AI News

A recent Stanford study found the more people use AI, the more convinced they it is saving them time… when the results show it isn't and the price of that illusion is about to get a lot more visible for organizations. The researchers tested nearly 2,700 people on exactly the kinds of tasks most Microsoft Copilot rollouts are built around. Summarization Rewriting Simple information retrieval Basic arithmetic Short drafting tasks Here's what the results found. --> People used AI far more than they predicted… 47% of the time versus the 33% they expected --> They believed AI would save roughly 56 seconds per task. It saved about 7 --> On the simplest tasks, AI actually made things slower. --> And after using AI, people became even more likely to use it again — and more confident it was creating value — regardless of whether it actually was. This isn’t efficiency. That's a belief of it. And it's costing organizations. Most AI transformation programs are built on a simple assumption… Train people. Increase adoption. Drive usage. And the value will follow. This study suggests something uncomfortable: The tools and metrics many organizations are using to demonstrate progress may not be measuring value at all. Usage may simply be creating more usage. For the past two years, that distinction barely mattered. AI was effectively subsidized. Uber recently disclosed it burned through its entire 2026 AI budget in just four months. Its COO admitted it's still difficult to draw a direct line between rising AI costs and customer value. Uber won't be the last organization facing that question. The discounts they received to adopt those tools are quicky disappearing. And leaders who built their business cases around adoption metrics are about to face a harder question: Where is the business value? Measuring AI activity is easy: Training completion Adoption Active users Prompt volume Measuring AI transformation is much harder: Workflow redesign Decision quality Knowledge architecture Operational change One creates dashboards. The other changes how work gets done. And that's where many organizations are stuck. Not deciding whether AI works. Deciding whether they're willing to fund the foundations that allow more than a handful of strategic projects to succeed. The foundation that creates lasting value isn't another tool. It's the semantic layer, decision logic, governed platforms, and embedded expertise that enable teams to redesign work instead of simply accelerating it. Usage without those conditions doesn't become transformation. It becomes habit. The organizations that scale AI won't be the ones with the highest adoption rates. They'll be the ones willing to fund the foundations that make deep operational redesign repeatable. | 24 comments on LinkedIn

PodHive AI als NotebookLM-Ersatz – ein EU-Anbieter im PraxistestIn dieser Folge schaue ich mir PodHive.ai an: ein Tool aus Bonn, das Audio-Podcasts aus eigen...

KI-Kennzeichnung wird Pflicht: Was Unternehmen ab 2026 zu Deepfakes, KI-Texten, Chatbots, Fristen und Bußgeldern wissen müssen.

𝘼𝙧𝙚 𝙬𝙚 𝙪𝙣𝙡𝙚𝙖𝙧𝙣𝙞𝙣𝙜 𝙝𝙤𝙬 𝙩𝙤 𝙩𝙝𝙞𝙣𝙠? Anthropic surveyed over 81,000 people globally to understand their real expectations from AI. The top responses are deeply human: Professional Excellence: 18.8% Time Freedom: 11.1% Learning & Growth: 8.4% On the surface, technology delivers: 32% experience immediate productivity breakthroughs and nearly 10% feel a direct lift in learning. 𝗕𝘂𝘁 𝘁𝗵𝗲 𝗯𝗮𝗰𝗸𝗹𝗮𝘀𝗵 𝗶𝘀 𝗶𝗺𝗺𝗲𝗱𝗶𝗮𝘁𝗲  16.3% point to a dark reality: 𝗖𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗔𝘁𝗿𝗼𝗽𝗵𝘆. This is not a vague prediction. Educators and academics witness this intellectual decline 2.5 to 3 times more frequently than the average participant. 𝗪𝗵𝘆 𝗶𝘀 𝘁𝗵𝗶𝘀 𝗵𝗮𝗽𝗽𝗲𝗻𝗶𝗻𝗴?  In corporate learning, AI is currently being abused as a cheap shortcut. Companies flood their systems with low-effort AI content, fueling an illusory productivity that leaves 18% of users deeply concerned. Employees blindly click through generated slides, offload critical thinking to machines, and enter the workplace as "human hard drives" - stuffed with rote memorization but entirely lacking real capability. 𝗧𝗵𝗲 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻: 𝗦𝘁𝗼𝗽 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗽𝗿𝗼𝗰𝗲𝘀𝘀. Technology only unfolds its true value when it helps us understand deeper, not when it replaces our intellect. The true learning tutor is not generative AI, but a biologically accurate, adaptive algorithm: ➡️ Time as a variable, competence as a constant:  It deconstructs rigid, time-based benchmarks and tailors the path in real-time to the learner's unique needs. ➡️ Zero Blind Spots:  It uncovers dangerous unconscious incompetence and drives every single employee to documented 100% proficiency. ➡️ Brain Training over Content Spam:  While generative AI accelerates content production, our algorithm ensures the human brain remains actively challenged and developed. 👉 Find deeper data, graphics, and the full qualitative breakdown of this global study in the attached PDF slideshow.

Aligning agent work with human will — 35 professions, graded against 2,066 fact-anchored criteria, evaluated by what workers actually want automated.

I just read all 245 paragraphs of the Pope’s encyclical on AI. Here are 8 fantastic ideas (even if you aren't religious). The Pope just released a major document on artificial intelligence. It's called Magnifica Humanitas. Cards on the table, I’m not a Catholic. I am Christian though. But even if you aren’t, good thinking is good thinking. And this is some of the clearest writing on AI I've read all year. So here are the 8 ideas that stood out: 1. We don't get to opt out of this moment The encyclical refuses to treat AI as "merely yet another theme to be studied or a crisis to be managed." It's already here. Pretending it’s not happening or that it's someone else's problem is not a responsible option. 2. We are building something. The only question is what. This is the heart of the whole document. The Pope uses two images from the Bible. Babel: a tower built on pride and efficiency that ends in confusion. And Jerusalem: a city rebuilt "piece by piece" through shared responsibility. "The primary choice is not between a 'yes' or 'no' to technology, but rather between constructing Babel or rebuilding Jerusalem." 3. Technology is never neutral. "Technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate and use it." I've said this for years, borrowing from Neil Postman. Good to see it stated so plainly. 4. AI is "cultivated," not "built (and we don’t really understand it) Even the people who design these systems have "only a limited understanding" of how they actually work. Nobody is fully in control. That should make all of us a little more humble. 5. Don't confuse this "intelligence" with yours. AI systems "merely imitate certain functions of human intelligence." They "do not undergo experiences, do not possess a body, do not feel joy or pain." They can simulate empathy without having any. The word "intelligence" may smuggle in ideas that simply aren’t true. 6. "Having more" is not "being more." Progress without moral growth produces "an increase in means without a growth in humanity." More capability does not automatically mean a better life. Having more is not an automatic good. 7. Watch who holds the power. AI "tends to amplify the power of those who already possess economic resources, expertise and access to data." A small number of private actors increasingly set the rules. That concerns everyone, believer or not. 8. A civilisation is judged by its care, not its power. Here’s my favourite line in the whole document: "The quality of a civilization is measured not by the power of its means, but by the care it is able to offer, by its ability to recognize the other as a face not merely as a function." Wow. And readily applicable to the classroom. Overall, this was a great read. I commend it to you regardless of your faith position. I’ve included a link in the comments below. Which of these ideas resonated with you most strongly? Let me know! | 117 comments on LinkedIn

Success in the age of agentic AI requires building systems that capture and amplify your company’s unique expertise

Join thousands of professionals working remotely to advance AI systems. Get matched to high-paying contract opportunities that fit your expertise and interests.

The next wave of skills management will not be about building a better skills taxonomy. It will be about building a better skills signal system. For years, companies treated skills like a static database: define the taxonomy, tag the content, ask employees to self-assess, then hope managers use the data. AI changes the equation - but only if we stop treating skills as HR vocabulary and start treating skills as living signals. A stronger skills strategy connects: * Work signals - what people actually do * Learning signals - what people are building * Career signals - where people want to go * Business signals - what capabilities the company needs next The real opportunity is not “AI-generated learning recommendations.” The real opportunity is helping leaders answer better questions: - Where are our capability gaps emerging? - Which skills are becoming strategically obsolete? - Which employees are ready to grow into adjacent roles? - Which learning investments are actually changing workforce capability? My point of view: the companies that win with AI-powered skills management will not be the ones with the biggest taxonomy. They will be the ones with the most "trusted skills signals" - and the operating discipline to act on them. #SkillsManagement #LearningAndDevelopment #AIinHR #FutureOfWork #LearningTechnology

I finally got around to reading the Insight222 PA & AI benchmarking report (https://lnkd.in/gGVBPTBb, yes, I still read reports rather than just the AI summaries of them... There's a time for a quick scan and themes and then a time to really wallow in meaning and patterns....) Lots of nuggets in here but the one that I think broke my heart the most? Single digit percentages of my PA friends and compatriots are using these advanced tools to forecast their talent pipelines. All of you people wringing your hands about SWP and you're not using what you already have in front of your faces? Come on folks. If I'm honest, the low utilization of skills inference kind of killed me too. People! Let's get on the stuff that lets us shepherd the organization (and people into it!) into the future! Also - L&D personalization is such low hanging fruit! There are such great tools! Super surprised by this one. There's lots more in this 88 (!) page report. Notably - as reported here, PA ratios and workloads aren't really changing that much, and mostly aren't shrinking... Appreciation to Madhura Chakrabarti, PhD, Jay Dorio, Heidi Binder-Matsuo, Jonathan Ferrar for their tireless work in support of this community. AI tools notwithstanding, building a report that's meaningful, insightful and useful is harder than it looks! Happy reading!

If you work in L&D, AI is likely devaluing you and your role. Here’s why, and how to fight back. As someone who works in L&D, you probably feel something I hear a lot: that your role is being quietly flattened by AI. AI promised innovation and growth, but in practice its impact has been to increase the volume of content you’re expected to build in compressed timelines. If you feel this too, you're definitely not alone. As one person on my bootcamp said last month: “I’m a learning designer, but since AI appeared I’ve never felt more like a content developer”. Here’s what I learned when I dug into the details..... __________________________ 👉 The Problem: → L&D has access to <15% of the AI tools available to most other industries. → Over 90% of the AI tools built for L&D help to do one thing: develop more content. Almost none help us with the strategic work — diagnosis, prioritisation, impact measurement — that elevate our roles and get a function taken seriously. → The result is a narrowing of our roles, job losses and flat-lining salaries. 👉 The Solution: → Stop waiting for vendors to build the AI tools you need to drive real value. They're not coming — or at least not soon enough for you. → Start building your own AI tech stack to augment tasks outside content development using tools you already use: ChatGPT, Claude and Microsoft Copilot. Think: GPTs for data analysis, research bots to select optimal design approaches and agents to automate slow but critical tasks like monitoring completion data and running QA. → Treat building with AI as a core L&D skill, not a technical side-quest. It's what's separating the L&D pros getting promoted (complete with a ~20% salary premium) from the ones getting flattened. This is exactly the pattern we've seen play out in Product Management over the last two years, where the rise of self-taught “AI PMs” has driven a 340% increase in hiring and 50%+ increase in salaries. __________________________ In this week's blog post, I explore: → Why L&D's AI tech stack is so much smaller than other industries' → What a fully-formed AI tech stack for L&D would actually look like → How to build your own AI tech stack in ~a month, with the tools you already use & pay for. TLDR: building with AI is what separates the L&D pros getting promoted & feeling fulfilled from the ones getting flattened. Start flexing your builder muscles this week, with help from this post (link in comments). Happy building! Phil 👋 | 13 comments on LinkedIn

Unternehmenswissen erschließen?

A quiet vindication. A year ago I published "Learning in the Flow of Context" and argued that Palantir's Forward Deployed Engineer (FDE) was actually L&D's Trojan Horse for AI advancement in disguise. I named the L&D analogue the Forward Deployed Learning Designer or FDLD. Today we have some big news. OpenAI launched the $4 billion OpenAI Deployment Company leveraging the FDE model. Anthropic announced a parallel platform within minutes. Salesforce, EY, Stripe, Workday, Adobe, Databricks, Poolside... all now running the same FDE play. I am not going to pretend I predicted the scale of this and any tie to it's cousin, the FDLD. Nobody did. "Deployment Is L&D's New Moat" -- the second part of last year's prediction -- is out today. It opens with the OpenAI and Anthropic news, walks through the broader wave, suggests the Top 10 critical FDE > FDLD skills, and includes the original Part 1 so you have the full context. Heck, maybe it's time for a new "L&D"? Learning and Deployment? I encourage sharing and critique, and special thanks to the folks in comments for enhancing, challenging my thinking and ongoing encouragement. https://lnkd.in/euZ-WSFz

In meinem Vortrag "𝗕𝗲𝘆𝗼𝗻𝗱 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆: 𝗪𝗶𝗲 𝗞𝗜 𝗪𝗶𝗿𝘁𝘀𝗰𝗵𝗮𝗳𝘁 (𝘂𝗻𝗱 𝗔𝗿𝗯𝗲𝗶𝘁) 𝘁𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗶𝗲𝗿𝘁" auf der TechShow in Frankfurt ging es um die Auswirkungen der KI auf Wirtschaft und Arbeit.

