Covering AI since ChatGPT burst onto the scene and changed everything, I’ve written hundreds of articles about artificial intelligence. And I can say that 2025 seems to be the year the AI industry is truly taking shape.
The tech world seems less focused on lofty dreams of artificial general intelligence (AGI), and more on practical matters, like building data centers. Take OpenAI, which is talking less about releasing its much-anticipated GPT-5 model. Instead, it’s working through limited GPU capacity and focusing on the features people are gravitating to, like the native image generator.
It’s also becoming abundantly clear that chatbots aren’t the be-all and end-all of the AI revolution. The most interesting products launched this year are more far-reaching autonomous bots, which can take control of your device and execute new tasks independently. This “agentic AI” might eventually change how we use our phones and computers.
Here are the most exciting AI releases from 2025 so far. But it’s early days, and there are still plenty of half-baked features—and full-blown flops—to go around. Stick around until the end for a look at the biggest fails of the year.
1. OpenAI’s Operator
OpenAI started 2025 with a bang, releasing Operator in January. This AI can surf the web and do tasks for you by “typing, clicking, and scrolling,” OpenAI says. It’s only available as a research preview right now, and OpenAI is likely still working out some kinks, but it’s a big step forward.
Those with a ChatGPT Pro account ($200 monthly) can try Operator at operator.chatgpt.com. You describe the task, and Operator will complete it using the same websites a human would; for instance, it can order groceries on GrubHub. When it needs to enter login or payment information or solve a CAPTCHA, it will pause and ask the human for help.
OpenAI is also experimenting with autonomous AIs trained in specific jobs. Shortly after launching Operator, it debuted Deep Research for Pro subscribers. It creates comprehensive reports on the subject of your choice. Suppose you ask it about the impact of President Trump’s tariffs on a sector of the US economy. It will “find, analyze, and synthesize hundreds of online sources…at the level of a research analyst.” At one point, OpenAI reportedly was looking into offering a version of it for $20,000 per year to replace PhD researchers.
2. Samsung Galaxy ‘Cross App Actions’
Samsung Galaxy S25 Ultra (Credit: Iyaz Akhtar/PCMag)
The Samsung Galaxy S25 Ultra beat the iPhone to the punch by releasing an AI voice assistant, called Cross App Actions, that can field your complex requests using Google Gemini.
“I tested this with the request, ‘Find pet-friendly restaurants in Clearwater, Florida, and send it to Robert,'” writes PCMag mobile analyst Iyaz Akhtar in his device review. “Gemini compiled a list of such restaurants with an online search, organized it, and then put it in a text message to my contact, Robert. You have the option to edit the text or send it as is. I then tried, ‘Can you create a workout plan for a 45-year-old man and put it in Notes?’ Sure enough, the S25 Ultra created a list for me in Google Keep.”
Although these simple examples may not seem groundbreaking, they are a huge leap forward for AI assistants. Amazon is working on adding the same capabilities to Alexa+ and Apple to Siri (but neither have quite figured it out yet).
3. DeepSeek Upstages the US AI Industry
DeepSeek app (Credit: Emily Forlini/PCMag)
The DeepSeek chatbot exploded out of China in late January, causing the US stock market to plunge and making American AI companies ask themselves, “How are they offering that ChatGPT-level of performance for a fraction of the cost?”
In our testing, we found it to be very similar to ChatGPT, but with more privacy concerns and CCP propaganda. Still, Microsoft, Amazon, and Perplexity swiftly integrated the tech, offering it as an option for developers looking for a cheap API.
Questions remain about how DeepSeek developed the product, particularly its claims that it used fewer Nvidia chips and its less-advanced ones. The Trump administration is investigating what happened and trying to restrict the flow of chips to Chinese AI companies, The New York Times reports. If indeed DeepSeek can offer comparable performance using less energy and fewer data centers, it could push the US AI industry to be more efficient.
4. Microsoft Copilot Vision
This AI product from Microsoft can see what you’re doing on your computer and chat with you about it. It can even tap into your phone’s camera to see your surroundings and answer your questions.
It’s free now in the Edge browser for those who opt in. “It’ll think out loud with you when you’re browsing online,” Mustafa Suleyman, CEO of Microsoft AI, said on Bluesky. “No more over-explaining, copy-pasting, or struggling to put something into words.”
Not having to craft the perfect text prompt can provide humans with a less stressful, more seamless relationship with AI software. But it’s not perfect. PCMag’s Microsoft expert Michael Muchmore says earlier versions of Cortana had more capabilities. Still, Microsoft no doubt will continue evolving Copilot Vision and releasing similar products, like Copilot Pages, which helps you draft and save documents.
5. ChatGPT Gets a Humanlike Memory
Chatbots are getting smarter and more humanlike; they can now remember what you’ve told them. They can reference your conversation history, recalling that spouse you mentioned earlier or a project you’re working on that might be related to a different topic you’re discussing.
