The basis remains the bi-weekly LinkedIn listing of Tom Van ‘t veld, Learning Innovator at OASE (powered by Xylos). Tom closely follows AI developments, and we translate his observations into what they concretely mean for organizations and the people working in them. Welcome to edition four.
EDITION 4 – MAY 26, 2026
Here are five stories from the past two weeks that are most beneficial to your organization.
Google I/O 2026: new models, a rebuilt search engine and agents that continue to work for you
Google dominated AI news over the past two weeks with its annual I/O conference. On the model front, two major newcomers landed: Gemini 3.5 Flash is faster and stronger than its predecessor, with added focus on programming tasks, and GeminiOmni is the first model to combine text, image, audio and video within the same system, on both the input and output sides. The video version is already live in the Gemini app and YouTube Shorts; a more advanced version for filmmakers will follow later.
The biggest shift is in Google Search itself. Google itself calls it “the biggest change to the search bar in 25 years”: the input field now accepts long, conversational questions like you would ask in ChatGPT. On top of that, the new Information Agents search on for you in the background, think ticket prices or sports results. Search sometimes even builds little interactive mini-apps on the fly based on your question. AI Mode, by the way, already reached one billion monthly users this year.
Google is also emphatically pushing Gemini toward agent work. Gemini Spark is a personal assistant that picks up tasks for you, such as following up on emails, updating your study schedule or scanning your credit card statement for unnecessary subscriptions. In Gmail and Docs, it comes on top of a conversation mode where you can brainstorm aloud using your own documents and emails as context.
For organizations, this also shifts search behavior, and thus the way customers find you. Time cites research showing that 26% of users close a session completely as soon as an AI summary appears at the top. For media companies and web shops, this is an important signal: valuable visitor traffic is tilting. For B2B organizations, it changes your website’s content strategy, as AI models summarize what they find there. Invest in clear, authoritative content that holds up even in a summary.
Microsoft listens to feedback, Apple chooses multi-model route
Microsoft updates Copilot based on user feedback. The floating Copilot button in Word, Excel and PowerPoint could cover formatting options or cells in certain scenarios. Microsoft is now rolling out a tweak that lets you hide or reposition the button in the Ribbon. In Edge, the separate Copilot mode will be integrated into the browser itself, making the user experience simpler. For organizations that guide their employees to work with Copilot, these UX customizations reduce real-world friction.
At Apple, the strategy is visibly widening. iOS 27 gets Claude and Gemini built in as alternatives within Apple Intelligence in addition to ChatGPT. A new menu lets you choose which assistant Siri passes a question to. Apple is also working on a screenless AI pin that cleverly leans on the existing ecosystem: audio via the AirPods, visuals on the Watch and the heavy computing on the iPhone. A design much more likely to succeed than the stranded Human Pin.
The common thread: the end user once again gains more control over which model does what. For IT managers, this brings both opportunities and headaches. The choice of which model to run for which use case becomes more important, and agreements about data, governance and compliance must evolve accordingly. Those who lay the groundwork here today with clear guidelines and a well thought-out model mix will later avoid the proliferation of uncontrolled AI choices within the organization.
Agent sprawl and the silent mistakes of AI agents
In the workplace, AI assistants are shifting from pilot to daily reality, and that reality is proving messier than the marketing flyers suggest. The Wall Street Journal documents a new phenomenon called “agent sprawl”: everyone in an organization is building their own AI helper, and some companies, meanwhile, are paying more for AI use than for the employees the agents were supposed to relieve.
Microsoft itself adds sobering research. On longer tasks, current AI models make rare but severe errors that corrupt documents undetected. Remarkably, agents that are allowed to invoke external tools perform worse than those that cannot. This reinforces the importance of a structured approach to AI coding tools. We previously wrote about vibe engineering as a framework for keeping these risks manageable.
For Belgian organizations, governance around AI agents is now really coming to the fore. A workable framework consists of three layers: a central overview of which agents are running in your organization, clear rules about which tasks an agent may perform independently versus with human review, and a recurring review cadence where you periodically check the output of longer tasks. Without that structure, the productivity benefit disappears back into correcting silent mistakes.
Cybersecurity: AI attacks, AI defends, and bug bounties still exist
The security story surrounding Anthropic’s Mythos continues to crack under its own hype. The lead developer of curl, a widely used Internet tool, tested the model and got five alleged vulnerabilities reported, only one of which was actually true. At the same time, Anthropic is positioning Mythos as the future of AI cybersecurity, while the company itself is launching a classic bug-bounty program in which human researchers report vulnerabilities in its products.
On the other side of the battlefield, the Google security team intercepted for the first time an attack written with AI help: a script that bypasses two-factor authentication on a popular admin tool. The AI origin betrayed itself in detailed explanations in the code and a self-created severity score, among other things. OpenAI meanwhile responded with a more widely available cybersecurity version of GPT-5.5.
The message for those responsible for IT security: the cat-and-mouse game between attackers and defenders is moving up a phase. AI-generated attacks are becoming more prevalent, and your organization’s detection capabilities must evolve with them. A thoughtful mix of technology, awareness and trained expertise continues to make the difference between a notification that gets noticed in time and an incident that stays under the radar for days.
Confidence in AI content takes a hit, AI literacy becomes more urgent
The pace at which AI output is flooding the Internet is accelerating the discussion of reliability. Consulting firm EY had to retract its own report after 16 of the 27 sources cited turned out to be fabricated or incorrect. Dutch web shops Bol and Standaard Boekhandel were caught selling hundreds of AI-generated books without clear labeling. And researchers warn of a form of AI cannibalism in which models trained on the output of other AIs gradually degenerate. Tom predicted that phenomenon more than a year ago.
At the same time, the countermovement is growing. American students are beginning to boo speakers who extol AI en masse, while top executives of AI companies are publicly expressing surprise at the continued dislike. The gap between enthusiasts and critics is widening.
For organizations, this entails a dual task. On the input side, it requires tight agreements about which sources to trust for which decisions, how to double-check AI output, and when human review is mandatory. On the competency side, AI literacy becomes a basic skill at every level of the organization. Training that teaches employees how to prompt, review and read critically remains a key link in any serious AI approach. That’s precisely where the overlap is with what Tom and his colleagues at OASE and Xylos Learning build every day: AI literacy that scales with the tooling you bring in.
We will be back in two weeks with the next edition.
About Tom Van ‘t veld
Tom has been working for Xylos for years, where he started as a Microsoft Office trainer and grew into the driving force behind innovative learning concepts. He is at the basis of OASE, Xylos’ online learning platform, and developed, among other things, the Digital Coach concept, a Microsoft Teams Escape Room app and the mAindset game that allows employees to playfully learn to promote AI. As a Learning Innovator, he has been increasingly focusing his gaze in recent years on what AI means for the way we learn and work. Respond or chat on? Find him at LinkedIn.
