Open New Learning Lab Resources

Fosway Group, in association with Learning Technologies, shares the first cluster of research from Digital Learning Realities 2026 - Strategy and Resourcing.

Fosway Group, in partnership with Learning Technologies, shares the second cluster of research from Digital Learning Realities 2026 - AI Value.

Every technology wave eats its own vocabulary. Words start sharp. Then the market discovers them. “Cloud.” “Platform.” “Transformation.” “AI.” At first, each points at something real. Then it becomes useful in a sales deck. Then it becomes useful in every sales deck. The word survives. The meaning does not. So here is Seale’s Law: ⚡ The precision of a word’s meaning is inversely proportional to its commercial value as a marketing term ⚡ It is a joke - but only just. The painful irony is that this is now happening to semantics. For years, those of us working with ontologies and knowledge graphs have made a fairly unfashionable argument: meaning matters. Context matters. Data is not ready for machines simply because it has been loaded into a warehouse, lakehouse, dashboard or API. When concepts remain implicit, identities are inconsistent and business meaning is trapped inside departmental dialects, the system may appear integrated. It is not. AI has made that impossible to ignore. LLMs have exposed a truth that many analytics programmes could postpone: enterprise data is not connected enough, described clearly enough or governed around meaning well enough for intelligent systems to use it safely. AI does not simply need more data. It needs data with identity, relationships, context and explicit business meaning. That is good news. It is also dangerous. 🔵 “Semantics” is about to become a beautiful container for vague promises. The word will be attached to products, features, copilots, roadmaps and platforms. Some of it will be real. Much of it will be rebranding. Organisations moving quickly will be sold noise at precisely the moment they need clarity. The message is simpler than the market will make it. 🔵 Every organisation has a data integration problem. To solve it for AI, you must connect your data and organise that connected data through an ontology: an explicit model of the things your organisation cares about, what they mean and how they relate. That is phase one: internal coherence. 🔵 Phase two is arriving fast: interoperability. No organisation operates alone. Supply chains, regulators, partners, customers, research networks and financial systems all cross organisational boundaries. If AI is going to work across those boundaries, meaning has to travel too. That requires open standards. Not proprietary semantic theatre. Not another silo with “ontology” printed on the brochure. Open standards for identity, connectivity and shared models of meaning. Semantics is having its moment. Our task is to protect the substance before the market hollows out the word. Your AI strategy is only as strong as your semantic foundation. Connect the data. Model the meaning. Use open standards. Prepare for interoperability. And if Seale’s Law is right, do it before we need a new word for the thing that finally made the old one profitable. 🔗 Building your semantic foundation? Let's talk: https://lnkd.in/ezHU2amU | 48 comments on LinkedIn

New research shows that AI can both harm *and* deepen human learning. What makes the difference is how we design AI use within the learning experience. Here's the TL;DR 👇 Maybe the most important finding from the research is that, left to their own devices, ~92% of learners offload thinking to AI meaning their learning diminishes (Cui et al., 2026). But with the right sort of design & direction, AI use can drive better engagement and improved outcomes for all learners. In practice, this means our job is no longer to decide whether AI belongs in the learning experience - it's already there. Instead, we need to develop new types of design expertise which enable us to be specific and intentional about how & when learners work with AI within the learning experience to optimise benefits and minimise risks. To get started on this journey, check out my carousel summary & latest blog post 👇 If you've designed an AI-assisted learning experience: how did you decide how to design learner-AI interactions? Happy designing! Phil 👋