Sign in or join now to see posts like this one and more.

New research shows that offloading learning tasks to AI can improve - rather than erode - human thinking and learning

Really solid insights in this new report diving deep into 51 successful enterprise AI deployments from Stanford Digital Economy Lab led by Elisa Pereira Alvin Wang Graylin Erik Brynjolfsson. Some of the most important takeaways with a little of my own commentary: 🧱 Technology is not the hardest part. The model is rarely the bottleneck. The real work is change, data, and process redesign, which accounted for 77% of the hardest challenges. Failed attempts are often part of the hidden cost of success. This is ultimately about organizational transformation. ⏱️ Timeline variance is organizational, not technical. Similar AI use cases can take weeks in one company and years in another. The accelerators are sponsorship, existing foundations, user willingness, and the ability to iterate rather than over-plan. It is and has to be understood as a journey. 🔁 Escalation-based models were associated with better results. The highest gains came when AI handled routine work and humans reviewed exceptions. Escalation models showed 71% median productivity gains, compared with 30% for approval or collaboration models. However this should not be generalized across decision types. 🚀 Executive sponsorship is about actions, not approval. Effective sponsors do not just endorse AI; they clear blockers, align business and technology, and make experimentation safe. The key implication is that AI adoption has to be led, not merely funded. 🛡️ Staff functions are the most frequent source of resistance, but some parts may become enablers after buy-in. Legal, HR, Risk, and Compliance were the biggest blockers at 35%, ahead of end users. But when they are brought into governance early, their caution is what makes scaling possible. Governance is a fundamental enabler of transformation, but for that the usual blockers need to understand the AI context. 🤖 Agentic AI works, but most firms have not used it, yet. Agentic AI represented only 20% of cases, yet showed strong gains. The real shift is not a new interface, but a redesign of who does what: AI takes larger chunks of work, humans move toward judgment and oversight. This requires redesign of workflows. 🗂️ Messy data is not a blocker if you design around it. Waiting for perfect data is often the wrong strategy. LLMs can help make unstructured, inconsistent, and scattered data usable, so the priority is to connect and expose data rather than clean everything first. Start with what you have and work at continually improving data integrity and architecture. 🔐 Security enables more than it blocks. Security did not kill successful projects; it made higher-value uses possible. Strong controls are a front-loaded cost, but once in place they allow AI to work with sensitive data at enterprise scale.

KI und grüne Technologien verändern die Arbeitswelt. Zwei aktuelle Reports zeigen, wie sich gefragte Kompetenzen verschieben und neue Herausforderungen entstehen.

