AI's Limitations: Humans Still Key to Making Things Work

9 𝐌𝐮𝐬𝐭-𝐇𝐚𝐯𝐞 𝐀𝐈 𝐒𝐤𝐢𝐥𝐥𝐬 𝐟𝐨𝐫 𝟐𝟎𝟐𝟔 Most people think AI skills mean better prompts. That worked last year. In 2026, it’s not enough. The people who get ahead aren’t just good with AI tools. They know how to create and operate AI systems. If you want to stay useful at work, these are the 9 AI skills that actually matter: 1. 𝐏𝐫𝐨𝐦𝐩𝐭 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 – Turning vague ideas into clear instructions. Why it matters: AI only does what you tell it. Clear prompts save time and frustration. 2. 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧 – AI helpers that run multi-step tasks. Why it matters: Knowing what’s useful is the real skill. 3. 𝐀𝐈 𝐎𝐮𝐭𝐩𝐮𝐭 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 – Checking AI outputs for accuracy. Why it matters: AI can sound confident and still be incorrect. 4. 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐀𝐈 𝐀𝐬𝐬𝐢𝐬𝐭𝐚𝐧𝐭𝐬 – Custom helpers for daily tasks. Why it matters: Hours saved every week. 5. 𝐀𝐈-𝐏𝐨𝐰𝐞𝐫𝐞𝐝 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 & 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬 – Process large amounts of info quickly. Why it matters: Faster insights = better decisions. 6. 𝐀𝐈-𝐀𝐬𝐬𝐢𝐬𝐭𝐞𝐝 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭 – Build simple tools without coding. Why it matters: Test ideas fast and learn faster. 7. 𝐀𝐈 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧 – Connect tools so work runs automatically. Why it matters: Protect focus and scale efficiency. 8. 𝐑𝐀𝐆 𝐋𝐢𝐭𝐞𝐫𝐚𝐜𝐲 – AI uses your files & data contextually. Why it matters: Context = more useful answers. 9. 𝐀𝐈 𝐄𝐭𝐡𝐢𝐜𝐬 & 𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 – Safe & responsible AI use. Why it matters: Trust is hard to rebuild once lost. 𝐅𝐢𝐧𝐚𝐥 𝐓𝐡𝐨𝐮𝐠𝐡𝐭: AI tools will keep changing. AI skills compound. People who know how to build AI systems will always outpace those chasing new features. Which skill are you learning first? AgenticDay is where founders, builders, and leaders go beyond AI hype. We focus on how agentic AI systems are actually designed, deployed, and scaled — not just prompts or tools, but real architectures, workflows, and execution. If you care about building AI that can reason, act, and adapt in the real world, AgenticDay is where those conversations happen. Follow us for insights, updates, and strategies on building and scaling agentic AI systems. #AI2026 #AISkills #FutureOfWork #AgenticAI #AIAgents #AIWorkflows #CareerGrowth #TechSkills #AIProfessionals #AITrends #GenerativeAI #AIProductivity #DigitalSkills #AIForWork #LearnAI #MachineLearning #DeepLearning #ArtificialIntelligence #NextGenAI #AILeadership #TechLeadership #Automation #RAG #PromptEngineering #AgenticSystems #AIResearch #Innovation #DigitalTransformation #AIForFounders #AIInBusiness #AIInsights #AIEthics #ResponsibleAI

AI tools are wiping away the usual technical barriers to technology, which means that more than ever before, getting value from these tools depends on mindset. Here are five mindsets I've come across that stop people from getting the full value from AI, and the one mindset that I think drives the biggest impact.   1️⃣ The sceptic: "I can't trust the outputs." How to disarm - Let them explore, learn about and document the limitations of outputs. - Teach techniques to improve the quality and reliability of outputs.   2️⃣ The dodo: "If I use it fully, I'll be out of a job." How to disarm - Map out the ways they don't currently add value, but would love to if they had the tools and time. - Find ways that AI can help them either directly do these things, or free up time to do these things.   3️⃣ The technophobe: "I'm not a techy person, it's all a bit complicated for me." How to disarm - Build confidence in playing with AI's capabilities in safe, simple scenarios, e.g. "Create a report about flowers". - Pair people up with someone more confident on the technology who can help them experiment.   4️⃣ The doomsayer: "AI is bad for the world." How to disarm - Work through the worries they have in a one-to-one chat and what that means for how they engage with the technology at work. - Get them comfortable using AI for specific use cases, starting with smaller uses they're comfortable with.   5️⃣ The literalist: "I don't question my AI outputs, they're always good enough." How to disarm - Share real examples of AI outputs that hadn't been tailored or verified and the negative impact they had. - Build in mnemonics to their AI usage, e.g. Always SIT before sharing AI assisted work, check: Sources, Impact and Tone.   🌟 The Innovator: "I'm excited to experiment and learn what actually helps me work better." How to encourage - Build in times for people to take a step back and reflect on how AI might help them in what they're really trying to do at work. - Reward experimentation and share learnings widely rather than trying to hide when things don't work well.   Which mindsets have you seen help or hinder getting value from AI tools?

🤖 𝐀𝐈 𝐃𝐢𝐝𝐧’𝐭 𝐆𝐞𝐭 𝐒𝐦𝐚𝐫𝐭𝐞𝐫 𝐎𝐯𝐞𝐫𝐧𝐢𝐠𝐡𝐭. 𝐖𝐞 𝐉𝐮𝐬𝐭 𝐅𝐢𝐧𝐚𝐥𝐥𝐲 𝐆𝐚𝐯𝐞 𝐈𝐭 𝐑𝐞𝐚𝐥 𝐖𝐨𝐫𝐤. For years, AI lived in demos. Chatbots that answered FAQs. Models that generated fun but useless text. Tools that impressed, but didn’t move outcomes. Then something shifted. AI stopped being a novelty and started becoming infrastructure. Not because the models magically evolved, but because we changed how we use them. 🧠 The real AI revolution is operational, not conversational The biggest mistake people still make about AI? Thinking it’s about “talking” to machines. Modern AI is less about conversation and more about coordination. • Planning • Deciding • Sequencing actions • Executing tasks • Learning from outcomes AI isn’t replacing thinking. It’s replacing friction. ⚙️ From tools to teammates Old software waited for instructions. New AI systems: • Observe environments • Interpret context • Choose actions • Escalate uncertainty • Improve over time That’s not automation. That’s agency. And agency changes everything. When AI can act, the question shifts from: ❓ “What can it generate?” to ❓ “What are we comfortable letting it decide?” 🧩 Why most AI projects fail quietly AI failure rarely looks dramatic. It looks, disappointing. The model works. The demo shines. The pilot launches. Then: • Adoption stalls • Trust erodes • ROI becomes fuzzy • Humans bypass the system Why? Because AI systems are often dropped into broken workflows. AI doesn’t fix chaos. It scales it. 🧱 AI needs structure more than intelligence Counterintuitive truth: The more powerful the AI, the more constraints it needs. Great AI systems are built on: • Clear objectives • Defined boundaries • Escalation paths • Human override mechanisms • Feedback loops Freedom without structure isn’t intelligence. It’s unpredictability. 🔍 The hidden cost of “just adding AI” Every AI system introduces: • New failure modes • New bias vectors • New security surfaces • New accountability gaps 🤝 Humans aren’t being replaced, they’re being repositioned The most successful AI deployments don’t remove humans. They move them upstream. Humans become: • Goal setters • Exception handlers • Ethical governors • Sense-checkers • System designers AI handles: • Repetition • Scale • Speed • Pattern detection The future isn’t human vs AI. It’s human with leverage. 👉If you’re working with AI today, ask yourself: • Where does this system act without supervision? • What assumptions does it quietly make? • Who is accountable when it fails? • What humans are empowered, and which are sidelined? AI doesn’t eliminate responsibility. It concentrates it. 💬 How comfortable are you letting AI make decisions, not suggestions? #ArtificialIntelligence #AI #AgenticAI #FutureOfWork #AILeadership #ResponsibleAI #AITransformation #GenerativeAI #Automation #TechStrategy #DigitalTransformation

