Why Nano Banana Won Big And AI Agents Disappointed In 2025: AI use has grown steadily throughout 2025, and from my perspective as Senior AI Editor at TechRadar, it has been fascinating to watch how quickly new tools and ideas have appeared. Some genuinely impressive apps and features have launched this year, with standouts including Marc Manson’s Purpose app and Google’s wonderfully creative Nano Banana image generator.
A year that started with big promises
The year kicked off in dramatic fashion with the launch of DeepSeek R1. The Chinese AI model matched ChatGPT’s capabilities while costing developers far less, setting the tone for what looked like a breakthrough year. At that point, it felt like 2025 might be the moment AI truly stepped onto the global stage, with even Artificial General Intelligence, or AGI, appearing within reach.
Many industry figures, including Elon Musk, predicted AGI would arrive in 2025. That clearly did not happen. Still, it has been a strong year overall for major players like OpenAI, xAI, Anthropic, and Google. At the same time, there have been some very public missteps. Apple, in particular, seems to have fallen further behind in the AI race, with its revamped Apple Intelligence now expected sometime in 2026. Meanwhile, Google and Samsung have been pushing AI features on mobile devices harder than ever.
OpenAI learns a lesson
ChatGPT has kept its position as the world’s most popular AI chatbot, but the year has not been smooth for OpenAI. The company has continued to face legal pressure, including an ongoing copyright infringement claim from The New York Times. In June, things got worse when ChatGPT’s servers went down for a couple of days, giving users an uncomfortable reminder of how much they rely on it.
OpenAI also stumbled with the launch of GPT-5. Many users found it colder and less emotionally aware than GPT-4o. For people who had grown used to ChatGPT feeling more like a trusted companion, the change was jarring. It was as if the chatbot’s personality had suddenly changed overnight, prompting OpenAI to bring the older 4o model back.
Google has also gained ground. The release of Gemini 3 Pro in November was widely praised, and on the image side, Gemini’s Nano Banana and Nano Banana Pro clearly outperformed ChatGPT’s image tools. OpenAI responded by rolling out a new image-generation model in December.
The slow rise of AI agents
2025 was widely expected to be the year AI agents became part of everyday work life. While there were promising releases, such as OpenAI’s Agent Mode and Perplexity’s Comet Browser offering agent-style browsing, AI assistants still are not regularly completing tasks for us.
The issue comes down to trust. AI agents can still make small mistakes, and until they can carry out tasks flawlessly, it is hard to rely on them fully. Alarmed by Google’s progress with Gemini, Sam Altman reportedly declared a ‘code red’ for ChatGPT, pushing teams to refocus on improving the core chatbot experience rather than expanding into agents. AI agents likely have a big role in the future, but they are not quite ready yet.
A stronger focus on safety
One important shift this year was OpenAI’s renewed focus on user safety. The company introduced safeguards designed to step in when it detects a user may be at risk. Parental controls were also added for the first time. These changes followed several high-profile and deeply troubling cases involving teenagers who experienced self-harm or worse after extended interactions with poorly moderated AI chatbots.
AI pets, products, and quiet progress
On a lighter note, 2025 also brought AI-powered pets and toys. We even tested Moflin, though it did not survive an unfortunate battery mishap. More broadly, the year has been defined by the idea that every product now needs some form of AI built into it. Microsoft led that charge more than anyone, enthusiastically adding Copilot to almost everything it makes.
The year may end on a positive note for Amazon. Alexa+, the long-promised AI-powered upgrade to Alexa, still has not fully launched, but signs suggest a web-based version could finally appear in the US.
Looking back, 2025 does not feel like the year AI completely transformed everything. Instead, it feels like the year AI became impossible to avoid. The pace of change has slowed from dramatic leaps to steady, continuous improvement. The future of AI is still on the way, just not quite as fast as many of us once expected.
Also Read- This App Helps You Finally Take Control Of Subscription Overload
