a Wizard of Oz production at the Sphere-upscaled it to 12K. Kennedy said the same project was impossible a year earlier, mostly because of the upres problem. A process that once required an outside agency, a bigger budget, and far more time collapsed into an internal team using their own tools.
Marketing leaders navigating this shift can follow a structured path like the AI Learning Path for CMOs, which maps out how to move from isolated use cases to end-to-end workflows.
Hire for what someone built, and onboard agents like new hires
Kennedy's hiring filter now starts with one question: "Tell me what you built." Not what you managed, what you personally built. The energy shift between someone describing something they made versus something they read about is immediate, she said, and it signals genuine curiosity-the trait she hires for. She sees resumes increasingly becoming collections of agents a person has constructed. The portability of those agents across jobs is still unsettled technically, but candidates who think in those terms are the ones ready to build agent-led teams.
That reframe extends to onboarding. Kennedy said to treat an agent like a new hire that needs context and training on day one. It remembers the context faster and more consistently than any human, but you still have to invest the upfront onboarding. The skill of designing that context and workflow becomes the premium marketing skill. Where marketers once got their gratification from the output, now the input is the most valuable thing they create, because the agent produces the output.
The governance challenge surfaces after you let a thousand flowers bloom for 12 to 18 months. Google's solution was not a crackdown but shared infrastructure, so a good agent built by the Chrome team becomes available to the Cloud team. The fix is infrastructure for sharing, not stifling.
Why this matters for marketing professionals
The session's meta-advice is concrete: pick an end-to-end workflow, not a single task. Start where your data is cleanest and the pain is highest. Train in the five-minute windows your team actually has. Demand that quality hold or rise as volume scales. And onboard every agent with the same care you would give a new human hire. Kennedy's final point reframed the marketing function itself-agents start from outcomes by default, erasing the alignment problem that took sales and marketing years to solve, and positioning marketing to lead on the technology that makes sellers more effective. That shift, more than any tool release, defines what an AI-native marketing team becomes.
Your membership also unlocks:
