CTNET Review of 2025 and Predictions for 2026

Annual Review / By /

In this post, I want to review what happened during 2025 and see how well my predictions fared. As an experiment, I asked Claude to summarise my post and suggest other highlights from 2025. It even made some mistakes and put forward its own suggestions.

I will then look to make some predictions for 2026.

How Did My 2025 Predictions Fare?

I’m not sure I should be publishing this – it puts more pressure on my predictions for 2026. But then again, the fun of making predictions is seeing how correct or wrong they turn out to be.

The fact is, I did quite well last year, according to the analysis carried out by Claude Sonnet, though I did have to correct it about Spotify offering lossless digital streams. This was probably due to the model’s training data cut-off being before this was announced.

That’s why you need to carefully consider any response you receive from generative AI, especially if you’re depending on it being factually correct.

Anyhow, let’s get back to reviewing 2025.

Prediction 1: Development of Personal AI Assistants Driven by LLMs

Verdict: ✅ Largely Correct

I predicted further development of AI in note-taking and knowledge management, specifically mentioning Google Notebook LM becoming an extremely useful learning assistant.

Google Notebook LM has indeed become an extremely useful learning device, especially with its new research tools.

I’ve also been able to start integrating Artificial Intelligence (AI) into my workflow, thanks to a combination of improvements to the Obsidian community plugin Obsidian Co-pilot and the potential for local-based clients to access my desktop.

At the start of the year, I had one question: “How can I use AI whilst protecting my cognitive abilities?” I think that question is still there, but I’ve added a more progressive one: “How can AI improve my cognitive abilities?”

I’ve published a number of blog posts on this exploration over the last 12 months. I’ll link to them at the end in the further reading section.

Prediction 2: More Vocal Concerns About AGI Development

Verdict: ✅ Correct

I predicted that more concerns would be raised surrounding the development of Artificial General Intelligence, and this proved accurate. However, the conversation shifted somewhat – the focus in 2025 has been less on theoretical AGI and more on immediate concerns: AI scheming, deception, job displacement, and the geopolitical AI race between the US and China.

The AI race between the US and China heated up at the start of the year with the launch of DeepSeek, a model that was challenging the top US models such as ChatGPT and Gemini at apparently lower cost.

The unemployment rate here in the UK reached just over 5%, and there does seem to be a debate on how much of an impact, if any, AI systems have had on this.

I would also add that there’s a debate on how AI will impact the cognitive development of young people, as it potentially takes away the friction of learning new skills.

I’m exploring the same question from the perspective of my knowledge system. AI, like most technologies, is a double-edged sword, and we must learn the best way to use this new technology.

Prediction 3: Spotify HiFi

Verdict: ✅ Finally!

I had this prediction partially as a joke, as we’d been expecting it for several years. To be fair, I would have been really surprised if it hadn’t happened, given that practically every other competitor now offered lossless streaming.

Thankfully, Spotify saw sense and rolled it out this autumn. You can read our review here.

A Review of 2025: Major Breakthroughs

Artificial Intelligence: The Year of the Agent

2025 has become the year that AI agents first appeared, marking the evolution of generative AI from reactive generation to autonomous tools that can carry out automated processes.

Agents first appeared in development tools, probably in part because AI developers have a good understanding of what other software developers want. Also, the end requirements of most software projects are often quite tightly defined.

I’ll be writing a full blog post, but I saw Claude Code spin up AI agents as it worked out how to create an A to Z index of links to my permanent notes. It was quite awe-inspiring.

A lot of this development is being driven by increased competition between OpenAI, Anthropic and Google DeepMind, which was disrupted even further by the arrival on the scene of Chinese start-up DeepSeek – delivering similar performance to other top models but with lower computing costs.

Quantum Computing: Developments Continue

This year, Google’s Willow chip achieved two major milestones. Firstly, it demonstrated that errors can be reduced exponentially as the system scales up using more qubits, reducing the impact of quantum error correction. Secondly, Willow performed a standard benchmark computation in under five minutes that would take today’s fastest supercomputers 10 septillion years.

Google estimates that real-world applications in medicine and materials science could arrive within five years.

An AI generated image of a man walking down a busy street seeing in 2026.

My Technology Predictions for 2026

Agent AI Goes Mainstream

Use of AI agents is likely to increase. As I mentioned earlier, Claude Code spun up agents to help solve a problem I’d given it. Each of these agents was another instance of Claude Code, spun up to help solve a particular task or subtask created from a task list that Claude had generated.

This is likely to spread to use by more casual users interacting with popular web-based models that will spin up agents if required to complete more complex tasks. In most cases, their use will be unknown to end users.

Others, like myself, will deliberately consider how their use can fit in with our own processes – what we can offload to them without having a negative impact on our own cognitive abilities.

Personal AI Assistants Get Long-Term Memory

One of the biggest problems with current Large Language Models is that they find it difficult to learn new things once their training is complete. An example of this occurred when I asked Claude to review last year’s post, and it wrongly stated that Spotify still didn’t support lossless streams.

This was almost definitely down to the fact that the data it was trained on had an earlier cut-off point, as it was quite a recent announcement within the last few months.

Traditionally, Large Language Models find it really difficult to learn new things without destroying what they’ve already learnt, unlike humans who can learn new skills without impacting skills we’ve already acquired.

Secondly, they only used to remember things within the context of your current exchange, and each conversation has a specific context window dependent on the model and your subscription for that model.

However, models are learning capabilities to remember things between conversations. You can ask them to add things to memory, and this trend is only going to grow as it will help the model become more helpful.

That’s why I’m looking to incorporate Large Language Model access to my Zettelkasten, as those notes will act as an external memory to my thoughts and ideas, giving any AI better context on what I’m trying to do.

Quantum Computing’s First Commercial Applications

Could the breakthroughs with Google’s Willow lead to the first commercial applications in quantum computing appearing?

The first applications would likely be modelling complex systems such as financial modelling or molecular simulation.

You have to wonder what impact quantum computing will have on artificial intelligence research.

Conclusion

2025 saw generative AI and the underlying Large Language Models continue to evolve, even if speculation suggests that the development of Large Language Models is starting to plateau. The impact on how we work and learn is only just beginning to be understood.

For me personally, 2025 was the year I began seriously integrating AI into my knowledge management workflow whilst maintaining the cognitive boundaries that make knowledge work valuable. It’s an ongoing experiment, and I’m excited to see where it leads in 2026.

You can follow my journey by signing up to my monthly newsletter.

Further Reading

Leave a Comment

Your email address will not be published. Required fields are marked *