As 2024 came to a close, I noted here that two big stories were beginning to crowd out everything else in tech: the rapid development and diffusion of artificial intelligence, and the shifting policies of tech giants as they prepared for life under a re-elected President Trump.

Twelve months later, those stories did indeed define the year here at Platformer. On the product side, this year saw the first consumer agents, deep research, Google’s AI mode, OpenAI’s hardware ambitions, Sora, and the Atlas browser, among other key developments. 

Meanwhile, AI policy got both looser and more restrictive. Frontier AI labs eagerly made deals with the US military, reversing long-held policies against building weapons of war, and began leaning into adult content, from erotica in ChatGPT to Grok’s sexbot companion. On the other hand, amid rising evidence that chatbots were fueling a new mental health crisis, AI companies placed new restrictions on teen use and added parental controls.

All that took place against the backdrop of the new Trump administration, whose impact on the tech world was felt almost immediately. The year began with Meta’s surrender to the right on speech issues, a move that included changing its policies to allow for more dehumanizing speech against minority groups. It also killed its DEI program, a move followed by many of its peers, and shut down systems that once prevented the spread of misinformation.

With DOGE, Elon Musk ran the same playbook for cost-cutting in the federal government that he had done previously at Twitter, to devastating effect. The new administration’s embrace of the tech right showed up quickly in its policy proposals, including most notably in its accelerationist position toward AI.

By mid-year, Musk and Trump split. But the broader relationship between Trump and Silicon Valley remained mostly positive — particularly for Meta — despite the fact that the government continued to pursue and antitrust cases against both that company and Google. (Meta won its case; Google lost.)

I spent much of the year feeling increasingly disillusioned by the platforms’ cynical embrace of Trump and how little it seemed to cost them in users or revenue. (Even Tesla stock, which plunged in the wake of DOGE’s outrages, is now up almost 10 percent year over year.) I noted in particular the disquieting silence from trust and safety executives, who once spoke up proudly for human rights but understandably went quiet amid death threats, job insecurity, and threats to their immigration status. Platformer’s traditional focus on issues related to content moderation came to feel somewhat quaint in a world where principled leaders had been largely replaced by Trump appeasers.

And so I found fun where I could: an unusual and revealing dinner with Sam Altman; a speculative preview of how Substack might change to live up to its lofty new valuation; and some hard-won lessons about productivity. 

And I found myself relieved that, at long last, more countries had begun to take child safety on platforms more seriously. Australia’s move to ban children under 16 from having social media accounts, despite the real harms to free expression, still strikes me as inevitable given platforms’ inability to protect their users from child predators, AI suicide coaches, sextortion, and other catastrophes.

Through it all, you all have kept me going with kind emails, thoughtful comments, and the occasional angry (but welcome) direct message on Signal. Thanks to all of you for reading Platformer over the past year, especially the paying subscribers who allow me to do this for a living and to employ our small but mighty team. (Consider becoming one of them, if you haven’t already?) 

This year we’ve made some changes to our format that you all have responded very positively to. Next year we’ll begin a new experiment in audio. Five years into this experiment, I remain more grateful than ever that I get to wake up and help make Platformer for you. And as bleak as the news cycle often seems, I remain excited to see what 2026 will bring. 

See you here on January 5.

As always, in the name of pundit accountability, I begin by checking in on my 2025 predictions. 

What I got right

The AI culture war begins. From executive orders banning “woke AI” to one attorney general’s suggestion that it should be illegal for ChatGPT to criticize President Trump, this one played out exactly as I imagined it would. 

The child safety fight comes to chatbots as well. “With more teens using chatbots — and some of them having tragic outcomes — look for there to be more pressure on AI labs to add restrictions, warning labels, and safety features for their under-18 users,” I wrote. And they did.

Google, Apple, or both will begin to offer age verification at the device level. Google did, at least in some states. Apple did, too.

The big AI companies remain competitive with each other. “Expect regular changes at the top of the Chatbot Arena leaderboard,” I wrote. And there were — although some of that was because Meta was cheating.

Meta has a Llama scandal. Did I mention that Meta was cheating?

Everyone copies Meta Ray-Bans. Google and Samsung announced Android XR glasses; Apple and Snap also reportedly have glasses of their own on their way or in development.

The flop of “Apple Intelligence” results in the company scrapping its existing strategy. Yup and yup.

“Remarkable Alexa” underwhelms. Yup.

Google loses its ad tech antitrust trial. Yup.

Mixed results

The TikTok ban goes into effect. It did go into effect … for about a day. I successfully predicted that the Supreme Court would side with Congress. I did not predict that Trump would ignore both other branches of government and extend the ban indefinitely. (News of a sale agreement was reported today, nearly a year after the original deadline.) 

AI will show continuous incremental improvement, but no exponential leaps. Depends on how you define “exponential,” really. In my own work, I found large language models to be significantly more capable than they were a year ago. Their ability to draft research briefs, write thoughtful interview questions, fact-check and copyedit dramatically increased over the past year. Software engineers reported seeing even larger productivity gains. For most people, though, AI still hasn’t changed their job all that much. 

