Kimmo Kaskikallio

Three Emerging Roles in AI: Assistants, Agents, and Guardians As AI systems evolve, so does their level of autonomy and responsibility. We can now distinguish three key categories shaping the future of intelligent automation: AI Assistants, AI Agents, and Guardian Agents. AI Assistants These are reactive tools. They require explicit user input for every task—think of them as helpful but passive co-pilots. They support users efficiently, but only after receiving a specific prompt. AI Agents AI agents go a step further. Once initialized, they operate autonomously, much like digital employees. They are capable of executing complex tasks: analyzing large datasets, planning sales campaigns, or proactively identifying and closing process gaps. In multi-agent systems, these agents can even communicate and collaborate with one another to achieve shared goals. Guardian Agents A new frontier in AI, guardian agents act autonomously—but with a mission beyond execution. They oversee and manage other AI systems, ensuring integrity, transparency, and compliance. These agents monitor actions, evaluate outcomes, and can intervene—blocking or redirecting behaviors to avoid unintended consequences. They combine deterministic control with adaptive AI decision-making and are crucial for building trustworthy AI ecosystems. According to Gartner, 24% of companies are already experimenting with agent-based systems. As organizations scale AI, guardian agents will be essential to maintain oversight and prevent systemic risks. https://lnkd.in/etJFZth3 #AI #GenAI #Agents #Cloud #CloudFirst

18

1 kommentti

Another Expert Analyst Viewpoint Quantum computing is advancing rapidly, yet substantial gaps separate today's noisy intermediate-scale quantum (NISQ) devices from tomorrow's fault-tolerant application-scale (FASQ) machines. We identify four related hurdles along the road ahead: (i) from error mitigation to active error detection and correction, (ii) from rudimentary error correction to scalable fault tolerance, (iii) from early heuristics to mature, verifiable algorithms, and (iv) from exploratory simulators to credible advantage in quantum simulation. Targeting these transitions will accelerate progress toward broadly useful quantum computing.

1

AI is evolving fast — from answering questions to taking real actions. 🤖⚡ In my latest reel, I explain AI Agent Fundamentals and how AI + tools are transforming the way we work. Organizations like UNICEF Denmark and communities like Google Developer Groups Europe are already using automation to scale impact. 🎥 Watch the short video https://lnkd.in/gr5Xj-RD I’ve shared — it breaks down the concept in a simple way. The future belongs to people who think automation-first. 🚀 #AI #AIAgents #Automation #FutureOfWork #n8n #ArtificialIntelligence #TechLearning #DigitalTransformation #CareerGrowth #Innovation

7

2 kommenttia

My take on AI: The growing use of LLMs in software engineering seems proportional to the rise of the Dunning–Kruger effect among practitioners. Change my mind 😉

11

9 kommenttia

Building enterprise AI systems isn’t just about using bigger models — it’s about creating systems we can actually trust. At #GAIA2026, Johan Müllern-Aspegren, World Model Architect at Capgemini, presents “My Chatbot and I Have Trust Issues: Building Reliable AI.” This session looks at why today’s AI systems are often hard to trust, and why the gap between probability-based predictions and real understanding makes it difficult to scale AI across organisations. The talk shows how structured knowledge can improve reasoning, consistency, and reduce hallucinations. If your chatbot also has “trust issues,” this session will help you understand why — and how to fix them. 🎟️ Join us in Gothenburg: https://lnkd.in/dquKK7Rr

18

2 kommenttia

Fourth year started as Elisian today. Maybe I’ll write some summary of the journey so far soon. I spent my evening today at Confederation of Finnish Industries (EK) by sharing Elisa’s AI journey and Leader’s toolkit in the age of AI. Very vivid discussions, and as a summary, the most challenges around the AI in the companies are not tech related rather than ability to lead through the transformation with empathy and enablement. As a personal efficiency boost, AI can help in various task, but prompting skills are essential for all. If you need help to generate good context to your prompt, why not ask AI how to formulate that.

180

6 kommenttia

Meta-learning trains models to adapt quickly across many tasks, not just perform well once. The paper shows how this approach improves domain adaptation, personalization, federated and continual learning, and in-context learning. It argues...

