Business Trends

Over two years ago, I wrote about the emerging synergy between LLMs and ontologies - and how, together, they could create a self-reinforcing loop of continuous improvement. That post struck a chord. With GPT-5 now here, it’s the right moment to revisit the idea. Back then, GPT-3.5 and GPT-4 could draft ontology structures, but there were limits in context, reasoning, and abstraction. With GPT-5 (and other frontier models), that’s changing: 🔹 Larger context windows let entire ontologies sit in working memory at once.   🔹 Test-time compute enables better abstraction of concepts.   🔹 Multimodal input can turn diagrams, tables, and videos into structured ontology scaffolds.   🔹 Tool use allows ontologies to be validated, aligned, and extended in one flow. But some fundamentals remain. GPT-5 is still curve-fitting to a training set - and that brings limits: 🔹 The flipside of flexibility is hallucination. OpenAI has reduced it, but GPT-5 still scores 0.55 on SimpleQA, with a 5% hallucination rate on its own public-question dataset.   🔹 The model is bound by the landscape of its training data. That landscape is vast, but it excludes your private, proprietary data - and increasingly, an organisation’s edge will track directly to the data it owns outside that distribution. Fortunately, the benefits flow both ways. LLMs can help build ontologies, but ontologies and knowledge graphs can also help improve LLMs. The two systems can work in tandem.   Ontologies bring structure, consistency, and domain-specific context.   LLMs bring adaptability, speed, and pattern recognition that ontologies can’t achieve in isolation.   Each offsets the other’s weaknesses - and together they make both stronger. The feedback loop is no longer theory - we’ve been proving it:   Better LLM → Better Ontology → Better LLM - in your domain. There is a lot of hype around AI. GPT-5 is good, but not ground-breaking. Still, the progress over two years is remarkable. For the foreseeable future, we are living in a world where models keep improving - but where we must pair classic formal symbolic systems with these new probabilistic models. For organisations, the challenge is to match growing model power with equally strong growth in the power of their proprietary symbolic formalisation. Not all formalisations are equal. We want fewer brittle IF statements buried in application code, and more rich, flexible abstractions embedded in the data itself. That’s what ontologies and knowledge graphs promise to deliver. Two years ago, this was a hopeful idea.   Today, it’s looking less like a nice-to-have…   …and more like the only sensible way forward for organisations. ⭕ Neural-Symbolic Loop: https://lnkd.in/eJ7S22hF 🔗 Turn your data into a competitive edge: https://lnkd.in/eDd-5hpV

Getting Enterprise AI right Enterprise ontology and knowledge graph initiatives have a historical reputation: they are often where brilliant engineering teams go to spend years "mapping the business" and ship nothing. But LLMs have changed the physics of that problem. My colleagues from the Cogentiq team at Fractal have just published a post with a sharp reframe of this architecture. The core concept here is the Enterprise Delta, and it’s a massive shift in mindset: Foundation models already know what a customer, an invoice, or a profit margin generally is out of the box. You don't need to re-model the world. You only need to engineer the gap - the highly specific, proprietary way your business defines reality, resolves context, and handles exceptions. It turns an infinite project into a finite one. And in a market where foundational models are rapidly commoditizing, that gap is your true moat. It’s not the LLM, and it’s not just the raw data - it’s the governed, structured way your business resolves its own context. Two things that stand out for anyone moving past the RAG honeymoon phase: (a) Build for agents that act, not dashboards that show: This isn’t the passive semantic layer of the BI era ("what does this column mean?"). It encodes rules, tools, and expert reasoning paths. It is an active context layer designed for autonomous action, not just data visibility. (b) Bottom-up emergence: Instead of a top-down, "boil-the-ocean" master blueprint, it's discovered piece-by-piece, starting with a single high-value use case. You build the city by developing it organically, not by planning every street corner in advance. If you are currently figuring out the cognitive architecture required to scale agentic workflows safely without losing structural control, read the full deep dive here: https://lnkd.in/dRmXr4AM #EnterpriseAI #AIAgents #AI

Large language models (LLMs) can improve their performance not just by retraining but by continuously evolving their understanding through context, as shown by the Agentic Context Engineering (ACE) framework. Consider a procurement team using an AI assistant to manage supplier evaluations. Instead of repeatedly inputting the same guidelines or losing specific insights, ACE helps the AI remember and refine past supplier performance metrics, negotiation strategies, and risk factors over time. This evolving “context playbook” allows the AI to provide more accurate supplier recommendations, anticipate potential disruptions, and adapt procurement strategies dynamically. In supply chain planning, ACE enables the AI to accumulate domain-specific rules about inventory policies, lead times, and demand patterns, improving forecast accuracy and decision-making as new data and insights become available. This approach results in up to 17% higher accuracy in agent tasks and reduces adaptation costs and time by more than 80%. It also supports self-improvement through feedback like execution outcomes or supply chain KPIs, without requiring labeled data. By modularizing the process—generating suggestions, reflecting on results, and curating updates—ACE builds robust, scalable AI tools that continuously learn and adapt to complex business environments. #AI #SupplyChain #Procurement #LLM #ContextEngineering #BusinessIntelligence