The companies that thrive after disruption aren't the ones with the best strategy. They're the ones that learn faster than everyone else. One of the biggest myths in Learning & Development is that resilience comes from having the perfect process. It doesn't. A systematic review of 59 studies (Evenseth et al., 2022) found that organisational resilience is fundamentally built through organisational learning. Not just after a crisis, but before, during and long after disruption. Here were the biggest takeaways that stood out to me: 🧠 1. Learning is a strategic capability, not a training event. The most resilient organisations don't wait for something to go wrong before they learn. They continuously anticipate, adapt and improve. 🔄 2. Reflection without change isn't learning. After-action reviews, debriefs and "lessons learned" only matter if they lead to different behaviours and better decisions. ❌ 3. Unlearning may be even more important than learning. One of the review's most interesting findings was that resilience often depends on deliberately abandoning outdated assumptions, routines and mental models. 🤝 4. Knowledge must flow. Resilience grows when experiences are shared across teams and even between organisations...not trapped inside individual experts. 💬 5. Culture matters. Psychological safety, trust and supportive leadership create the conditions where people are willing to experiment, speak up and learn from mistakes. For me, this reinforces something I've believed for a long time. The organisations that outperform over the next decade won't necessarily be those with the smartest people or the biggest training budgets. They'll be the ones that create systems where learning is continuous, knowledge is shared, and outdated thinking is challenged before it becomes a liability. Resilience isn't a capability you buy. It's a capability you learn.

The Summer 2026 Edition

I freakin' love studies that match with reality - Boston Consulting Group (BCG) just nailed it with their new AI at Work Survey. These are three trends we see in real life, all the time. Let's go through them. TIME SAVED ISN'T THE SAME AS VALUE CREATED BCG found that 42% of frontline employees who use AI regularly save a full day or more per week. But 66% get little or no guidance on what to do with that time, and more than half never redirect it to strategic work. This is our experience as well - including the biggest company we've ever worked with (a Fortune 5). They were seeing gradual productivity gains with AI - which is great! But there was a problem. They weren't finding the radical gains that leadership knew AI could produce. They invited us to come in and investigate why they weren't seeing the kind of sharp gains AI is capable of producing. The culprit was a very human one. Seems that when people get time back, they don't redirect it to innovation. They just keep filling their time up with more emails. Getting to ROI means recognizing that productivity can be the enemy of innovation. Time redeployment is the responsibility of leadership. (I wrote a newsletter on this called The Million Little Hockey Sticks.) USE CASES ARE A FRAGMENT, PROCESSES ARE THE WHOLE BCG found that companies that redesign workflows end-to-end outperform the ones that just deploy tools. Invent initiatives have nearly doubled since 2025. This is one of the reasons why we don't do use cases, we do processes. A use case is a fragment of a workflow. You have to think about why that process exists in the first place, how it drives value, start with first principles, and then imagine you have a brilliant new colleague (or 10) to help you. You have to think big, not small. UPSKILLING ISN'T A KNOWLEDGE PROBLEM BCG found 88% of respondents believe they'll need major upskilling in the next five years, yet only 36% feel properly trained. The problem isn't that leadership doesn't care about upskilling. It's that we keep believing this is a proficiency to learn rather than a set of habits and behaviors to master. That's why we call it mindset. Because it is not 'proficiency.' It's not 'competency.' Still skeptical? Just know that Anthropic, McKinsey & Company, KPMG US and many others are using that language: Behavior, not proficiency. Microsoft ran an entire research project showing this was the case (that actually proved our methodology was 2x as effective in producing quality work, but whatever.) Redesigning how you work is everything. The focus, as BCG says, must be on strategy, not on the tool. The tools matter way less than we think. Strategy over tools. Well said, BCG. | 68 comments on LinkedIn

Otto Scharmer has been helping leaders' sensemaking for decades. His insights are extraordinarily relevant in the AI era. Here he proposes four levels of collective action and strategic innovation, each with its own structure of attention. In this powerful article he starts with the same thesis as in my book Thriving on Overload: "Overwhelm has become a shared planetary experience." The key danger is intelligence monoculture, "the assumption that AI is the only intelligence worth investing in." He identifies three forms of intelligence: Artificial Intelligence: Predicting from patterns Organic Intelligence: Understanding through relationships Source Intelligence: Creating from emergence The four levels of collective action each require different leadership styles 🪜 Level 1.0: Pattern-Executing — Automating AI automates well-defined tasks, with leaders focused on delegation, oversight, and ensuring reliable execution. 🪜 Level 2.0: Pattern-Adapting — Augmenting Humans and AI work together to adapt existing approaches, with leaders providing judgment, context, and direction. 🪜 Level 3.0: Pattern-Shaping — Co-Sensing People and AI combine insights to understand complex systems, challenge assumptions, and shape new possibilities through collaborative sensemaking. 🪜 Level 4.0: Pattern-Originating — Deep Sensing and Cocreating Leaders create the conditions for breakthrough ideas and collective creativity to emerge, with AI supporting rather than driving the process. The fundamental leadership choice is the allocation of attention and budget across the levels. If there is insufficient attention at the higher levels, the organization's ability to create distinctive value will inevitably erode. Otto is absolutely right. Amid the extraordinary AI transition, leaders need to step back to truly see the big picture, and ensure that organic and source intelligence remain central to how organizations create and fulfil their vision and purpose.