What’s the “state of AI” at your company? How are you approaching it? Skeptical? Cautious? Curious? Or are you dipping your toes in? Are you Tinkerers? Builders? Experimenters? Or better yet, are you All In? I’ve been hearing from many of you and it’s clear, there’s no single “state of AI.” The reality is, we’re still early on in people’s understanding of AI and their willingness to accept it, understand it, leverage it. 𝟭. 𝗦𝗸𝗲𝗽𝘁𝗶𝗰𝗮𝗹 𝗮𝗻𝗱 𝗖𝗮𝘂𝘁𝗶𝗼𝘂𝘀 Some of you are hesitant, resistant. You see AI as risky, untrustworthy. You’re worried about protecting your data. You also worry about job displacement. Your conversations are dominated by questions of safety, not opportunity. You’re looking for assurances around data protection with clear evidence that AI augments your people’s work rather than replaces them. 𝟮. 𝗖𝘂𝗿𝗶𝗼𝘂𝘀 𝗯𝘂𝘁 𝗧𝗲𝗻𝘁𝗮𝘁𝗶𝘃𝗲 Others have moved past the skepticism; you’re experimenting in small, low-stakes ways - using AI to summarize meetings, generate simple content. For you though, AI still feels like “magic”; novel but not yet operationalized. You’re intrigued, but you haven’t connected it to business outcomes. Your focus is on discovering what’s possible and building basic literacy. 𝟯. 𝗟𝗲𝗮𝗿𝗻𝗲𝗿𝘀 𝗮𝗻𝗱 𝗧𝗶𝗻𝗸𝗲𝗿𝗲𝗿𝘀 A growing group of you are learning how to use AI effectively. You’re refining your prompting skills, experimenting with assistants, embedding tools into workflows. You may not have enterprise strategies yet, but you're developing internal champions; “AI fluent” teams who can translate curiosity into capability. This is where internal AI innovation often begins - in small, bottom-up experiments. 𝟰. 𝗕𝘂𝗶𝗹𝗱𝗲𝗿𝘀 𝗮𝗻𝗱 𝗘𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝗲𝗿𝘀 Others of you have moved beyond exploration. You're building tangible things: proofs of concept, lightweight “agents,” pilot workflows. You’re testing how AI can automate decisions, assist employees, connect across systems. These efforts often begin in innovation labs but they’re increasingly reaching business units. Your focus is on showing early wins, learning fast, de-risking through experimentation. 𝟱. 𝗖𝗼𝗺𝗺𝗶𝘁𝘁𝗲𝗱 𝗮𝗻𝗱 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝘃𝗲 Finally, some of you are going “all in.” You see AI not as a tool, but as a foundation. You’re embedding AI literacy into job expectations. For you, AI isn’t optional or experimental; it’s part of your operating model. You're building sophisticated agents, agentic workflows, fully agentic systems that make decisions, execute tasks, and learn continuously. Your goal is to make AI an invisible layer that accelerates the business. Wherever you find yourself in this spectrum, this isn't a binary moment. Sure, there are valid reasons for each of these groups. But AI adoption is a journey of mindset before it’s a journey of tools. The question at hand isn't whether AI will shape your business. It’s how you will consciously choose to shape that future. #discoverthefuture

I've been thinking about what’s driving a lot of the polarization around AI usage across industries. Let's set aside environmental, socio-economic, and quasi-religious arguments for a moment. They already carry a lot of embedded bias. What I keep running into are differences in how people experience AI in practice. One group (A) values AI primarily for its ability to alleviate friction in activities they previously couldn’t perform effectively, or at all. Another group (B) sees AI as a way to access perspectives, skills, or identities they didn’t previously have access to, and uses it as an accelerated learning surface. Both groups benefit from reaching levels of output that, until recently, were largely reserved for people who had invested years building that competence. A third group (C) struggles with this. Not necessarily with AI itself, but with what feels like a false sense of accomplishment. The issue is exacerbated when AI collapses distinctions that were previously enforced by time, practice, and irreversible trade-offs. In other words, by “paying dues.” These three groups tend to stay in tension, and that tension becomes visible at predictable fault lines. A) When Group A experiences working with Group C as friction, they tend to minimize the effort it takes to build the competence AI now makes accessible. This isn’t always intentional; it’s often a consequence of distance from the process and incentives that reward efficiency. Group A may appear aligned with Group B, but that alignment often breaks down around questions of responsibility. Specifically, they may disagree on what work should be “handed off” to humans. B) Group B is learning quickly, often at speeds that were previously improbable. They celebrate outcomes like Group A, but also celebrate learning like Group C. However, their apparent alignment with Group A can make sustained dialogue with Group C difficult, until they encounter problems AI cannot solve (yet). Those moments tend to recalibrate what “learning” actually requires. C) Group C may appreciate Group B’s genuine effort to learn with AI, but still experiences them as potential competition. In environments where hiring and recognition increasingly favor AI-augmented output over process mastery, Group C may find an ally in Group B, recognizing the challenges that may present themselves when the limits of AI usage are reached. These groups aren’t mutually exclusive. Most people I talk to occupy two or more of them, depending on the domain or the task. The tension isn’t about identity so much as state: where someone is relative to a competence boundary, and how much of that boundary AI is currently able to mask. It’s a weird time to be a polymath. It also feels like the early formation of a different kind of Renaissance figure: someone less defined by individual mastery, and more by the ability to integrate human judgment and machine capability across those boundaries, especially where things start to break.