The first year of "the agentic era" mostly disappoints. Certainly it underperformed relative to the hype. Consumer agents from OpenAI, Anthropic and Google struck me as something closer to a party trick than a useful tool — and the party trick typically took 20 to 30 minutes to complete. That said, enterprise agents did have some notable successes — witness the $100 million in annual recurring revenue for the B2B agent company Sierra, for example, or Claude Code hitting $1 billion in ARR after just six months of existence. (See my ethics disclosure!)

Bluesky and Threads remain in a productive rivalry. For the most part, I’m not sure either platform really even thinks of the other as a rival. Each has developed into its own thing, neither of which has the heat or relevance of Twitter in its heyday. (Nor does X, for that matter, no matter what the dead-enders there still insist.)

What I got wrong

An activist investor group demands changes at Snap. They simply did not. 

The bro-ligarchy collapses, as Trump comes to resent being seen as a tool of Silicon Valley. The thing I got the most wrong this year. On one hand, yes, Trump and Musk had their predictable blow-up. Aside from that one relationship, though, the tech right and Trump are still painfully close. Just this month, the Andreessen Horowitz wing of the Republican party pushed through an executive order that seeks to ban states from passing most laws seeking to regulate AI. This is despite the fact that the ban is hugely unpopular with scores of elected Republicans. Perhaps Trump will resent being seen as a tool of Silicon Valley someday, but it did not happen in 2025.

Predictions for 2026

With that out of the way, here are some predictions for next year.

The AI bubble won’t burst. The bubble talk that dominated tech discourse in 2025 will continue all year long. We shouldn’t be surprised if there are some spectacular flameouts next year, particularly if models from the frontier labs improve enough that businesses can deploy them directly and not pay $50 per user per seat per month for some AI startup’s product. And some public companies will undoubtedly lose tens of billions of dollars trying to catch up to the leaders. (The odds of Meta catching up are long, and growing longer.) But I predict that 2026 ends more or less as 2025 did, with OpenAI, Google, Anthropic, and mayyyyybe xAI jockeying for prime position. As the analyst Benedict Evans recently wrote: “The fact that AI is working really, really well, and has no signs of hitting a wall, and will change the world… Does NOT mean that there cannot also be a bubble in AI. In fact, that’s generally the kind of thing that *causes* bubbles.” If you want to prepare yourself for 2026, internalize that point.

AI will have a dramatic impact on software engineering in 2026, and a much less dramatic impact on other jobs. At current rates of improvement, benchmarks that measure the ability of LLMs to write code should be almost entirely saturated by the end of 2026. That should lead to reduced hiring rates for software engineers, rapidly changing job descriptions for those who remain, and perhaps even the beginnings of large-scale layoffs. Outside of coding, though, expect another year full of Nano Banana-scale improvements: moments where the technology becomes obviously more powerful than its predecessor, opening up new possibilities for work and entertainment, but in ways that augment what humans can do rather than replace it.

AI talk won’t dominate the midterms. Some pundits predict that a massive anti-AI movement will coalesce by November, delivering huge gains to whichever party more convincingly adopts a tech-hostile posture. I predict AI concerns will take a backseat to the economy and a backlash to Trump’s authoritarianism. That said, I do think we’ll see the beginnings of an anti-AI movement — and perhaps a bipartisan one. Look for at least three Democrats to win House seats promising to stop the construction of data centers in their districts, or creating new protections for people who lose their jobs to automation.

AI companions will become a cultural flashpoint. In 2026, more Americans will turn to AI companions for companionship, sex, and love — and exit the traditional dating market altogether. Platforms will become more skilled at extracting money and attention from this cohort, leading to a societal schism that results in warnings from religious leaders, Congressional hearings, and several painful essays in The Atlantic. 

An LLM-powered cyberattack brings AI safety fears to the forefront. Leading frontier models already show significant promise in finding exploits, and only the labs’ guardrails prevent them from being used for crime. In 2026, look for a nation-state or group of cybercriminals to weaponize an LLM to cause significant damage, and give AI safety advocates the concrete evidence they’ve been looking for to push for stronger regulations on AI development. 

Meta’s AI reboot gets rebooted. About half a year into Meta’s organizational reset, stories are already leaking out about frustrations between executives and challenges in competing with the top labs. By the end of 2026, expect Meta’s models to still fall short of state-of-the-art — and Mark Zuckerberg to shuffle his executive lineup again in response.

OpenAI retires Sora. A year from now, I suspect that we will look back on Sora as a moment when OpenAI was at its most unfocused. At a time when Google and Anthropic were wowing customers with advancements in AI code, OpenAI was launching a TikTok-like social network that immediately came under fire for its lack of copyright guardrails and vulgar embrace of brainrot. While it was a hit on release, Sora usage has tapered off sharply since then. Assuming Sam Altman is serious about diverting more resources to ChatGPT, he’ll wind down Sora development early next year and bring those creative tools and social features to the place where he’s already winning: ChatGPT.

Social media bans for children under 16 will become the norm. After more than a decade of parents demanding stronger platform protections and mostly disappointing results, this year Australia said “the hell with it” and moved to delay teens’ exposure to social media until the middle of high school. Amid rising evidence that smartphones in schools are bad for kids in general, and regular exposés documenting the way platforms have sought to bury their own research into child safety issues, expect other countries (and US states) to follow suit. 

A judge rules that Google must spin out its ad exchange. The government handily won its case that Googl