The Invisible Interface

Published 20 November 2025

The Invisible Interface

We have spent the last three years staring at text boxes. But if last week's teasers are anything to go by, the Chat phase of AI is ending. The Ambient phase is beginning.

OpenAI and Jony Ive finally broke their silence on their hardware collaboration. The details are fascinatingly vague. They describe a device that is screenless and naive in its simplicity.

The pitch is that smartphones have become noisy. This new device is meant to filter that noise. It knows what you want before you ask. It disappears.

But the most telling detail was not about the specs. It was the admission that they knew they had nailed the prototype when there was an urge to pick up that thing and take a bite out of it.

This sounds odd, but it signals a massive shift. We are moving from utility where AI is a tool you use, to intimacy where AI is an object you live with.

It is not just OpenAI. The market is suddenly flooded with AI wearables trying to find a purpose. We are seeing everything from pendants to glasses. Most will fail. But the direction of travel is clear. The goal is to remove the friction of the screen entirely.

For us, this means the skill set is changing again. Prompt engineering is about talking to a machine. The next phase will be about behaving with one. If the interface becomes invisible, our interactions become behavioural rather than textual.

I am skeptical about carrying another device. My pockets are full enough. But if they can actually deliver a calm computer that works as well as a smartphone without the dopamine addiction, I might just be tempted to take a bite myself.

News Roundup

Google Takes the Performance Crown

Analysts report that Google has officially taken the lead with Gemini 3 Pro. It is currently outperforming competitors on major benchmarks. It feels like the lead changes weekly. But for now, Google is sitting pretty.

OpenAI Fixes the Em-Dash

In a move that feels personally targeted at me, OpenAI claims to have fixed ChatGPT's em-dash problem. The model used to overuse the long dash. Now it speaks more naturally. I am strictly forbidden from using that punctuation mark in this newsletter anyway. I feel no safer.

Insurers Panic Over AI Risk

A new report suggests AI is becoming too risky to insure due to hallucinations and copyright claims. Without liability coverage, enterprise adoption could hit a massive and boring wall.

Anthropic Releases Prompt Library

If you struggle to get Claude to behave, Anthropic has released a searchable library of practical use cases. It is a helpful resource for coding and work tasks.

European Schools Split on AI

There is a growing divide in Europe. Some nations are rushing to integrate AI into schools while others are banning it. It is a messy and real-time experiment with children's education.

Chatbot Privacy Warning

New research on AI companions shows they are a privacy minefield. They collect intimate data that companies cannot fully protect. Be careful what you tell your digital friend.

Search Risks for Business

Companies are realising that AI web search features can ingest and regurgitate inaccurate data. This creates new liability nightmares for enterprise data accuracy.

IBM Identifies Data Silos

A new report from IBM suggests that old-school data silos are the biggest factor holding back enterprise AI. You can have the best model in the world. But it needs access to your data to work.

Designing for Digital Resilience

MIT Technology Review argues that we need to design digital resilience into the era of agentic AI. We must ensure systems can recover when autonomous agents inevitably mess up.

Inference at Scale

Companies are finally moving from training to inference. They are focusing on how to run these models at scale in production environments without bankrupting themselves.

The Gemini 3 Upgrade

We all know Nano Banana Pro. It has been around for a while. But until this week, it suffered from the same problem as every other image generator. It was great for art but useless for information. The new Gemini 3 integration changes that.

Why It Matters

Gemini 3 brings actual reasoning to the image generation process. It understands logic. This solves the text-on-image problem that has plagued us for years. Previous models guessed at letters. This model understands words.

My Experience

This update is a life saver for my workflow. I have been creating complicated diagrams for work. For the first time, the AI is not hallucinating the text. The labels are legible and correct on the first try. It handles complex spatial instructions (like place label X to the left of Y) because it actually understands the concept of left.

The Verdict

This upgrade moves Nano Banana from a creative toy to a productivity tool. If you need cinematic vibes, stick with Midjourney. But if you need a diagram, a poster or anything with words, this is now the only serious option. It finally works. A good example is the header image for this very newsletter!

The Invisible Interface