AI News: The most important developments of the past week (July 13, 2026 – July 19, 2026)

Dear Community,

we receive news about new tools, security updates, and regulatory changes in the field of AI almost daily. For medium-sized businesses, it is often a challenge to keep track of it all and separate the wheat from the chaff.

This is precisely where we come in. As your bridge between science and practice, we filter the week's news and present you with the developments that are truly crucial for your competitiveness.

1. From notes to analysis: Google turns NotebookLM into Gemini Notebook

On 16 July, Google renamed its research tool NotebookLM to Gemini Notebook. What at first sounds like mere branding is more than that: along with the new name, every notebook gains a secured working environment in which the program can write and execute code itself.

The difference from before is easy to grasp. Until now, the tool could search your uploaded documents, summarise them and answer questions about them. Now it can also calculate with the same content – comparing values, forming totals or looking for anomalies in a table. The detour of preparing the figures manually before you can even ask a sensible question falls away.

For companies, this is above all interesting where the same documents are analysed regularly: quotation lists, project reports, complaints or meeting minutes. Anyone who has so far shied away from setting up a dedicated analysis project has a very low-effort entry point here. The feature is initially available to Google AI Ultra users and to Workspace customers with extended AI access, with further Pro users to follow in the coming weeks.

A note for use in the DACH region: in Workspace you can set data to be held in European data centres. This setting should be checked before sensitive documents are uploaded.

2. Less explaining twice: OpenAI expands ChatGPT's search and memory

On 14 and 15 July, OpenAI rolled out two changes to ChatGPT. These are not a new model generation but improvements to everyday handling – which are nonetheless clearly noticeable in daily work.

First, there is now a unified search across all your own chats, projects, images and documents, with filters by content type. If you know a particular formulation or analysis has already been produced, you do not have to create it again. Second, the space for persistent working instructions has been increased from 1,500 to 5,000 characters.

The second point in particular deserves a closer look. These instructions let you store what the program should know permanently about your company: how you address your customers, your technical terms, your product names, and also what must never appear in any text. Until now the space for this was so tight that most users re-explained the same context in every conversation. Set up properly once, this saves rework in the same places every week.

In addition, on 15 July OpenAI released app synchronisation for enterprise and edu environments with their own key management. All the changes mentioned are available globally and can therefore also be used in the DACH region.

3. The race is tightening: Google's Gemini 3.5 Pro is months behind

On 16 July it emerged that Google is months behind its own schedule with Gemini 3.5 Pro. At the developer conference in May, CEO Sundar Pichai had still named June as the date. For context: this is not about the notebook tool described above, which is already available, but about the most capable model in the Gemini range, which is still outstanding.

At the end of June, Google had reworked the training data in order to improve programming capabilities in particular, but fell short of its own expectations. The stock market reaction was clear: Alphabet shares closed 4.4 percent down, which corresponds to roughly 200 billion US dollars in market value.

The context is what makes this interesting. In the same month, two competitors delivered: OpenAI released GPT-5.6 on 9 July, and Anthropic made Claude Sonnet 5 the default model for all users at the beginning of July. Who leads the field is therefore less clear-cut than it was a few months ago.

For companies, one thing above all follows from this: it does not pay to bet on a particular provider. In everyday work, the gap between the leading models is smaller than the headlines suggest. It makes more sense to build applications so that the model behind them can be swapped out later, and to base a rollout on what is available and proven today.

Our conclusion: This week shows two movements on different levels. At the top, the field is tightening, and even a corporation like Google is postponing an announced date by months. At the same time, the tools your employees already use every day are becoming steadily more useful. For practical purposes, the second movement is the more important one: the progress that actually reaches you usually sits in the programs you have long had in house, not in the next model announcement.

As always: technology is the lever, but strategy decides success.

Fiona & Maureen | Tailor-made AI Consulting