In 2026, AI will undoubtedly continue to be a huge part of our social, economic, technological, and political progression as a species. Here are a few things I think we’ll see in 2026 in the world of AI and elsewhere.
Autonomous Driving Accelerates and Runs Into the Real World
Waymo is planning to aggressively expand its areas of services throughout 2026 and beyond. Buoyed by impressive safety data and growing (albeit slowly) acceptance of autonomous driving, they will continue to increase their YoY miles logged and cover more geographic areas than they previously have. I’m very bullish on autonomous driving and think ideally designed and deployed autonomous driving systems will be widely available to riders in the coming decades and will dramatically reduce (>95% reduction) road fatalities once we hit the point at which a human is no longer necessary to drive the vast amount of miles logged in this country.
However, my prediction for 2026, is that Waymo will see its first rider fatality. My hunch is that it will actually be the fault of the human driver in the other vehicle in the accident (i.e. a blind merge/lane change on a freeway), but it will happen nonetheless. At some point, the law of averages dictates that it must. I hope that I’m wrong, but we know that Waymo has been preparing for that moment whenever that first fatality does happen.
What will the public think when this accident happens? Judging by other cases where fatalities (both human and feline) have involved autonomous technology, they will receive outsized media attention compared to a run-of-the-mill road fatality. Safety stats make for a boring story – a first of its flavor tragedy does not. Autonomous driving, like many other new technologies, will have skeptics that point at edge cases encountered in the real world, like when Waymos were more or less rendered inert during a PG&E outage in San Francisco in late 2025. These considerations are important – the world is messy! It’s full of edge cases. But they should be weighed appropriately against the preponderance of evidence of current road experience (excessive speeding, impaired driving, texting and driving, sexual assault by Uber drivers) to adequately assess the pros and cons of bringing a new technology into the real world.
Chatbot Wars Intensify
OpenAI’s ChatGPT is still by far the dominant market leader in consumer AI applications. Due to many factors, including explosive early adoption, rapid early iteration, and the very public personas of OpenAI’s leadership, ChatGPT is the “Kleenex” of AI chatbot applications today. Gemini will continue to increase its marketshare, as Google’s renewed and intense focus on consumer facing products and AI technology was an important 2025 trend. Expect lots of Super Bowl ads for AI again, really leaning in on the chatbot experience and form factor. It is still relatively early in the chatbot game, and I think that other than for superusers, the switching costs are still low. If Google can show that Gemini’s capabilities are on par with ChatGPTs and thread the needle of more seamlessly integrating into G-suite and Android phones (something I’m admittedly a complete novice regarding considering my pseudo-religious devotion to the Apple ecosystem) I think they have a real shot at taking a bite out of ChatGPT’s marketshare.
Apple Kills Siri – But “Phoenix” Rises from the Ashes
Apple has been one of the weirdest players in the AI space. Over the past several years they have overpromised on features that never shipped or made AI a footnote in keynote addresses. I don’t necessarily blame them for it – they have continued to perform well both in the consumer marketplace and in the stock market and probably actually have made a wise decision to let the other companies duke it out in the capabilities war. Apple as a company has long had a corporate philosophy of not necessarily doing it first, but doing it best. As someone who thinks a lot about the creative process of product design, the thing that is most important to me when considering creating something new is to truly understand what the problems are that your product is going to solve for. Right now I think it’s the lack of well-defined and well-specified problems that are keeping AI adoption relatively low when you consider the actual capabilities of these tools. My bet here is that Apple retires Siri, sending it respectfully afloat on a barge, down the digital river to join other storied relics of the past. They use their fall keynote to launch Phoenix, Apple’s Ambient Intelligence.
Ambient Intelligence is a Key Theme in Consumer AI
An ambient intelligence, one that lives with you, hears what you hear, sees what you see, and knows what you experience in the world, would be capable of solving the problems that you encounter on a daily basis. Imagine a personal Clippy that rides around on your shoulder, adding to your to-do lists, ordering Ubers when you have a reservation, surfacing that ticket in your email as you arrive at a concert. This kind of ambient intelligence will make your life more frictionless. This is my bet for the consumer product category in 2026. Whether it be pens, pendants, pucks, or phones – ambient intelligence is coming. A truly personal assistant that takes the annoying things off of your plate and positively adds to your life.