Seven firms published AI reports this year.   None blamed the technology. Financial services leaders will recognise the diagnosis: programme governance. I read all seven back to back this week. McKinsey, IBM, Bain, BCG, Accenture, Anthropic, OpenAI.   Different research teams, different client bases, same conclusion. The operating model is the bottleneck, not the model, not the compute, not the vendor. Every one of them names the same gap.   Different language, same root cause. Decision rights aren't defined.   BCG found 58% of leading firms expect governance and decision rights to change under agentic AI. Most haven't started that work. Accountability isn't mapped.   When an agent makes a consequential call, three teams claim it and none of them own it. Data governance isn't there yet.   Agent output can't be verified, audited, or trusted at scale. 🚀 Subscribe to The Transformation Constant — my newsletter — https://lnkd.in/d_mtyWzr The firms actually scaling this in financial services and asset management aren't moving faster than everyone else. They did the boring work first: decision rights, an AI governance framework, and a data foundation that can survive an audit. Everyone else is running pilots on an operating model nobody redesigned to carry them. AI shows the cracks.   It does not fill them. The full 2026 agentic AI strategy pack, all seven reports: 1️⃣ McKinsey: The Agentic Organisation   ↳ Five dimensions that decide whether AI agents create value or just noise: strategy, operating model, technology, talent, and governance.   https://lnkd.in/evEzBQh8 2️⃣ IBM: Agentic AI Operating Model   ↳ Why efficiency-first AI rollouts stall, and the execution systems that turn agents into structural advantage instead.   https://lnkd.in/diWGccJr 3️⃣ Bain: Foundations for Agentic AI   ↳ The unglamorous work that actually determines scale: data readiness, governance, and execution discipline, before the agent goes live.   https://lnkd.in/ePgxJBqH 4️⃣ BCG: Leading in the Age of AI Agents   ↳ When an agent makes the call, who owns it. BCG's framework for closing the accountability gap before it opens.   https://lnkd.in/eCV-x6wM 5️⃣ Accenture: Agentic AI Platform Strategy   ↳ Why agents force a rebuild of platform strategy and vendor decisions, not just another layer of tooling.   https://lnkd.in/e_RWUZfr 6️⃣ Anthropic: State of AI Agents (2026)   ↳ Guardrails that scale with the agent, not against it. Built for regulated environments, governance-first.   https://lnkd.in/e4w3FnRb 7️⃣ OpenAI: State of Enterprise AI   ↳ Where enterprise AI value is actually showing up right now, measured from real usage data, not demos.   https://lnkd.in/eqQHwZd8 🔔 Follow Justin R. for more Enterprise Transformation content | 169 comments on LinkedIn

Vor zwei Wochen hatte ich auf LinkedIn gefragt, welches Thema euch aktuell am meisten interessiert. Future Skills lag klar vorn, also zeige ich in diesem Vid...

Seit gestern steht (m)eine erste, noch überschaubare Auswahl an interaktiven Lernmaterialien und Tools, die in den vergangenen Monaten mithilfe von generativer KI und dennoch erheblichem Aufwand (Ideenfinden, Konzipieren, Strukturieren, Orchestrieren, manuelles Coden und Debuggen) erstellt wurden, online zur Verfügung: 3ducation (Link in den Kommentaren).

«Ist es eigentlich schlimm, wenn ich ChatGPT für die Schule nutze?»