Every time you ask your AI a question, it forgets everything it learned the last time you asked. That's how most AI knowledge systems work today. They're called RAG systems — Retrieval Augmented Generation. You upload documents. The AI searches fragments on every query. Builds an answer from scratch. Closes the session. Next question? Same process. Zero accumulation. Andrej Karpathy — the person who built Tesla's AI vision system and co-founded OpenAI — published something quietly last week that might change this fundamentally (https://lnkd.in/dNB5swS8). He calls it "LLM Wiki." The concept: Instead of searching your documents every time, the LLM builds and maintains a persistent knowledge base. A structured wiki of markdown files with cross-references, concept pages, and contradiction flags. New document arrives? The LLM doesn't just index it. It reads it, extracts the key information, updates existing pages, notes where new data contradicts old claims, and strengthens the evolving synthesis. His key insight: "The tedious part of maintaining a knowledge base is not the reading or the thinking — it's the bookkeeping." Cross-references. Keeping summaries current. Noting when new data overrides old claims. That's exactly what nobody in your team has time for. And exactly what LLMs don't get bored doing. What this looks like in practice: Step 1 — Pick one knowledge domain. Not everything. One area where your team wastes time re-finding information. Customer onboarding. Product specs. Compliance requirements. Step 2 — Set up the structure. Claude Projects or Obsidian + Claude Code as the wiki layer. Raw sources go in one folder. The LLM-maintained wiki lives in another. Step 3 — Feed sources one at a time. Let the LLM summarize, cross-reference, and file. You review. Redirect. Ask follow-up questions. The wiki grows with every session. Step 4 — Query against the wiki, not the raw documents. Answers are faster, more contextual, and cite specific pages. The open-source project Graphify already implements this pattern — claiming 70x fewer tokens needed to answer questions compared to raw-folder RAG. For now, this is still a developer concept. No plug-and-play enterprise solution exists yet. The benchmarks are anecdotal, not peer-reviewed. The wiki could drift or hallucinate if not curated. But the direction is clear: AI that builds knowledge, not AI that searches it. Save this for when your team's knowledge base fails you again. — 📌 Save this post for later ♻️ Share it to inspire your network Follow Hartmut Hübner, PhD for AI insights that work.

Most enterprise AI right now isn't intelligence. It's structured lookup… wrapped in an LLM. More specifically... it's digital-era thinking applied to systems that require machine-understandable meaning. The pattern looks like this: -> Take their data dictionary… terms, definitions, business rules -> Drop definitions and rules into spreadsheets -> Store them in PostgreSQL -> Embed documents and data into vector databases -> Retrieve the context through RAG -> Inject intent and logic into a prompt Clean? Yes. Structured? Technically. AI-first? Not even close. This is still a digital mindset applied to AI architecture.… and the wrong foundation for what LLMs actually need. Digital-era systems were designed to store and retrieve information efficiently. AI changes the requirement entirely. Now systems need shared meaning, decision logic, and machine-interpretable relationships. A relational database tells your AI what something is called. "Customer = someone with an account." But it can't tell it when "customer" means billing entity vs. legal counterparty… that one implies a different approval chain… or that crossing a threshold changes the outcome. This isn't a problem you can solve by adding another table, another vector store, or another instruction in a prompt. It's a knowledge representation problem. LLMs don't struggle with definitions. They struggle with equivalence, hierarchy, and implied logic… the connective tissue that gives concepts contextual meaning, not just labels. And it's why a semantic layers matter. 1️⃣ A digital mindset stores definitions. A knowledge mindset encodes meaning SKOS isn't a prettier data dictionary. It encodes preferred labels, alternate labels, examples, and semantic relationships… creating shared meaning across systems speaking different business dialects. 2️⃣ A digital mindset connects records. A knowledge mindset encodes relationships and rules OWL goes further than a relational model ever could. It encodes classes, relationships, constraints, inheritance, and inference rules… allowing systems to reason from concepts, not just rows. LLMs can't reliably infer business meaning from foreign-key relationships, table joins, or disconnected rules buried across systems of record. 3️⃣ A digital mindset retrieves information. A knowledge mindset reasons across it A knowledge graph is not SQL or a vector DB with more connections. Relational databases organize records. Vector stores retrieve semantically similar information. Knowledge graphs organize relationships between concepts and entities… enabling contextual reasoning across systems instead of isolated retrieval from them. This is what it means to be AI-first. It doesn't mean layering LLM capabilities - RAG, agents, MCP - onto digital architecture. It means rethinking the architecture itself… moving from systems designed to store information toward systems designed to represent meaning. Because AI doesn't create knowledge. It depends on it.