𝗔𝗜 𝘀𝗸𝗶𝗹𝗹𝘀 𝗮𝗿𝗲𝗻’𝘁 𝗷𝘂𝘀𝘁 𝗳𝗼𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀. 𝗧𝗵𝗲𝘆’𝗿𝗲 𝗳𝗼𝗿 𝗲𝘃𝗲𝗿𝘆𝗼𝗻𝗲. If your execs can’t articulate AI’s value — you’re stuck. If your experts can’t translate use cases — you’re stalled. If your employees don’t trust the tools — adoption fails. This is the AI literacy gap — and it’s killing transformation before it even begins. 𝗪𝗵𝘆 𝗔𝗜 𝗹𝗶𝘁𝗲𝗿𝗮𝗰𝘆 𝗺𝗮𝘁𝘁𝗲𝗿𝘀? It’s not just about new roles or flashy tools. It’s about enabling everyone to understand, trust, and challenge AI. Gartner calls AI literacy a major trend for 2026 — and here’s why: → It’s tied to regulation (like the EU AI Act) → It drives responsible, real-world adoption → It prevents the two biggest risks: blind trust and blind rejection The idea is simple: The more people understand AI, the better they use it. That includes non-technical teams too. AI literacy means: → Knowing where AI fails (hallucinations, misuse) → Navigating compliance, ethics, and governance → Cutting through hype to focus on business value Gartner’s framework breaks it down into four key level: 𝗟𝗲𝘃𝗲𝗹 𝟭 – 𝗡𝗼𝗻𝗲: → No clue how AI works. Still far too common. 𝗟𝗲𝘃𝗲𝗹 𝟮 – 𝗕𝗮𝘀𝗶𝗰: → Understands AI concepts. Can follow, not lead. 𝗟𝗲𝘃𝗲𝗹 𝟯 – 𝗜𝗻𝘁𝗲𝗿𝗺𝗲𝗱𝗶𝗮𝘁𝗲: → Applies AI meaningfully in their work. The SME sweet spot. 𝗟𝗲𝘃𝗲𝗹 𝟰 – 𝗦𝘁𝗿𝗼𝗻𝗴: → Leads AI strategy. Evaluates trade-offs. Connects models to mission-critical goals. There’s no one-size-fits-all training when it comes to AI literacy. A tailored approach is essential. Technical teams need different training than executives or middle management. So what’s needed? Targeted upskilling — by role, by depth, by design. Because AI success isn’t just about smarter models. It’s about smarter people. ⇣ 𝗘𝘃𝗲𝗿𝘆 𝘄𝗲𝗲𝗸 𝗜 𝗯𝗿𝗲𝗮𝗸 𝗱𝗼𝘄𝗻 𝘁𝗵𝗲 𝗺𝗼𝘃𝗲𝘀 𝘀𝗵𝗮𝗽𝗶𝗻𝗴 𝗔𝗜 - 𝗹𝗲𝗮𝗸𝘀, 𝗹𝗮𝘂𝗻𝗰𝗵𝗲𝘀, 𝗳𝗶𝗲𝗹𝗱 𝗻𝗼𝘁𝗲𝘀 - 𝗶𝗻 𝗺𝘆 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 𝗛𝘂𝗺𝗮𝗻 𝗶𝗻 𝘁𝗵𝗲 𝗟𝗼𝗼𝗽 [𝗙𝗿𝗲𝗲]: https://lnkd.in/dbf74Y9E | 68 comments on LinkedIn

An eight-month study of 2,500 employees at KPMG discovered indicators that can help identify superusers.

Every CEO is talking about AI. I surveyed 703 about AI at work. The scary part? Most executives deciding what to automate and who to cut?  They don't use AI for their own work. And they'll never admit it. In big companies, the AI strategy starts like this: Inside the room: ↳ CEO: read about AI in the airport lounge. ↳ Board Member: loves the Wall Street hype. ↳ CIO: sees this as a chance to be seen as more strategic. ↳ Consultant: has the perfect slide to make you feel behind. ↳ Vendor: built a beautiful wrapper and called it AI. Outside the room: ↳ The middle managers who are the glue who cover years of ops debt. ↳ The director still waiting 9 weeks for a business intelligence report. ↳ The Customer Success lead who hears what frustrates customers daily. Then nothing in that AI strategy scales but the PowerPoint. But the execs already announced the transformation. Got the stock bump from Wall Street. 👉 The people expected to make it work? We find out from the headline. Adoption flatlines. Six months later HBR writes the case study: ‼️ How 95% of AI fails in enterprise. Here is how to join the 5% heroes 📌 : 1️⃣ Start where people won't block AI to protect their 19 years of doing work the same way. Let them build AI in, not defend how they have always done it. 2️⃣ Assume everyone is a little worried about looking foolish and admitting what they don't know. Small masterminds allow people to learn and apply AI together. 3️⃣ Don't listen to what the AI vendor can do, start with how people actually work. The manager knows what breaks when you delete the workaround. Listen to them first. 4️⃣ Build AI operating agreements for what the AI can and should never decide. A client gave an AI agent a simple prompt: clean test data. It ran the SQL command perfectly and incorrectly deleted 1.3M rows of real customer data. 5️⃣ Start with a specific work where failure is cheap. Internal FAQs. Meeting notes. Formatting documents. The stuff people do that no one has time to check twice. 6️⃣ One team. Real feedback. Not KPIs to impress the board with how fast AI is. A test to see how people adapt so you can go deeper. The press release doesn't fail when you try to scale it. The work does. Who's the smartest person about AI in your company that no one's asked yet? ♻️ Repost to help other leaders trying to work out AI strategies. ➕ Follow Betsy Tong for frameworks on how AI works. | 279 comments on LinkedIn