🚀 Mercedes-Benz bündelt HR und Konzern-IT in einem Vorstandsressort: 𝗣𝗲𝗼𝗽𝗹𝗲 & 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗧𝗲𝗰𝗵.

Irgendetwas ist gerade anders.

Deutschland hat ein eigenes offenes KI-Modell: Soofi S, trainiert in der Telekom-Cloud, unterstützt vom Bundeswirtschaftsministerium.

I'm going to try to explain this crazy viral story, because I'm not sure most people actually understand it: Claude seems to have an inner monologue. It thinks stuff without telling us, and that may change how we see AI. Here's what's happening: Here's what Anthropic researchers have learned... DUDE - NOBODY BUILT IT. WUT? First, there's a small internal workspace inside Claude where it holds and moves concepts around without ever writing them down. They call it J-space. Nobody designed it. It emerged on its own during training. This is where it gets confusing. I had to read so many articles to really understand this, and here's how I'd describe it: WHAT WE THOUGHT WAS HAPPENING BEFORE We thought an LLM just predicted the next word, over and over. Ask it the capital of France and it says "Paris" because "Paris" is the word most likely to come next. Not because it *knows* anything about France. We also thought that when a model wrote out its reasoning, those words were just more predicted text. They looked like thinking. They weren't actually driving the answer. So the picture was: math underneath, convincing words on top, nothing in between. WHAT'S ACTUALLY HAPPENING There's something in between. Claude holds real ideas in its head while it works - ideas you never see. Researchers found those. Then proved they matter by reaching in and changing one mid-thought. The answer changed with it. So there's a step between the question and the answer. It's not just saying words that sound like thinking - it's more than that. And that hidden step is what actually decides what Claude says. THE NEUROSCIENTISTS NOTICED J-space looks a lot like global workspace theory - the leading model for how the brain broadcasts conscious thought. Deliberate reasoning routes through it, routine stuff bypasses it entirely. Anthropic makes no claims about consciousness. But Stanislas Dehaene and Lionel Naccache, who built that theory, called this a landmark finding. MY TAKE We keep saying that Claude and LLMs in general behave like humans because they kind of learn and act and 'think' like humans. And we always get pushback, but this isn't just an academic conversation. But if you want to master AI, or drive that across your company, you have to understand that implementing AI is much closer to merging with a new company - a very human problem - than it is like a software implementation. I would argue it has almost nothing in common with the latter. Which means we have to treat upskilling completely differently - because nobody upskills to deal with new colleagues, they think about how they act, every day, and how to get the best out of these new relationships. That's AI, friends. It's weird. | 60 comments on LinkedIn

Today's AI Mindset Newsletter! This is the speech I would give as a CEO to my people about AI. The first thing we do in our senior leadership workshops is messaging. Get that wrong, everything else implodes. Here's some thoughts: For the full message, and to get this stuff in your inbox, subscribe to our newsletter! Many say it's the best thing they've ever read in the history of reading anything ever. (Find it at our AIM website.) Okay, here's the TL;DR... Nobody has ever won an argument by saying "get over it." Yet that's basically the CEO message on AI: you may not love it, but this is the future, so use it. THE FEARS ARE RATIONAL Your people are not children refusing to use the new CRM. They're good at their jobs. They have muscle memory. They spent twenty years building skills that AI now threatens, along with their ability to feed their family and their kid's job prospects. Nobody here is being irrational - they're paying attention. THE MANDATE IS EMPTY ANYWAY Every tech mandate that ever worked had an old thing to kill. Salesforce killed the spreadsheet, Slack killed the fifty-person email chain, and nobody needed convincing because the alternative got taken away. AI has nothing to kill because there's no legacy system for thinking bigger. Which means your people choose, every single day, whether to open the thing. It's voluntary whether the CEO admits it or not. THE SPEECH So I wrote the speech I'd give if I ran a large company. It starts with "yeah, this is scary" and it ends with why hallucinations are your people's job security. The whole thing hinges on one move: hand people something instead of demanding something. AI training is a gift from the company to the employee, not the other way around. You stop asking for productivity and you get it anyway. WHY IT WORKS Leading with vulnerability is the most underrated unlock in AI adoption (preach, Brené Brown!). When a CEO admits they were intimidated for a year, everyone in the room is suddenly allowed to be behind - which means everyone in the room is allowed to start. AI adoption is an employee engagement strategy, not a technology rollout. It comes down to whether your people believe you're on their side. | 14 comments on LinkedIn

