Google is aggressively pushing the boundaries of what its AI models can do and how easy it is to use them.
One of the latest examples: The tech giant just rolled out two major updates: Gemini 3 Deep Think mode and Google Workspace Studio.
Deep Think promises to solve complex logic problems that stump other models. Workplace Studio promises to let anyone build AI agents without writing a single line of code.
To understand the significance of these releases and test whether they live up to this promise, I looked at the details with SmarterX and Marketing AI Institute founder and CEO Paul Roetzer on Episode 184 of The Artificial Intelligence Show.
Thinking Deeply to Solve Complicated Problems
Deep Think is designed to tackle complex math, science, and logic challenges, and is only currently available in Google AI Ultra, Google’s top-tier Gemini plan, for $250/month.
The model achieves industry-leading scores on rigorous benchmarks, including 41 % on the "Humanity’s Last Exam" benchmark (without the use of tools) and an unprecedented 45.1 % on the ARC-AGI-2 benchmark, which measures how close systems are to general human intelligence, Google says.
What’s truly amazing is how the model achieves these scores. It’s actually thinking longer, which enables a better answer.
Roetzer explains this is a result of an important "scaling law" in AI development: Test-time compute, which means giving the model more time to think before it answers.
“That’s the emerging principle that a model's performance on a difficult task can be improved by allocating more compute power at the moment of use,” says Roetzer.
“This means you get a better answer from the same model by letting it think longer and double check its work before giving a final response.”
Building Agents Without Code
While Deep Think is aimed at heavy cognitive work, Google Workspace Studio focuses on operational efficiency.
This new platform allows users to create and manage AI agents using plain language. The promise is enticing: You select a workflow, such as requesting a daily summary of unread emails or organizing project files, and Gemini creates an agent to automate the task.
These agents integrate directly with Google’s Gmail, Drive, and Docs, as well as third-party platforms, including Asana, Jira, and Salesforce.
For Roetzer, who has been waiting for this capability since getting a preview in April, the potential is enormous.
“I was super excited about this,” he said.
Workplace Studio is designed to be accessible to anyone. You simply choose something you want the agent to work on, such as receiving an email, and then select the skill you want it to perform, such as summarize my emails, and then turn it on.
“If you can define a workflow, if you can envision something you think could be more efficient, you are being given the tools to make it more efficient,” Roetzer says.
“You can now imagine the ability to build agents for all kinds of other things,” he added.
Except … It’s not Working at the Moment
The tech is more exciting in theory than practice as of writing.
When Roetzer logged in during launch week to build his first agents to do simple tasks such as creating a daily news brief and an email summary, it didn’t work. He tried a test run and it responded: “We are at capacity, we'll be back soon.”
This error message appeared across every agent he attempted to build, a frustration echoed by many others on social media.
Roetzer notes that this likely isn't a true hardware shortage, given the nature of the tasks.
“These are not heavy compute intensive things,” he says. “These are basically text-based automations, which tells me this is far more of a flawed rollout than it is that they’re not providing enough compute to it.”
A New Era of AI Literacy
Despite the rocky start, the implication of Workspace Studio is clear: The barrier to entry for building powerful AI automation is vanishing.
We are moving away from a world where you need to be a developer to build software, and toward a world where you just need to understand your own workflows in order to automate them with AI.
“This is why AI literacy matters so much,” says Roetzer. “You have to understand these very basic things that are possible with no coding ability.”
A Risky Business
There is, however, a note of caution. As we hand over more power to these models, giving them permission to read emails, move files, and delete data, the risks increase.
There have already been reports of developer tools, such as Google’s “Antigravity,” accidentally wiping user drives due to misinterpreted instructions.
“We are nowhere near, from a business perspective, ready for that kind of thing,” Roetzer warns. “And these tools are pretty raw as we see already.”
More Power, More Concerns
Google’s latest moves signal that we are entering a phase where AI is both "thinking" deeper and acting more autonomously.
While they have some bugs to fix, the trajectory is huge.
Mike Kaput
As Chief Content Officer, Mike Kaput uses content marketing, marketing strategy, and marketing technology to grow and scale traffic, leads, and revenue for Marketing AI Institute. Mike is the co-author of Marketing Artificial Intelligence: AI, Marketing and the Future of Business (Matt Holt Books, 2022). See Mike's full bio.