The 4 Layers That AI Cannot Automate On Wednesday, I promised to break down the "Human Value Stack" - the specific areas where professionals still hold the strategic advantage over algorithms. This is the most important concept for career durability in 2026. Here is the reality we are facing: AI is driving the cost of Prediction (drafting, summarizing, coding, pattern matching) toward zero. In economics, when the cost of a resource drops, the value of its complements rises. That means as AI gets cheaper and faster at generating outputs, the human ability to evaluate those outputs becomes priceless. We need to stop competing with the machine on speed. We need to compete on depth. Here are the 4 layers of the Human Value Stack where your real value now lives: 1. CONTEXT (The "Where") AI knows the data. It does not know the room. It doesn't know the political history of your department, the specific tension between two stakeholders, or the unwritten rules of your client's culture. ->The Skill: Moving from "generating content" to "aligning content with reality." 2. JUDGMENT (The "What If") AI predicts the most likely next word or outcome. It does not understand consequence. It cannot weigh the reputational risk of a bad email against the benefit of a fast one. It cannot feel the weight of a strategic trade-off. ->The Skill: Moving from "doing the task" to "owning the decision." 3. EMPATHY (The "Who") AI can simulate empathetic language. It cannot build trust. In negotiation, conflict resolution, and leadership, people do not want an optimized answer. They want to feel heard. Trust is a biological reaction, not a data output. ->The Skill: Moving from "transactional efficiency" to "relational intelligence." 4. RESPONSIBILITY (The "Ownership") This is the bottom line. An AI agent can execute a transaction, but it cannot go to jail, be fired, or lose its reputation. Only a human can take moral and legal accountability for an outcome. ->The Skill: Moving from "executor" to "owner." ---->The Lesson: If your work creates value primarily through speed and volume, you are in the danger zone. If your work creates value through Context, Judgment, Empathy, and Responsibility, you have never been more valuable. Don't try to out-run the robot. Out-human it. Discussion: Which of these 4 layers do you find yourself leaning on most heavily this year?

"𝘈𝘐 𝘪𝘴 𝘢 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯 𝘭𝘰𝘰𝘬𝘪𝘯𝘨 𝘧𝘰𝘳 𝘢 𝘱𝘳𝘰𝘣𝘭𝘦𝘮." I can't remember where I first heard this, but after almost 2 years of researching and working with AI in AEC, I see it play out constantly in our industry. There's a pattern: New technology emerges. Looks promising. You think, what's my biggest headache? Let me try solving it with this new tool. It doesn't work. You assume either the tech isn't good enough yet, or you're not using it properly. Everyone in the field had that early AI experience. But the real issue is simpler. You're likely trying to solve the wrong problems. 𝘔𝘦𝘵𝘢𝘱𝘩𝘰𝘳 𝘪𝘯𝘤𝘰𝘮𝘪𝘯𝘨: A power drill is a useful tool for building wood cabins. But if you try using it for chopping down trees - it just won't cut it (hah). If you understand what it actually does, you'll find dozens of problems it solves perfectly. Of course, AI is not comparable to a power drill, but the metaphor stands; the challenge isn't learning to use it. It's knowing which problems to bring it to. I see people trying to use AI to automate tasks that require absolute consistency. To make final decisions that need human judgment. To replace expert review processes. These attempts fail, and their conclusion is always the same: "AI isn't ready yet" or "I need to learn better prompts." 𝐖𝐫𝐨𝐧𝐠. These are simply the wrong problems. AI generates variations, it doesn't ensure consistency. It can suggest options, it doesn't make decisions. It accelerates and complements thinking, but doesn't replace expertise. It makes educated, statistics-based solutions, not (always) real answers. Understanding this changes everything. Instead of throwing AI at your hardest problems, you start asking: what does this technology actually do well? Then you find problems that match. At Henning Larsen, we built our AI approach on four principles: 𝐋𝐢𝐭𝐞𝐫𝐚𝐜𝐲, 𝐓𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲, 𝐀𝐠𝐞𝐧𝐜𝐲, and 𝐀𝐜𝐜𝐨𝐮𝐧𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲. Literacy comes first. We teach and learn how AI works: how it's trained, how it "thinks", what it can and cannot do. Not prompting techniques or workflow tutorials - but the technology itself. Then, people on the ground, working on actual projects, identify not just better solutions, but better problems. This bottom-up approach is proving itself in our practice every day, in what is arguably one of the most successful AI adoption processes in the entire field. The best AI implementations I've seen don't come from people with the fanciest prompts or the most complex agentic workflows. They come from people who understand the tool well enough to know when 𝘯𝘰𝘵 to use it. If you have any thoughts about this, feel free to reach out or share - you know the drill <3 #ArtificialIntelligence #AEC #AIStrategy #AILiteracy

