Another year into society's great ChatGPT experiment, AI is proving adept at amplifying workers' productivity, or taking over work altogether.

Why it matters: OpenAI first released ChatGPT on Nov. 30, 2022. Three years later, every worker's future hinges on avoiding automating themselves out of a job.

The big picture: The past three years have felt like 30 in terms of ChatGPT's infiltration into our lives. And if you believe the AI accelerationists, there are few signs of slowing down.

• As AI models evolve, chatbots will make fewer mistakes, requiring humans to adapt from fixing AI errors to directing AI agents' work.
• Understanding how to manage AI will be the key skill. And any good manager knows that delegating all of their concrete tasks to direct reports means that their own skills quickly atrophy.

Zoom in: Researchers now have enough data to show how AI can be a time saver or a time sucker.

• OpenAI researchers found that frontier models can complete certain tasks roughly a hundred times faster and cheaper than experts.
• Yes, but: Workers spend an estimated 41% of their time on fact-checking and reworking AI-generated memos, reports and emails, according to research from Stanford Social Media Lab and BetterUp Labs.

Between the lines: AI systems are powerful amplifiers for people who already bring expertise to the table, business leaders say.

• That leaves college grads and early-career workers who lack deep experience struggling to stand out or get hired at all.
• As the technology spreads and drives layoffs, only "using AI" at work doesn't necessarily translate to job security.
• Humans are training AI in their own domain expertise, which then risks devaluing those same skills when the models take over.

Zoom in: There's also a growing belief that AI is making us intellectually lazy, but that really depends on how you use it.

• Vasant Dhar, author of the new book, "Thinking With Machines: The Brave New World of AI," tells Axios that humans who stay "AI-proof" tend to share three traits: grounded expertise in a specific field, insatiable curiosity and a habit of asking lots of questions.

The bottom line: No one knows what's next — and the AI bubble hype can't be discounted — but on ChatGPT's third birthday, significant signs point to humans, agents and bots marching together into the uncertain future.

AI is upending everything, including retirement planning.

The big picture: Our comfortable golden years have always rested on assumptions of a steady full-time career, employer-sponsored savings and a predictable job exit. Now those assumptions are fraying.

Zoom in: AI is already reshaping hiring and workforce strategies for big and small businesses alike.

• Gig and temporary work are flourishing in the uncertainty, but that means limited access to retirement savings options and reduced future Social Security benefits.
• Previously well-paid office workers funded their retirement accounts by performing tasks that chatbots can now do just as well or better.

All the AI bubble anxiety isn't helping.

• Heavy exposure to AI stocks like Nvidia and the Mag 7 in 401(k)s is prompting warnings that a correction could hit retirement savings hard.

Reality check: AI can help with retirement planning too. But it's still prone to errors and bias.

• AI can democratize sophisticated planning and reduce the need for a high-paid planner. Investors at any stage of their career can use AI for scenario modeling, identify savings gaps, tax optimization, Roth conversions, and "what if" questions like, "What if I retire at 55 versus 65?"
• It can free them from busywork, allowing them to spend more time talking to clients about their retirement needs. However, experts warn that using it on your own can be risky.
• "Clients get really emotional about their money during volatile markets, and emotions often lead them to making bad decisions," Richard DellaRusso, wealth management managing director at UBS, told Axios. "We do a lot of handholding."
• AI is particularly unsuited for handling complex life events like divorce, inheritance or emotional questions about money. And be wary of turning over your personal and financial information to the bots, DellaRusso says.

Yes, but: AI has the ability to reduce emotional influences in decision-making, offering a more rational, data-based approach.

• For retirement planning, in particular, those who have the means might be better off combining AI with professional expertise.
• "Our clients like to talk to humans," DellaRusso says.

The holiday shopping wars have sparked an AI arms race — the retail industry's biggest technology experiment since the dawn of e-commerce.

Why it matters: Retailers and tech giants — from Target and Walmart to Amazon, Google and Meta — are rushing to deploy smart shopping assistants that promise to help consumers find gifts faster, spend smarter and even complete purchases.

The big picture: The same generative tech that powers chatbots is now shaping everything from product discovery to in-store navigation — testing whether consumers are ready to let algorithms do the gifting.

• It's turning 2025 into the first true AI holiday season, experts tell Axios.

Keep reading.

• Here's a highly technical look at Google's latest TPU — the chip giving Nvidia a run for its money. (SemiAnalysis)
• Alibaba's latest Qwen model can spot a "needle in haystack," finding a single detail in long videos and reportedly outperforming recent models from Google and OpenAI at some tasks. (Decoder)
• DeepSeek released an open model that it says can deliver gold medal-level results on key math tests. (South China Morning Post)

It turns out that crows can really hold a grudge. (h/t Stefan Jon Silverman)