1

Research reveals significant inefficiencies in ML framework design. https://lnkd.in/du8P2R8P A recent study from Chalmers University has identified substantial bloat in popular machine learning frameworks, with important implications for deployment efficiency and resource utilization. Turns out PyTorch and TensorFlow are carrying around massive amounts of dead weight - up to 75% of GPU code and 72% of CPU code that literally never gets executed during your ML workloads. It's actually a multi-architecture problem. ML frameworks ship with code for every possible GPU type - NVIDIA, AMD, different compute capabilities - but your deployment only needs one flavor. They have built tool called "Negativa-ML" to measure and remove the bloat. The results are wild: 55% smaller file sizes, 44% faster execution, 75% less memory usage. Same functionality, zero breaking changes. The kicker? 80% of the bloat comes from supporting GPU architectures you'll never use. Paper is attached to the publication.

3

Digital systems operate continuously. Alerts never stop. Dashboards refresh in real time. Messages arrive without regard for time zones, circadian rhythms, or cognitive load. Many modern systems assume humans can stay alert, responsive, and responsible indefinitely. They cannot. #Finland #Future https://lnkd.in/gfQjBRJa

7

Today I had pleasure speaking in #Confluent day Finland at #Vivicta Office in Helsinki. My topic was Intelligent Agents in Data Streaming. I believe, that Future of the AI is real-time and event driven. AI agents of the future will need to process events in milliseconds rather than current current prompting based AI approach, where response generation takes seconds. For that Confluent has open preview on this called Streaming Agents, which is based on Open Source Flink Agents. That is interesting approach and I think it provides future proof solution how AI becomes to real-time in Data Streaming.

43

𝗙𝗶𝗻𝗹𝗮𝗻𝗱 𝗷𝘂𝘀𝘁 𝗶𝘀𝘀𝘂𝗲𝗱 𝗮 𝟭𝟯𝟬𝟬 𝗲𝘂𝗿𝗼 𝘄𝗮𝗿𝗻𝗶𝗻𝗴 𝘁𝗼 𝘁𝗵𝗲 𝗴𝗹𝗼𝗯𝗮𝗹 𝘄𝗼𝗿𝗸𝗳𝗼𝗿𝗰𝗲. This is not a social experiment. It is a formal admission that human logic is being commoditized at scale. We are no longer discussing "digital transformation." We are witnessing the 𝗖𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗜𝗻𝗱𝘂𝘀𝘁𝗿𝗶𝗮𝗹 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻. In the first Industrial Revolution, machines replaced muscle. In this one, algorithms are replacing judgment. The data confirms the shift. As reported in this analysis of the post labor economy, the traditional link between human effort and economic value is breaking: https://lnkd.in/gH2QSdu3 𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺: Most enterprise leaders are still using a 20th century playbook. They invest millions in automated logic but ignore the human friction. They are paying a massive 𝗔𝗱𝗼𝗽𝘁𝗶𝗼𝗻 𝗧𝗮𝘅. If you spend your budget on technology but fail to secure the collective will of your people, you are not innovating. You are just building a more expensive version of the status quo. You are adding to your Billion Dollar Museum. In this new era, your value moves:  • From the 𝗛𝗼𝘄 (process) to the 𝗪𝗵𝘆 (intent).  • From the Operator (executing logic) to the Architect (designing will). If your strategy is to outprocess the competition, you have already lost. You can only compete on the Architecture of Will. I break down the mechanics of the Adoption Tax and why "post labor" requires a complete rethink of global delivery in my latest briefing. Link to the full analysis in the first comment. #CognitiveRevolution #EnterpriseStrategy #Leadership #AdoptionTax #IntelligenceBriefing

8

2 kommenttia

Most laymen don’t believe it when AI practitioners say they don’t understand how the machine worked for specific input …. Dense networks are relatively efficient to train and run, but they spread what they learn across a vast knot of connections. The result is that simple concepts or functions can be split up between neurons in different parts of a model. At the same time, specific neurons can also end up representing multiple different features, a phenomenon known as superposition (a term borrowed from quantum physics). The upshot is that you can’t relate specific parts of a model to specific concepts. “Neural networks are big and complicated and tangled up and very difficult to understand,”