This vision is probably going to have a hard time being accepted outside of techo-optimist circles. There’s going to be a very thorny set of social norms to navigate in a world where omnipresent recording devices are now just not within phones, but in peripheral accessories as well. What does obtaining consent for recording look like? How is that data stored? What about separating personal and work data, much of which comes with far stricter rules and regulations? I think there’s going to be significant cultural and social pushback on further fortifying the distributed panopticon that our modern world already represent. Add to that the general public’s distaste for AI and it’s a receipt for backlash and ridicule.
Utility over Capability – Benchmarks Fall Out of Favor
LLMs are going to saturate most benchmarks by the end of 2026 and I think they will no longer be a great measure of an LLM based AI tool. LLMs have made remarkable progress on human created benchmarks to date and I think that progress will continue. We will see intelligence per dollar across these benchmarks continue to fall, as smaller models improve and better computing is available, but the world starts to shift in favor of evaluating “utility” over “capability”. These means a lot, and at some point will probably be its own blog post in the future, but mark it down here – increased focus on “what have you done for me lately” not “what could you do in a theoretically perfect scenario.”
White Collar Job Growth Will Be Non-Existent
Slightly related to the capability vs. utility distinction in the preceding prediction, I think we see a continued holding pattern in the white collar world. C-suites all over are ecstatic about the promise of AI tools, wooed by AI industry players taking showcasing capability and promising utility. Huge, multinational organizations are going to be very wary of bringing in new workers to onboard and train if there’s a potential to automate their tasks in the next few years. While many companies may be using AI as cover for layoffs that occur for other, more traditional business reasons, I think it’s likely that organizations generally will take the stance of trying to keep teams relatively lean as they wait for the AI aha moment in their specific industry/use case.
AI Will Be a Major Issue in the 2026 US Midterms
At a time where a historically unpopular president already appears to be in his lame duck era, and has gone all in to appease the AI industry, it’s very likely that Democrats take a decisively anti-AI stance. It’s going to be the politically popular thing to do. I want Democrats to win back power and position themselves well for 2028, so I’m not going to necessarily begrudge them for harnessing popular sentiment and pointing it at the current punching bag.
Good politics doesn’t always mean good policy. What I would like to see is this issue framed like is focusing on building technology the works for us and increase our ability as humans to thrive, not technology that simply increases the valuation of a handful of trillion dollar companies. That’s a paradigm shift though that’s complicated and requires a lot of focus and effort – not something we are going to want to tackle in the 11 short months before these interim elections.
There will be lots of angles from which AI enters the political debate this year – nonconsensual AI image generation, data center construction, electricity prices, job displacement are just a few. I don’t think the population is in a receptive mood to the idea that “if we just go all-in on AI, the future will better.” Technology has done an incredible amount of good for human civilization, but we find ourselves in a precarious world currently, where we are chipping away at the last few percentage points of tangible improvements to our lives that entities are financially incentivized to go after, while majors problems like hunger, disease, violence, and inequality remain.
I want us to choose to use AI to help us maximize human thriving. We should be skeptical that a technology that has already provided outsized benefits to such a small percentage of people and corporations will magically enable to solve all the issues we face on this planet. The upcoming midterms will be a preview of the 2028 election, where I believe AI will be the single biggest issue.
AI Achieves a Scientific Breakthrough
And for now, my most positive prediction: 2026 will be the year where AI for science goes mainstream within the scientific community and aids in/discovers a scientific breakthrough.
We are around the time when then capability level of tools combined with a greater acceptance within the scientific community of integrating these tools into their work may be capable of producing something genuinely novel. Perhaps it’s a novel drug discovery, a math conjecture, or a materials science problem – all areas of knowledge with huge search spaces that are better approached with the intellectual horsepower of AI systems.
I don’t think that the Nobel Prize will be awarded (at least in part) to an AI system quite yet (I think that will happen around 2028-2029), but I think scientists are really starting to understand the parts of the scientific process that are amenable to AI tool inclusion, and are going to be better able to deploy this familiarity with the tools in an experimental setting.
Predicting, Updating, Iterating
I am really excited to put out my first formal predictions post. I want to hear what you have to think about these predictions and some of your own predictions. Putting a marker down at a specific point in time feels a bit treacherous, but I’ll be happily updating and iterating on these predictions over the years. I’ll also revisit this in December 2026 for my first prediction self-assessment. Subscribe if you want to make sure you track these through to the end.
Updated January 10th, 2026 – Added in an article that posits that AI is being used as cover for layoffs rather than AI actually replacing human labor. Kudos to my good friend Dr. Paul Hook for sending along this piece.
Leave a Reply