The past months have shown that generative AI is developing at an unprecedented pace. What were chatbots answering simple FAQs just a few years ago has evolved into systems that plan complex workflows, prepare decisions, and optimize their own results.
From Chatbots to Reasoning Agents
Between 2018 and 2020, rule-based chatbots dominated the market. They were useful but rigid, ideal for standardized customer queries. From 2021, the first systems with natural language processing arrived, able to understand context better.
The real paradigm shift came in 2023 with large language models (LLMs): for the first time, it was possible to proactively plan and independently execute chains of tasks.
In 2024 and 2025, development reached its next stage: multi-agent systems took the stage. Different specialized agents worked together, one analyzing data, another writing reports, a third designing strategies. Today, in summer 2025, we see a new generation of AI models that not only answer but also question, explain, and improve strategically.
The New Generation of Models
The first major step came from OpenAI. With GPT-5, a unified system was introduced that combines a fast chat model, a deep reasoning model, and an intelligent router. The model itself recognizes when it can give short answers and when longer reasoning is needed. Microsoft integrated GPT-5 directly into Copilot as early as August 2025.
Shortly after, Google DeepMind presented Gemini 2.5 Deep Think. It allows several solution paths to be examined in parallel and returned in consolidated form, a publicly available multi-agent system.
The third step came from Anthropic with Claude Opus 4.1. This model was developed specifically for agent workflows and coding in the enterprise and is already integrated into AWS Bedrock and GitHub Copilot. Anthropic simultaneously received 13 billion US dollars in growth capital, underlining the enormous demand for enterprise-ready agent solutions.
Practical Examples: AI in Core Business
These developments are not an end in themselves; they have long since arrived in companies. Microsoft has rolled out the reasoning generation in M365. Meta is working with Reliance on LLaMA-based enterprise solutions for the Indian market.
A particularly vivid example comes from Klarna: the company reports double-digit revenue increases after its AI assistant took over tasks previously handled by more than 700 full-time employees.
The message is clear: AI agents are no longer pilot projects, they are part of core business.
What Does This Mean for Companies?
With the new generation of AI systems, the responsibility for decision-makers also grows. Boards and division heads should now ask themselves three questions:
Which workflows benefit from generative reasoning? The new models realize their potential precisely where planning, tool use, and flexible re-planning are decisive.
How do we handle the diversity of models? GPT-5, Gemini 2.5, and Claude 4.1 are not purely competing products but complement one another. Orchestration and routing matter more than one-sided dependency.
How do we ensure governance and cost control? Reasoning-augmented agents are powerful but also resource-intensive. Companies need clear cost structures, accountabilities, and governance.
The Role of BLACKFIELD AI
Many organizations now face the challenge of deriving scalable strategies from spectacular technological progress. This is where BLACKFIELD AI comes in:
We support the piloting of agentic end-to-end workflows.
We establish governance structures to avoid risks.
We guide the integration of new models into existing IT landscapes.
The result: less pilot theater, more P&L impact. Companies benefit from a fact-based, secure path into the world of agentic AI.
Conclusion and Outlook
The last 100 days have impressively shown that AI is not just developing but transforming rapidly. GPT-5, Gemini 2.5, and Claude 4.1 mark the step from tools to autonomous partners. Those who act now gain not only efficiency but a decisive competitive advantage.
In the next blog in this series, we show which success factors and pitfalls companies must consider to integrate AI sustainably and responsibly.