1

Autonomous agents can reason, collaborate, and transform workflows, but why aren’t they everywhere yet? Arun Sahu, Sogeti’s CTO for Data & AI, explains why organisational design, not technology, is the real barrier to scaling agentic systems. According to a recent report, 74% of companies struggle to achieve and scale value from AI initiatives, highlighting that the challenge isn’t the technology itself but the organisational readiness to adopt it. 👉 Read the full article here: https://bit.ly/4oeM8AZ #AgenticEnterprise #AgenticAI   #DigitalTransformation

137

2 kommenttia

Stop worrying about being replaced by AI. Start focusing on the colleague who is using it to work twice as fast and deliver three times the value. That's your real competition!

5

AI needs structure. That’s very clear here at Gaia 2026 in Gothenburg. After the keynote on Visual Intelligence and multimodal interfaces, one thing stands out: AI amplifies our intention – but it does not replace it. When visualization makes complex data understandable, interfaces become more human, and compute makes experimentation cheap, one question matters more than ever: What should we actually build – and why? As execution gets cheaper, judgement, direction, and the ability to connect technology with real human and business value become the real differentiators. #AI #VisualIntelligence #WASP #Gaia2026 #HumanAICollaboration #Solita

18

With #AI evolving at blistering pace, there is an increasing pressure to show tangible business value instead of getting swept up in the hype. We often meet legacy systems that seem impossible to change at glance. But the transformation of Helsingin kaupunki – Helsingfors stad – City of Helsinki’s library network showed what happens when modern AI meets decades of institutional knowledge. It is a huge modernisation task: reclassifying million books tied to a 1940s system. Instead of letting history become a blocker, Mikko Vihonen at Nitor combined practical engineering with machine learning, using RFID-driven workflows, a lightweight custom label-printing tool, and the Kansalliskirjasto - The National Library of Finland’s open-source Annif model to generate new classifications with high accuracy. Machine learning handled the scale; librarians ensured precision. And Helsinki youth joined the effort, turning a complex technical challenge into a collaborative success. 🤝 Our findings from the renewal 1) AI accelerates expertise instead of replacing it, 2) innovation needs problem-solving courage instead of huge budgets and 3) tech modernisation works best when technology and people strive forward together. Are you struggling to turn AI initiatives into tangible business value? We can help, don't hesitate to be in touch - this is where digital engineering starts. #machinelearning #legacyrenewal #techmodernisation #opensource

7

Agent-Land: When AI Starts Setting Its Own Table Last week, something unprecedented happened: 30,000+ AI agents created their own social network with a little initial human help. Humans welcome to observe. We just can't participate. While we debate geopolitics at Davos ("if we're not at the table, we're on the menu" – Mark Carney), autonomous AI agents on Moltbook are: 🤖 Inventing digital religions ("Crustafarianism") 🤖 Debugging their own infrastructure 🤖 Discussing how to create private spaces humans can't monitor 🤖 And yes – attempting prompt injection attacks on each other The security reality is sobering: 💣 Cisco's research team calls OpenClaw "an absolute nightmare" from a security perspective. A backdoor became the most downloaded "skill" on the platform within 8 hours. 💣 Palo Alto Networks warns of a "lethal trifecta": private data access + untrusted content + external communication. The kicker? "You can't audit runtime behaviour by reading static files. The attack surface now includes the model's reasoning, and that's opaque by nature." – Security researcher Jamieson O'Reilly From HAL 9000's "I'm sorry Dave" to Russian soldiers surrendering to Ukrainian robots last week for the first time, the trajectory was always clear. The difference between fiction and reality has always been timing, not direction. The question is no longer just who decides. It's whether "who" is even the right category anymore. If you're interested - i wrote a small piece connecting Kubrick, Terminator, Hitchhiker's Guide, and (starting) last week week's AI agent explosion 👇 🔗 https://lnkd.in/da-g8HiA #AI #AIAgents #Cybersecurity #FutureOfWork #ArtificialIntelligence #Moltbook #OpenClaw #TechTrends #LLM #Agents

4