DIE LOGIK DER KI PASST NICHT ZU UNSEREN ORGANISATIONEN

𝗦𝗼 𝘆𝗼𝘂 𝗺𝗶𝗴𝗵𝘁 𝗵𝗮𝘃𝗲 𝗵𝗲𝗮𝗿𝗱 𝗼𝗳 𝘁𝗵𝗶𝘀... ⇣ A guy just became the first 1 person $1B+ company.  One guy. Zero employees. $1.8B run rate. Read that again. Medvi - selling weight loss drugs - did $400M+ in its first 12 months. One person. No team. No office. No org chart. Year two? Pacing nearly $2B. This isn't a startup success story. It's a structural shift in what a single human with the right tools can build. And here's what gets me. Every week I talk to enterprise leaders managing teams of hundreds of people who tell me they're "exploring" AI. Meanwhile one guy automated an entire pharmaceutical business end to end - customer acquisition, fulfillment, compliance, support - and printed a billion in revenue. The leverage gap between people who use AI and people who "plan to use AI" is becoming obscene. You don't need to build a billion-dollar company. But if this guy can automate everything, you can automate the thing you've been putting off since January. This is your weekly reminder to get started on using AI in your business. You can just do things. Start. | 18 comments on LinkedIn

Every aspect of organizational value creation will be Humans + AI. Our focus needs to be on architecting how they are integrated across all layers, and building loops where humans, AI, and systems all continually improve their capabilities. The Individual Augmentation layer is already happening apace by giving GenAI tools to knowledge workers. What is often still missing are the skills and mindsets for effective use and leverage. Many are now recognizing the reality that Humans + AI teams are moving to the center of work, requiring redesigned workflows, new skills, effective governance and accountability structures, and structures for better decision-making. Workflows must be redesigned to bring to bear the most relevant capabilities across the organization as work becomes more fluid, requiring Humans + AI orchestration processes to dynamically address challenges and opportunities. Every aspect of the organization needs to evolve based on data, insight, and learning. Learning loops need to encompass humans, AI systems, and how each of the other layers are architecting in constantly evolving systems. And since value creation largely happens across organizational boundaries, not just within companies, Humans + AI processes need to enable shared value creation, based on interoperability, trust layers, and platform structures. This diagram is an evolution from a previous one I created, simplifying and consolidating based on extensive work with leaders in helping them apply Humans + AI principles to transform their organizations. Would love any thoughts and comments

A major event happened in the world of AI today. Anthropic accidentally leaked the future of AI agents. 512,000 lines of Claude Code's source code were accidentally included in a routine software update. Within hours, developers had reverse-engineered the entire thing. The security story is interesting. But the feature roadmap story? That's what every learning professional should be paying attention to. Buried in the code are 44 unreleased features. Fully compiled, production-ready code. Three of them fundamentally change the game for anyone producing learning content at scale: KAIROS is the always-on agent. Named after the Greek concept of "the right time," this is a persistent background mode. It watches your working environment, consolidates its own memory while you're idle, and proactively acts on what it notices. Imagine an agent that monitors an SOP version, notices it's updated, and automatically regenerates your performance support experience without being told. Coordinator Mode is multi-agent orchestration. One Claude instance spawns and manages multiple worker agents running in parallel. Not "I'll run three things one at a time." Actual parallel execution with convergence. This is for anyone managing content pipelines with multiple derivative outputs. Think of a single training video that needs to become an email communication, step-by-step instructions, and a micro lesson. Agent Triggers + Proactive Mode: This is the self-running pipeline. It's like setting a recurring calendar event, except instead of reminding you to do something, it tells the AI agent to do something. Oh, and it can sleep, wake on a condition, and resume. Anthropic accidentally showed its cards, and it looks like it's holding 4 aces (agents?). We all talk about AI-driven content production efficiency, but this is so much more. It's about the fundamental nature of how L&D teams will operate. When AI agents can orchestrate multi-step workflows autonomously, including watching for triggers, running parallel processes, and maintaining a persistent record of where a project stands, we stop thinking in terms of "tools I use" and start thinking in terms of "systems I supervise." And it applies everywhere: course development pipelines, assessment generation, content localization workflows, compliance update cycles, and onboarding content refreshes. This is content-zero, where production time and cost go towards zero. The organizations that figure these new agent capabilities out first will not just move faster. They'll operate at a fundamentally different scale with the same-sized team.