Optimization, machine learning, and LLMs are NOT competitors. They are three different layers of the same decision system. Machine learning helps us understand the world as it is likely to unfold. It turns messy historical data into signals, probabilities, forecasts, classifications, and patterns. It helps answer questions like: What is likely to happen? What demand might show up? Which customer behavior is changing? Where is risk increasing? Where is the system drifting? Optimization helps us decide what to do about it. That distinction matters. A forecast is not a decision. A prediction is not a plan. A dashboard is not an operating model. Knowing that demand may increase, supply may be constrained, or a customer may churn does not tell you what action to take when capacity, cost, timing, inventory, labor, service levels, and business tradeoffs are all competing at the same time. This is where optimization becomes essential. It translates predictions into economically coherent actions. It forces tradeoffs into the open. It asks: given what we believe, what should we do now? What should we buy, build, route, allocate, price, schedule, or prioritize? Then LLMs add another layer. LLMs help humans interact with the decision system. They explain. They summarize. They retrieve context. They help users interrogate assumptions. They generate scenarios. They make complex systems more usable. In many cases, they become the conversational interface between the business and the machinery underneath. But an LLM by itself is not a decision engine. It can sound intelligent while having no real concept of feasibility, constraints, economic tradeoffs, uncertainty, or downstream consequences. It can describe a plan beautifully without being able to prove the plan can actually work. That is why the future is not “AI versus optimization.” The future is systems where machine learning estimates uncertainty, optimization makes disciplined decisions under constraints, and LLMs help people understand, challenge, and interact with those decisions. The companies that win will not be the ones that simply add a chatbot to a dashboard. They will be the ones that build decision intelligence systems where data, prediction, optimization, simulation, and human judgment work together. Because the goal was never to predict the future perfectly. The goal is to make better decisions before the future arrives.

LLMs force a shift from performing work to specifying work. Because the machine only responds to what you articulate, you end up writing constantly: intent, constraints, context, expected output. The act is functional rather than literary. You strip away tone, polish and performance. What remains is the core of knowledge work: clarifying the problem, defining the goal, summoning the factual context. The immediacy of feedback is captivating. You see instantly how a clearer description yields a better result. Over time this trains the fundamental skills of high-leverage white-collar work: documenting ideas, setting objectives, defining deliverables, and enriching the brief with the right context. A generation now growing up with this muscle memory will expect work to run this way. Their default workflow will be: capture the idea in writing, specify the task, let the machine execute or assist. Meetings without agendas, projects without specs, assignments without context will feel amateurish. The norms of today’s office—already bloated by vague communication and performative collaboration—will look as dated as the mid-century rituals of secretaries typing dictated letters and executives spending afternoons in performative chatter over drinks. The shift is not that AI replaces office work; it rewires the discipline of doing it.

LLMs are going vertical → and functional. We’re moving from “everyday AI” to functional AI: domain-specific agents embedded in real workflows where enterprise value is trapped Proof the shift is led by LLM providers themselves: Banking: OpenAI is partnering directly with banks (e.g., BNY Mellon’s multiyear deal to upgrade its Eliza platform; NatWest’s UK-first collaboration). These are not generic chats, they’re deeply embedded, regulated-industry builds. Life sciences: Anthropic’s Claude for Life Sciences adds connectors to tools like Benchling, PubMed, 10x Genomics and offers domain skills: from protocol QA to bioinformatics workflows. That’s vertical by design. Healthcare: Google’s MedLM + Vertex AI Search for Healthcare targets clinical documentation and medical record retrieval: out-of-the-box isn’t enough; it’s workflow-native. Industrial: Siemens Industrial Copilot (with Microsoft) is scaling across factories and engineering teams. LLMs tuned to PLC code, Teamcenter, and shop-floor realities. The takeaway: The real value isn’t a model. It is configuration and customization: grounding in your systems of record, domain ontologies, governed connectors, policy guardrails, eval harnesses tied to domain KPIs, and change management. Off-the-shelf chat interface won’t clear the bar for accuracy, compliance, or UX in complex functions. Verticalization is the on-ramp. Customization is the unlock. #EnterpriseAI 