An extremely useful research paper investigates the relationship between human and agentic organization structures, and the optimal organizational forms. It is substantially framed around an idea very aligned with my current thinking (will share in a bit), that the core issue is "contextual transaction cost". What are the gaps and costs in transferring work between agents, human or AI? Canhui LIU ran 8,000 identical knowledge-work tasks through seven organizational forms, spanning human-imitation, hybrid, and agent-native. Agent-native forms were 395% more efficient than the human-imitation family. However notably single-expert (so zero transaction costs) did very well. Best by far: agentic meta-organization, using an orchestrator to switch between organizational forms depending on the task. Dead last (surprise, surprise): committee. The short version: there are reasons why human organizational forms work for humans, given attention, collaborative transaction costs, and relationship value. However these to not map into agentic structures. The BIG question: what are the optimal Humans + AI organizational forms. I'm doing a lot of work on this and will be sharing more in the coming weeks and months.

Very interesting data from Anthropic on global use of AI, showing dramatic differences in usage (#1 Australia) and augmentation vs automation use (#1 Norway), as well as other instructive insights. A few of the standout findings from the data that point to GenAI adoption around the world: 🌍 Global usage is spreading unevenly across countries. The top 20 countries account for 48% of all per-capita usage, up from 45% in the previous report, so the leaders are pulling further ahead. Inside the US, usage across states is evening out, but between countries the divide is slowly widening. 🏙️ Small, technologically advanced countries punch far above their weight. The US, India, Brazil, Japan and Korea lead in absolute usage. But once you adjust for working-age population, the ranking flips to Israel, Singapore and Australia. On that per-capita measure Australia now sits first of 121 countries, using Claude at 6.4x its predicted rate. Capability density beats raw scale. 🤝 Higher-income countries collaborate rather than automate. Wealthier, higher-usage countries delegate less independent decision-making to Claude, using it as an assistant with humans in the loop. Just over half of Claude.ai conversations are now augmentation rather than automation, and that collaborative tilt is strongest at the top of the income range. 🧩 Lower-adoption countries often show more complex use. Where adoption is thin, the early users are the technical elite, so their usage looks sophisticated relative to richer markets. More casual, everyday use only emerges once AI diffuses to the wider population. 📚 Purpose of use shifts along the income curve. Higher-GDP countries lean toward work and personal tasks. Lower-GDP countries lean toward education and coursework, where students are a larger share of early users. 💰 Income alone does not explain the middle. For the 37 middle-income countries where most of the world's population lives, income is a weak predictor of relative usage. Education, connectivity, language access, regulation and local competition matter more than GDP.

Learn the basics of Human Performance Improvement (HPI) using Gilbert's Behavioral Engineering Model (BEM) in our informative blog.

„Mit KI gemacht!" ist zum Vorwurf geworden – als wäre KI-Nutzung eine Frage von Ganz-oder-gar-nicht und das Resultat immer schlecht.

Does AI have values? If you believe The Economist, the answer is “yes… AI models are Godless hippies”. As more people turn to AI to help them plan, communicate and make decisions about their work and personal lives, what values these models reflect and express are incredibly important. Values will influence what advice AI models give, what news they deem worth sharing, what facts they choose to surface and how they write content. For the most part, it seems, the values these models reflect is based on the language data used to train them. For example, when The Economist asked several of the most popular AI models to complete a World Values Survey questionnaire (in English), their values aligned almost exclusively with English-speaking and wealthy countries (see graphic). Elsewhere, researchers have found that changing the language used to ask a question changes the way models respond - because it changes what training data is used to inform the response. At Common Cause Australia we spend a lot of time talking about the importance of values in messaging and other forms of community engagement because values strongly influence our attitudes and behaviours. As the use of AI explodes across all areas of life, we need to start asking some serious questions about what values AI models are reflecting and expressing as they influence our daily lives. You can read the full Economist article here: https://lnkd.in/gVGbEM6K

KI muss Probleme lösen und einen Mehrwert haben, technisch umsetzbar sein und auch einen Bedarf treffen.

