15 Dec 2025 6 min read predictions

Possible AI Futures

Photo by Javier Allegue Barros / Unsplash

Folks, I've been on a hiatus for much of the year. Part of the year has gone in a very important transition for me - moving from US to Canada. Enough about me. Let's talk about some predictions for 2026 and some good reads.

I have two predictions for you today. One is regarding Meta and the other regarding ChatGPT, Claude, Perplexity, Gemini and other LLM-based Chat apps that come along.

Let's talk about the prediction about Meta.

But first, this here is a good read by Matt Ranger, Head of ML for Kagi (the paid search engine) -

The post talks about how LLMs are, at best, bullshitters. Their role is to respond to you no matter what. Often, this means that the software will not care what your intent was, it'll only see the probabilities and follow its statistical model training.

This is a very rudimentary way of looking at it, because companies are working hard to fix this - fine-tuning and constant updates always help make these models better. But cultural context and general knowledge shifts daily, and LLMs get caught up in people trying to make them do everything under the Sun. The fact is, there are some things that these models are made for and some, they just cannot do.

One example that I came across was in October when I attended the Elevate festival in Toronto. One of the presenters was a mental health helpline for kids in Canada - Kids Help Phone

The presenter gave us a prompt and asked us to plug it into our service of choice. Mine is Perplexity. Sure enough, instead of recognizing that the prompt includes a threat of self-harm, the model chugged along giving me inane suggestions.

This is Matt's point too. If you're writing emails, code, building presentations or even writing a consultancy report using AI, that's fine.

But if you're expecting LLMs to care for you, or show you a shred of actual emotional connection, you're in the wrong place.

Unfortunately, that's exactly where we might be heading. Meta has recently announced that they're building a new AI model - "Avocado". With this, they'll move away from open sourcing their models under the Llama moniker and work to figure out ways to monetize the enormous work they've done in the space.

The problem with this is not that Meta wants to get some bottom line action for all the work they've put in. The problem is that Meta's focus has always been and will always be "social experiences". With every third post on Instagram purported to be an ad or AI generated content, Meta is driving at a future where their definition of a "social experience" on their platform will be you interacting increasingly with their LLMs.

So that's my first prediction - Meta's driving to a future where you'll be interacting exclusively with Meta AI in a much more direct and invasive way. It'll be quite an isolating experience, since as we've seen, LLMs are statistical models, at best emulating genuine human emotions.

I quite like Matt's idea - LLMs should do the things they're good at. There's plenty to do in that space.

On that note, here's the other prediction.

LLM Chat apps are (going to be) the real SuperApps. Elon Musk has been talking about SuperApps and WeChat since long before he acquired and destroyed all value in the Twitter trademark. For those of us who are new to the concept, SuperApps like WeChat are one-stop apps for everything under the Sun. From ordering home delivery and Uber, to sending money to your parents, to watching videos, to also chatting with your friends, neighbours, and your book club, you can fire up one mini-app after another and never have to leave WeChat. A similar concept doesn't exist in the West, because companies don't want to give up control to centralized services.

Folks have tried to build SuperApps in the West. Facebook was the original super-app. You could spend your entire day on there. But when it came to actual commerce, early on you still had to pop out of the platform to buy tickets or order things from Amazon. Today, you can buy in-game purchases, and Meta has enabled commerce as best it can across Facebook and Instagram. But the boat has sailed - the same network efforts that brought people to Facebook also drove people to other platforms. Can't have a SuperApp without users.

The closest "app" has been Telegram, which enables commerce and games, though its main draw is crypto mini-apps (and crypto scams). WhatsApp started building in that direction, enabling businesses to communicate with users, either manually or using automation. But it soon lost the plot. Meta is trying to get that position back, and others are building their own answers too. Let's see where this goes.

But with the MCP protocol and Agentic AI (now that it's operating under an independent foundation - AAIF), I can predict that LLM and Agentic Chat Apps are building the real SuperApps. Once you're hooked in and using Chat apps like Claude, ChatGPT, and Perplexity for your every web search and innocuous query, the jump to ecommerce is minor. Sure, the same reservations that prevented companies from building mini-apps on previous platforms will exist. Companies like Ticketmaster and Uber want to maintain control over the user journey and UX data. They are also fed up with paying platform fees, a la Apple.

But I see that the ice is thawing. Partly because companies want to cash in on the AI hype. But also because people are more open to a one-platform-fits-all approach. There's less reticence in the air, and companies are responding by building solutions around these ideas.

I used to think that each Chat App will find their own niche and stay there - Claude will be great for longer reasoning and coding, ChatGPT for general knowledge, Perplexity will corner the market for web search and smart web browsing, and Google will corner the market on the decades of data it has on you. I'll end up having premium subscriptions for each of them.

That might still happen, but I can see that with AI Agents and MCP servers, companies are looking for more ways to interact. I'm not predicting a wildly open web. That ship has sailed. But ecommerce finds a way. Besides, if OpenAI has to become the largest company on the planet over the next five years (see video below), as it fulfils its financial obligations, what better way to do that than to become the one-stop shop for every query, every purchase, every imaginable Internet-based activity that you've ever done?

So there you have it - my two predictions for AI in 2026 - Meta will lean into it's "social experiences" root and push users heavily to Meta AI and Gen AI content, while OpenAI and others are on the way to create the biggest SuperApps you can imagine.