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 two.
EDITION 1 – APRIL 29, 2026
Here are five stories from the past two weeks that are most beneficial to your organization.
Model race, quality and the productivity paradox
OpenAI launched GPT-5.5 as “one step closer to a super app,” with greatly improved agentic coding and computing, immediately positioning Codex as a broader work agent that can also work in your browser, Sheets and Slides. Anthropic simultaneously announced Claude Opus 4.7 and a few days later came DeepSeek-V4 out, with performance close to Opus 4.7 and GPT-5.5 but at one-sixth the price. Impressive, on paper.
Practice tells a different story. ChatGPT was down again, users are complaining about performance degradation in Claude, Opus 4.7 was labeled an overzealous query cop, and Anthropic itself published a mea culpa about recent Claude Code quality issues. The race to ever more powerful models seems so intense that somewhere along the line the foundations are beginning to crack.
Microsoft, meanwhile, made the Copilot’s agentic functions in Word, Excel and PowerPoint generally available, with the assistant also able to handle longer autonomous tasks in Excel and Word. An important step that further accelerates work in Office, though PCWorld rightly describes the assistant as an enthusiastic but inexperienced trainee: readily employable for rough work, provided you look critically at the output. This is consistent with what research among CIOs finds: about 40% of productivity gains from AI are lost again to fixing mistakes. A separate analysis draws an important conclusion from this: companies that opt for AI augmentation rather than full automation reap greater long-term benefits .. For those embarking on AI projects, this is an important guide: invest as much in skills and governance as in tooling.
Claude Design: AI co-designs
Anthropic launched Claude Design, a tool that uses natural language to put together UI mockups, slides and infographics in no time. For those who want to design a first version of an interface, a pitch deck or a report without immediately hiring a designer or an external agency, this is a remarkable step forward. It doesn’t get up to speed right away, though: users report that they quickly hit their usage limits.
On the other side, OpenAI brought a revamped GPT-Image-2 which can also pull info from the Web, though the generated layouts were judged superficially compelling but typographically and editorially unusable. For those who want to keep track of what this means for creative workflows, offers this analysis on what GPT-Image-2 is really changing offers a good start. The message for marketing, communications and product teams is the same as in the previous section: AI speeds up the design process, but the finishing touches and editorial check are better left to a human.
The Mythos saga: how to learn to read an AI announcement
In early April, Anthropic announced with much fanfare that its new Mythos model managed to run a full cyber-attack simulation for the first time, including finding weaknesses and building working exploits. In fact, the company deemed Mythos too dangerous for public release and, through Project Glasswing, released it only to a select club of partners. An announcement that landed squarely in the headlines.
Only the hype immediately began to show cracks. Security researchers especially found a “nothingburger: most findings were already feasible even with Opus 4.6 or existing open-source tools. Ethical hacker Inti De Ceukelaire built live a working phishing tool with Claude in a few minutes. The “too dangerous to disclose” model also proved accessible to those who could simply guess the URL.
The political story was also less pure than it first appeared. Despite the Pentagon’s earlier “blacklisting” statement, the the NSA was deploying Mythos via a separate route anyway, while the U.S. cybersecurity agency CISA was denied access. When, a week later, OpenAI GPT-5.4-Cyber released, clearly positioned as a response, the tone was set. Mashable therefore asks the pertinent question of whether Mythos is a PR stunt rather than a real breakthrough.
For organizations having to assess AI claims, this is a learning pattern: spectacular announcements rarely get the immediate weight they claim in press releases. The proper reflex remains to look at independent research, security specialists and the impact on your own risk profile.
Europe opts for sovereign alternatives
The tone in Europe is noticeably shifting toward digital sovereignty. Canada’s Cohere is acquiring Germany’s Aleph Alpha to build a transatlantic AI powerhouse, explicitly focused on sovereign AI for European governments and businesses. With support from the German and Canadian governments, and an investment by Lidl owner Schwarz Group, the combination is aiming for a valuation of about $20 billion.
Along the same lines, the Dutch government is pushing a proprietary, self-hosted alternative to GitHub based on the open-source Forgejo. The reasoning is always the same: less dependence on U.S. Big Tech for critical infrastructure. For Belgian organizations overhauling their cloud and AI stack, that means a growing ecosystem of alternatives, and the need to make conscious choices about where data, models and code reside.
Man remains the quality control
AI-generated content is flooding the Internet at an astonishing rate. 44% of all new music on streaming service Deezer was created by AI and China’s iQiyi (often called the “Netflix of China”) is converting its entire platform to mostly AI-generated movies and series. At the same time, more attention is being paid to provenance: VRT NWS will from now on show an AI check for content that may have been generated by AI. Rightly so, because new research shows that people barely manage to detect when their WhatsApp or e-mail messages have been composed by AI.
Two studies also painfully expose the limits of AI as a source of advice. AI models are wrong in primary care patient diagnosis in more than 80% of cases, and popular chatbots regularly give incorrect medical advice without users realizing it. When researchers posed to Grok as psychotic, the chatbot actively fed their delusions instead of referring them to help.
At the same time, the backlash is coming from surprising quarters. Indeed, the strongest backlash appears to be coming from Gen Z, the generation that grew up with the technology. In the same vein, The Verge argues that Silicon Valley is blinded by AI features that the average user finds irrelevant or even annoying. And the earlier prediction that AI agents would roam en masse as virtual employees by 2025 appears to be exaggerated for now, although a Dutch advertising agency is openly scrapping jobs due to AI tools and LinkedIn launched an AI job market for experts.
The common thread is clear: AI does not exempt us from critical thinking. Therefore, training employees in prompting, assessment and critical reading remains an indispensable part of any AI strategy.
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.