Wie kann und soll sich die L&D-Funktion im Zeitalter von KI weiterentwickeln? Insbesondere vor dem Hintergrund, dass die Produktion und Bereitstellung von Lernressourcen zunehmend von KI-Werkzeugen geleistet werden kann? Egle Vinauskaite und Donald Taylor zeigen in ihrem aktuellen Arbeitsbericht dre

Using data from more than 72,000 Word users, researchers compared sustained Copilot adopters with similar coworkers who adopted later, and found the pace of work in Word changed markedly after adoption. The study introduces a new way to measure AI impact as products and people evolve together.

🎙️ "If you want people to feel like they will lose their job to AI, or can be easily replaced by AI, then put it on the org chart." This really resonates... A study by Boston Consulting Group (BCG) challenges a worrying trend gaining traction in many organisations: treating AI agents like employees. The research, by Matthew Kropp, Julie Bedard, Emma Wiles, Megan and Lisa Krayer, PhD, finds that anthropomorphising AI may make systems feel more familiar, but it also creates unintended consequences. When AI is framed as a teammate rather than a tool, accountability becomes blurred, review quality declines, escalation increases, and trust and professional identity suffer. Managers exposed to “AI employee” framing caught 18% fewer errors and were significantly more likely to defer decisions upwards. The authors argue that the real opportunity lies not in putting agents on the org chart, but in redesigning work for responsible human-AI collaboration. That means redefining workflows and human roles, making accountability explicit, building managerial capability to oversee human-agent systems, resisting one-to-one “digital employee” thinking, and deliberately redesigning work around judgment, creativity, and ownership. The organisations that create value will be those that redesign operating models and workforce capabilities together. 🔗 The article is featured in the May edition of the Data Driven HR Monthly, which you can access here: https://lnkd.in/enxVdRH9 🔗 | 13 comments on LinkedIn

Diese Studie zeigt präzise, wie nicht-resilienter Umgang mit KI aussieht…

🤖 AI is ready but most workforces aren’t.

Many new learning tools have appeared, and they look impressive. But why don’t they always create real impact? 👇 Today, faster than ever, you can: 🎨 create beautiful infographics 💻 build a digital course 🤖 launch an AI practice coach 🎭 add role-play scenarios 📩 send nudges 🧩 provide job aids And still see no real behavior change. Not because the tools are bad. Many of them are actually very good. The problem starts when we treat tools as the strategy. Tools can support behavior change. But they do not create it by themselves. Real behavior change comes from the journey around the behavior. That journey may include digital tools, live conversations, manager involvement, practice, feedback, reinforcement, and support in the flow of work. The real design challenge is not to use more tools. It is to place the right elements in the right moments of the journey. This is why I find ADKAR such a useful lens for learning design. Originally, ADKAR comes from change management. But it fits beautifully into learning design when the goal is behavior adoption. You can use it as a planning framework when you design a new learning journey. It helps you make sure that the journey includes all the key conditions needed for change: A — Awareness Before people learn the skill, they need to understand why this behavior matters now. D — Desire Before people adopt the behavior, they need to see why it is relevant and valuable for them. K — Knowledge People need to know what good looks like: the principles, steps, examples, and criteria behind the expected behavior. A — Ability Knowledge becomes useful when people practice the skill in realistic situations and receive feedback. R — Reinforcement Behavior does not stick at the end of the training. It needs to be supported in daily work until it becomes part of the routine. You can also use ADKAR as a checklist when your current learning project is expected to create change but does not. It helps you identify which element was skipped, underestimated, or not properly supported in the behavior change process. For me, this is one of the most important shifts in learning design today: from choosing tools 👉 to designing journeys. Because tools matter. But the journey decides whether they create impact. A tool can train a skill. A well-designed journey makes behavior stick. Which part is most often underestimated when learning teams design for behavior change? | 60 comments on LinkedIn