Last week, I explored what 2025 taught us about AI transformation. The insight settling in across industry: impact requires human work, not just technology deployment. Enter 2026, the realization deepens. Even incremental improvement is a challenge without addressing the underlying structure. Real gains need comprehensive transformation: reimagining work itself, not just adopting tools. And that requires widespread fluency across your organization, especially in less technical roles. What makes this moment different We're talking about a wholesale shift to innovation thinking, skills not traditionally distributed widely across organizations. For decades, people have been taught to maintain and adjust systems, not fundamentally reimagine them. That's a lot of muscle memory to work through. So, besides support and training to start this cultural shift, we need to develop the enthusiasm to engage, alongside AI fluency. Work going forward won't be about coasting; it will be about discovery and evolution. We're a generation of pioneers doing what's never been done. All hype aside, this world will be amazing if we build it that way. That's why we built Wish Bigger™ Wish Bigger™ is our people-focused method for building AI fluency and distributed innovation capacity. And it's fun— which matters more than you think when you're asking your entire organization to reimagine work. The cornerstone: Relevance and Trust Can your teams trust that this won't be another boring training initiative? That it’ll truly help their work, not just add to their workload? Wish Bigger™ builds that trust through personal relevance and real wins. People don't just learn to use AI—they become aware of their own work in new ways and discover how AI can transform it. Now Desktop Innovators™: people who've experienced AI making their work genuinely better. That changes how they show up. Individual fluency becomes team capacity becomes distributed innovation, crossing functions, levels, and tenure through project work that solves real problems. Building the adaptive capabilities your organization needs because just running it up the chain won't pace the rate of change ahead. Traditional consulting struggles here. Our approach? We nurture and scaffold, building the temporary structure that supports your people as they develop these capabilities. When our part ends, their innovation capacity continues to grow. Wish Bigger™ is a foundational element of our Human AI Rails™ system, the people-focused part that makes everything else possible. Along with Emergent Discovery™, this is how we build the path to ROI and competitive advantage. You can't transform work if people can't join in reimagining it. Your workforce is waiting It's not about them finding inspiration It's about bringing it to them Wish Bigger™ Next: How Emergent Discovery™ reveals the unseen realities of day-to-day work.

Let me tell you about the most expensive AI trap I see. I call it Side-of-the-desk AI. Here's how it works: Leadership decides AI is important. But they're not sure exactly how to approach it. So they send out a message: "We encourage everyone to experiment with AI tools when you have time." Feels responsible, right? Low risk. Let people explore. See what bubbles up. Here's what actually happens: A few enthusiasts dive in. They're the people who would have experimented anyway. They build some cool prompts. They save themselves some time on specific tasks. Most employees don't touch it. They're busy. They don't have "extra time." They're not sure what they're supposed to do. They're worried about doing it wrong. Nothing mission-critical moves. The enthusiasts are solving their own problems, not organizational problems. Twelve months later, you have scattered small wins and no compounding learning. The enthusiasts are frustrated because nobody adopted their discoveries. Everyone else feels even further behind. Leadership is confused about why their "AI initiative" didn't produce results. The hidden cost of Side-of-the-desk AI isn't the wasted tool spend. It's the wasted time. You burned a year. Your competitors who took a more systematic approach are now a year ahead. The gap widened while you were waiting for magic to happen. Side-of-the-desk AI isn't a strategy. It's hoping the problem solves itself. If AI matters enough to talk about, it matters enough to resource properly: Specific people accountable for specific outcomes. Time allocated for learning, not just "when you have time." Clear expectations about what adoption looks like. Otherwise you're just checking a box while your competition builds real capability.

What if the secret to mastering AI isn’t about crafting perfectly clear instructions, but about how you respond when things don’t go as planned? Nate B. Jones walks through how even the most detailed specifications can lead to flawed results if users neglect a critical skill: intent refinement. This analysis challenges the assumption that “clearer is better,” instead highlighting the nuanced relationship between human guidance and AI interpretation. Whether you’re a seasoned pro or a curious beginner, it reveals why some users consistently achieve exceptional outcomes while others remain stuck with disappointing results. https://lnkd.in/dqqeRmPK xTheAiSkills #ThatSetTopUsers #ApartIn2026