Get Our Best Stories!
Your Daily Dose of Our Top Tech News
By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
ChatGPT might say, “Are you asking about the risk of toxic paint fumes because last week you told me you’re painting your house?” Then it could customize its response: “OK, since that’s what you’re talking about, here are the statistics on air quality in freshly painted houses. Since you’re only painting a few small bedrooms, I wouldn’t worry.”
OpenAI CEO Sam Altman said that ChatGPT will “get to know you over the course of your life and become extremely useful and personalized.” In April, Microsoft and X added the same capability to Copilot and Grok, respectively.
6. Waymo’s Rapid Self-Driving Car Expansion
Waymo self-driving cars (Credit: JasonDoiy / iStock Unreleased via Getty Images)
There are more self-driving cars on the road than ever. Waymos are now commonplace in major cities like San Francisco and Los Angeles, where many riders love not having to talk to a human driver. You can call a Waymo in the Uber app in Austin and, soon, Atlanta.
Waymo expanded to new parts of the Bay Area this year, and has plans to test in 10 more cities, starting with Las Vegas and San Diego, according to The Verge. Its next big challenge is operating in places without perfect, sunny weather, such as rainstorm-prone Miami and snowy locations in the Midwest and Northeast.
7. ChatGPT’s Image Generator
(Credit: ChatGPT/Emily Forlini)
ChatGPT had arguably its most successful product launch this year since the chatbot’s original 2022 debut. It swapped out its DALL-E image generator for a homegrown one that operates on its GPT-4o model—and people went bananas for it.
Social media filled up with images in the style of Studio Ghibli (even though its creator Hayao Miyazaki has called AI an “insult to life itself”). It was so popular that OpenAI briefly restricted access, citing a shortage of GPUs—a nagging problem for the company. OpenAI also announced a more lenient, user-policed approach to image guardrails. People are already using ChatGPT’s image generator to create disturbing pictures and deepfakes, and controversial symbols like swastikas are now allowed.
Recommended by Our Editors
OpenAI’s Altman has big plans for the image generator. The company reportedly is developing a social media network centered around it.
8. Claude Plays Pokémon
Claude plays Pokémon on Twitch (Credit: Twitch)
A funny thing happened in February: Anthropic’s newest model started playing Pokémon Red. For days, fans gathered on a Twitch livestream to see if Claude could navigate through the correct doors, collect badges, and beat levels.
The model, Claude 3.7 Sonnet, uses its reasoning capabilities to solve problems without human input. It progressed further in the game than earlier models. This was an easy way to demonstrate AI advancements without a confusing spreadsheet of metrics and benchmarks, which is what AI companies usually publish.
9. AI for Birdwatchers
A dickcissel visits the Bird Buddy Smart Feeder Pro (Credit: Jim Fisher/PCMag)
We snuck this one into the end of the list, mainly because it’s so fun. But also because it shows how AI is changing everyday gadgets. The Bird Buddy Smart Feeder Pro identifies the species of feathered friends visiting your feeder (without subscription fees). It’s more accurate than other smart birdfeeders we’ve tested, indicating that the tech is improving.
We’re seeing AI-powered cameras crop up everywhere, from Meta’s Ray-Ban glasses to deadbolts with facial recognition. Soon, Amazon Alexa+ will hook into Echo Show and Ring cameras to answer questions about what’s happening in your house. Apple is also reportedly considering adding cameras to AirPods and the Apple Watch.
Fantastic Fails: Worst AI Tech of 2025 So Far
Even with all this excellent AI progress, the industry is still in “throw everything at the wall and see what sticks” mode. That means a lot of half-baked features—and some total flops.
One Tesla Cybertruck totaled itself by crashing into a pole. OpenAI is running out of GPUs to power new features and still kicking the can down the road on GPT-5. Apple has been sued three times for allegedly falsely advertising the iPhone 16’s AI capabilities. Unfortunately, Amazon is seemingly adopting the same “over-promise, under-deliver” approach with the Alexa+ rollout.
Meanwhile, PC makers are adding AI to every nook and cranny, resulting in lots of bloatware. The Asus ProArt PX13, for example, added a “creepy cartoon robot mouse as an always-on-top AI assistant—like Clippy, but somehow not as cute,” our PC analyst Eric Grevstad writes. No one needs that!
Society is also learning more about the negative impacts of AI. It can make us lonelier, push out fake news, gobble up electricity, and harm reputations. One big question this year will be whether the risks are worth it, and how the public will continue to wield this powerful new technology.
Disclosure: Ziff Davis, PCMag’s parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
About Emily Forlini
Senior Reporter
Leave a Comment
Your email address will not be published. Required fields are marked *