I. Situation  1. AI is poised to reshape work but lacks a predictive framework for where it is truly viable 2. A deep ontology of ~40,000 activities enables systematic mapping of AI across tasks 3. The strategic challenge is shifting from fragmented use cases to activity-level deployment logic II. AI Usage Concentration 1. AI adoption is highly skewed, with the top 1.6% of activities capturing over 60% of market value 2. 92% of applications cluster into just 6.8% of activities, revealing extreme concentration 3. This signals both saturation in core domains and outsized opportunity in underpenetrated tasks III. Dominance of Information Work 1. 72% of AI value sits in information-based activities, led by creation (36%) 2. “Think” activities dominate (~92%), especially creating, transferring, and modifying information 3. Physical work remains underpenetrated, positioning robotics as the primary expansion frontier IV. Core Activity Clusters 1. High-frequency use cases concentrate in generative creation (text, image, video, code) 2. Information transfer and summarization act as scaling layers across workflows. 3. Information modification (editing, converting, optimizing) operationalizes and refines outputs V. Interaction vs. Creation Dynamics 1. “Interact” (~32%) is largely driven by information transfer rather than standalone interaction 2. Most interactive systems are structurally extensions of information-processing workflows 3. AI advantage accrues to systems that control and transform information flows, not interfaces VI. Conclusion  1. AI impact is concentrated, not universal, with dominance in information-centric work 2. The ontology exposes both saturated domains and high-value whitespace 3. Competitive advantage will come from targeting underpenetrated activities and recombining workflows into AI-native systems Make sure to check out the insightful article by Alice Cai, Iman Yeckehzaare, Shuo SUN, Vasiliki Charisi, Xinru Wang, Aiman Imran, Robert Laubacher, Alok Prakash, Thomas Malone here: https://lnkd.in/eTQgjrAy _______ Stay Ahead of Transformative Innovation Follow The Futuring Alliance for regular insights, foresight, and practical tools to help your organization thrive in times of change. We support leaders across industries in turning future-focused ideas into real-world impact—through collaboration, innovation, and bold action. Let’s shape what’s next—together. #ai #ontology #robotics #futureofwork #transfomativeinnovation #futures #foresight #systems #systemschange #strategy

There's a paper in Nature that deserves more attention than it got. The finding: if you train AI models on data generated by previous AI models, the models progressively forget the edges of reality. Not dramatically but quietly. Rare knowledge disappears first. Outputs converge toward the average. Each generation becomes a slightly blurrier photocopy of the one before. They call it model collapse, and it's not hypothetical, they demonstrated it across LLMs, image generators, and other architectures. The implication that nobody is talking about loudly enough: as AI-generated content floods the internet, the data future models train on becomes increasingly synthetic. Which means the thing that made these models powerful in the first place, exposure to the full messy diversity of human thought and expression, slowly erodes. The researchers make a pointed observation at the end: genuine human interactions with AI systems will become increasingly valuable training data precisely because of this dynamic. The most important raw material for the next generation of AI is something that gets harder to find the more AI succeeds. | 156 comments on LinkedIn

I’ve been rereading one of my favourite books, Never Let Me Go, a 2005 dystopian novel by Kazuo Ishiguro. It starts as a story of children at an idylic English boarding school, full of friendship, creativity, and small obsessions. Gradually the truth emerges: these children are clones, created to provide organs for others. Their art, their self-expression (all supposedly proof of soul) exist as evidence for a system that will never allow them to flourish or extend their lives. This, I think, resonates with how we encounter generative AI. Both situations confront a mismatch between signal and life, between evidence and moral response. In Ishiguro’s novel, the clones’ art is never considered as sufficient evidence to support their requests to extend their lives. With Generative AI, our reading of outputs, untethered from life and accountability, may nonetheless misattribute agency, while obscuring the moral work of attending to real human minds whose life and cognition do depend on relational recognition. In other words, the ethical and existential question is “what do we owe generative AI?” Generative AI, like Ishiguro’s clones, shows us how we measure and relate, or fail to, when life and mind is truly at stake.

Manchmal gibt es Organisationen, die machen bei der AI-Adoption vieles richtig.Die Berliner Volksbank gehört dazu.Im Podcast habe ich mit Niclas Breier gespr...