Most enterprise conversations about what LLMs and agents can or cannot do are still driven by anecdotes: “it worked in my pilot” vs “it hallucinated in my demo.” This new paper forces a cleaner frame: the core gap is not fluency or even accuracy. It’s epistemology, meaning how a system forms, justifies, and regulates “knowledge.” The authors claim LLM can mimic the surface of knowing (answers, explanations, rationales) while lacking the machinery of knowing. That distinction matters because it explains why hallucination is not a corner case. It’s a structural consequence of “always produce the next step” even when the right epistemic move is abstention, escalation, or evidence-seeking. They also point to a workflow-level shift that’s easy to miss: we used to “retrieve then judge.” Generative systems collapse retrieval, selection, and explanation into a single authoritative artifact. That collapse changes user behavior. Evaluation becomes optional by default, and plausibility becomes a substitute for verification. Their term for this failure mode is “Epistemia”i.e. the feeling of knowing without the labor of judgment. The paper then decomposes the human vs machine gap into seven fault lines: 1. Grounding: LLMs can speak as if they have access to the world, but they do not carry commitments to a system-of-record unless you force it through tooling and policy. 2. Parsing: they may miss hidden constraints, because token-level continuity is not the same as constructing a situation model. 3. Experience: embeddings are not lived priors; the model can be statistically “familiar” without having a robust smell test for what’s missing. 4. Motivation: humans choose objectives and tradeoffs; the model cannot decide what should be optimized unless you specify it. 5. Causality: coherent narratives are cheap; causal validity is not. 6. Metacognition: humans can stop; the model is incentivized to continue. 7. Value and accountability: humans own consequences; the model cannot carry responsibility for harm. It is not that LLMs are bad at reasoning. It’s that they are good at producing reasons and bad at the epistemic control loop: grounding, calibration, and accountability. And that leads to the real risk: even correct outputs can degrade decision quality if they bypass the evaluation process. I think there are three main implications for the future of work: 1. Output production gets cheap and evaluation becomes scarce. 2. The org chart shifts from creators to governors. 3. Agents fit best where tasks are bounded and evidence is accessible. The right mental model is a Venn diagram, not replacement. Machines expand throughput and consistency. Humans provide grounding, tradeoffs, counterfactual checks, and ownership. In the overlap you get compounding value: faster cycles without outsourcing judgment. One plus one equals three. Where in your workflows is plausibility becoming the decision criterion, and what gates have you designed to prevent it?

Is code still intellectual property in a world where AI can generate it on demand? The rise of LLMs is transforming the foundation of software development, challenging traditional notions of intellectual property. Historically, IP was straightforward to identify: it was the code itself. Companies invested months or years building proprietary algorithms and implementations that represented their crystallized knowledge of how to solve complex problems. Today, LLMs can generate sophisticated, production-quality code in minutes rather than months. While this code still has inherent value in its utility, its nature as intellectual property has fundamentally changed. When multiple developers can prompt an LLM to generate similar solutions, the code itself no longer represents unique intellectual capital. This capability not only democratizes software development but raises a profound question: If code can be generated so easily, where does the real value in software development now lie? The answer is that intellectual property in the LLM era isn't in the code itself, but in the empirical knowledge of what actually works in practice. While LLMs can generate theoretically perfect code, they can't capture the practical insights that only come from deploying solutions in the real world. Consider a complex integration between enterprise systems - an LLM might generate syntactically correct code, but only practical experience reveals which approaches handle edge cases, maintain performance at scale, and remain maintainable. The true IP becomes this accumulated knowledge of real-world implementation challenges. This shift fundamentally changes how we build software businesses. Success is no longer about accumulating proprietary code, but about systematizing practical knowledge of making LLM-generated solutions work reliably at scale. When an LLM can generate ten different valid approaches to a problem, the competitive advantage lies in knowing which approach will actually succeed in production, how to implement it robustly, and how to maintain it efficiently. This knowledge, unlike code, can't be easily replicated or generated. We're entering an era where the winners won't be those who write the best code—that's becoming commoditized. Instead, they'll be those who most effectively accumulate and apply practical knowledge about making LLM-generated solutions work in production. This represents a profound shift: while LLMs can generate multiple, seemingly valid solutions to any problem, their non-deterministic nature means each generation might produce different code. This uncertainty actually increases the value of implementation knowledge - success requires understanding not just one solution, but the patterns and principles that work reliably across many possible implementations. The true intellectual property is the battle-tested knowledge of how to navigate from AI-generated possibilities to production-ready solutions.