Kategorie: Artificial Intelligence

While many AI companies mainly talk about benchmarks, model sizes and chatbots, is pursuing a far more radical approach: the startup Andon Labs is deploying AI agents into real economic environments with real budgets, real people, real contracts and real consequences.

It sounds like an internet joke, but it was a real issue: in newer versions of ChatGPT, references to goblins, gremlins and similar fantasy creatures began appearing with unusual frequency – even in entirely serious contexts. What initially looked like a quirky glitch turned out, on closer inspection, to be a revealing case study in how modern AI systems behave.

As agentic AI systems become more capable, a subtle but important shift is taking place. The focus is moving away from the model itself and towards the environment in which it operates....

ChatGPT 5.5: the shift from answer engine to work engine With GPT-5.5, OpenAI moves ChatGPT further away from the classic chatbot model and closer to a system that can carry out real work over extended sequences. Released in April 2026 for ChatGPT and Codex, the model is currently available to paid users across Plus, Pro, Business and Enterprise tiers. OpenAI positions it as a new class of AI built for practical tasks, with a clear focus on coding, research, data analysis, documents, spreadsheets and multi-step tool workflows. The real leap is not in conversation GPT-5.5 is not primarily designed to be a better conversationalist. Its progress lies in how it handles complexity: understanding messy inputs earlier, planning tasks more independently, using tools more deliberately, checking intermediate results and persisting longer until a task is complete. The difference from earlier models becomes obvious in execution. Instead of simply responding, GPT-5.5 breaks problems into steps, works through them, evaluates outcomes and adjusts when things go wrong. This marks a clear transition from a system that explains work to one that actively performs it. Coding becomes the central proving ground This shift is most visible in software development. GPT-5.5 is significantly more capable of handling end-to-end coding tasks, including architecture planning, implementation, testing and debugging. It maintains coherence over longer workflows and is less likely to stall when encountering issues. The improvement is less about individual responses and more about continuity. A high-level instruction can now lead to a structured project, with the model iterating through errors and refining its output along the way. At the same time, efficiency improves: many tasks require fewer tokens than before, as the model operates more directly and with less redundancy. Computer use becomes more mature Another major focus is interaction with real computing environments. GPT-5.5 is designed to work across tools, navigate websites, process data and coordinate tasks within digital systems. It can operate within browsers, complete forms, gather information and interact with local or cloud-based environments. It also shows stronger capabilities in interpreting visual inputs such as screenshots and translating them into actions. This brings the model closer to functioning as an active participant in everyday digital workflows rather than a passive generator of content. Knowledge work becomes multi-step For traditional knowledge work, GPT-5.5 represents a clear step forward. It handles large volumes of information more effectively, identifies connections, and produces structured outputs with greater consistency. The key difference lies in process handling. Instead of answering isolated questions, GPT-5.5 supports the full journey from raw material to finished result. It can research, organise, evaluate and synthesise information into reports, analyses or decision-ready documents. Variants, usage and cost structure Within ChatGPT, GPT-5.5 is available in a Thinking mode for more complex tasks, alongside a Pro variant for particularly demanding workflows. In Codex, it is used for agent-like tasks such as software development and automation. The context window has been significantly expanded, reaching into the million-token range in certain configurations. At the same time, latency remains broadly comparable to the previous generation, while overall efficiency per task improves. In the API, GPT-5.5 sits in the upper pricing tier. However, actual costs depend heavily on how efficiently the model completes tasks. Because it often requires fewer tokens to reach a result, overall costs can remain stable or even decrease in practice. Safety becomes central As capability increases, so does risk. GPT-5.5 therefore comes with stronger safety mechanisms, particularly around areas such as cybersecurity, exploits and misuse of automation. The model is more cautious in sensitive domains and more restrictive in potentially harmful scenarios. For organisations, this shifts the focus towards governance. The more autonomous the system becomes, the more important it is to define access controls, approval processes and monitoring. The value of the model is no longer just in its intelligence, but in how safely it can be deployed. The underlying paradigm shift GPT-5.5 is not simply a model that produces better text. It represents another step towards systems that can carry out work. The key change is that tasks are no longer single prompts, but extended workflows involving planning, execution, verification and iteration. This changes the role of ChatGPT itself. It becomes less of an interface for answers and more of an environment for digital work. For developers, analysts, consultants and knowledge workers, that is the real shift. Not every response is more impressive. But more tasks are completed end to end. And that is ultimately what defines GPT-5.5: a move from assisting work to actively getting it done.

With Claude Design, Anthropic is trying to solve a problem that many AI tools have so far struggled with: moving beyond generating visually appealing one-off outputs towards something that actually supports a real design workflow. The new product from Anthropic Labs is neither a classic image generator nor a simple prompt-driven toy. Instead, it is a browser-based environment that combines chat, a visual workspace and a pathway into production-ready development. Anthropic positions it as a research preview for Pro, Max, Team and Enterprise users

Autonomous AI agents are widely seen as the next evolution of software. They plan, decide and increasingly act on their own. That is precisely what makes them so powerful and, at the same time, so risky. The more autonomy a system is given, the closer it gets to a point where control is no longer guaranteed. The so-called “Rule of Two”, a security principle introduced by Meta, is a direct response to this tension. It is neither a complex framework nor a new technology, but a deliberately simple rule addressing a fundamental issue: the concentration of power within a single system.

At CHI 2026 in Barcelona, Apple is not showcasing major product launches, but something arguably more significant: a set of research contributions that offer a rare glimpse into how the company is thinking about the future of interfaces, accessibility and data-driven design.

Generative AI is not just transforming processes, products or business models. It is going deeper — into the core of organisations: leadership, accountability and structure. What once felt relatively stable is now being redefined. New roles are emerging, organisational charts are becoming more fluid, and leadership itself is undergoing a fundamental shift.