Between AI hype and real added value – Google I/O 2025 decoded

The AI technology landscape is evolving rapidly. New tools, models and updates are constantly changing the way we work and develop innovative solutions. Given this dynamic environment, it is challenging to maintain an overview and accurately assess the relevance of new technologies for specific projects. This became clear once again at Google I/O 2025. In this blog post, I have filtered out the most strategically important insights for you, summarised and classified key innovations, and provided an initial assessment and recommendations on how you can leverage this potential.

Gemini at the centre – the latest developments

At Google I/O 2025, Gemini's central role in Google's AI strategy was reaffirmed. The announced innovations will have a direct impact on the development opportunities and use of AI in projects.

• Gemini 2.5 Pro and Flash – Specific strengths for diverse use cases: The Gemini model family has been further developed. Gemini 2.5 Pro is now integrated as standard in Gemini Advanced, Google's premium subscription for access to the most powerful AI models. It features an advanced thinking mode called ‘Deep Think’. This is designed specifically for complex tasks such as sophisticated programming logic or detailed data analysis. At the same time, Gemini 2.5 Flash has been optimised as an efficient and cost-effective variant. It offers improved logical reasoning, multimodal processing and context understanding of over one million tokens. Developers now have a wider choice when selecting the right model for specific AI-powered applications, from resource-efficient on-device applications to highly complex reasoning tasks.

• Gemini Live – Natürlichere Interaktion: Deutlich verbesserte Konversationen durch native Audio-Ausgabe direkt aus dem Modell, was natürlichere Dialoge mit steuerbarer Tonalität (über 24 Sprachen) ermöglicht. Features wie die Emotionserkennung („Affective Dialogue”) und das Ignorieren von Hintergrundgeräuschen („Proactive Audio”) eröffnen zusammen mit Kamera- und Bildschirmfreigaben neue Dimensionen für interaktive Anwendungen wie Support oder Visualisierungen.
• Vertiefte Integration und erweiterte APIs: Gemini wird weiterhin tief in Google-Dienste wie Workspace oder Android integriert, um KI-Funktionen einer breiteren Masse zugänglich zu machen. Für Entwicklerinnen und Entwickler sind insbesondere die erweiterten Gemini APIs und die Vertex AI-Integration von Bedeutung. Diese ermöglichen es, die neuesten Modellfähigkeiten direkt in eigene Anwendungen und Arbeitsabläufe einzubinden und Googles zentrale KI-Plattform zu nutzen. Die Updates zielen somit auf leistungsfähigere Modelle und deren einfachere Implementierung in professionellen Anwendungen ab.

The Gemini 2.5 model family also introduces other important new features and improvements for developers.

• Optimised thought processes and cost control: Thought Summaries in the Gemini API and Vertex AI make the models' thought processes more transparent. At the same time, Thinking Budgets, now also available for Gemini 2.5 Pro, enable more precise control of latency and costs for complex queries.
• Enhanced security architecture and open source connectivity: Google has significantly strengthened the security mechanisms of the Gemini 2.5 family, particularly against indirect prompt injections when using tools. Native SDK support for the Model Context Protocol (MCP) also facilitates integration with open source tools.
• Preparation for agent-based computer interaction: The capabilities of ‘Project Mariner,’ which enable models to use and control computer applications, are being gradually integrated into the Gemini API and Vertex AI. This will begin in the summer with experimental capabilities.

Boost for developers – new tools and platform power

At Google I/O 2025, in addition to advances in Gemini models, the focus was on tools and platforms that help developers leverage this AI potential.

• Vertex AI – Further developments of the central AI platform: In addition to the integration of the latest Gemini 2.5 models, generative AI media models such as Veo 3 (video and audio) and Imagen 4 (images up to 2K) were introduced. This opens up expanded possibilities for multimodal AI applications. The Agent Development Kit (ADK) and the new Agent Engine UI in the Google Cloud Console are also designed to make it easier to create and manage AI agents.
• Firebase Studio & Firebase AI Logic – support for AI-powered app development: New features for app development were introduced in Firebase. Firebase Studio (with Gemini 2.5 integration) is a cloud-based work environment that supports the creation of apps using prompts or code. Firebase AI Logic integrates AI tools, frameworks (such as Genkit) and SDKs to embed Gemini and other models into apps.
• Android Studio with Gemini – integration of AI features into the IDE: The integration of Gemini 2.5 Pro into Android Studio aims to support developers in various phases of app development. An announced ‘Agent Mode’ will autonomously handle more complex, multi-step development tasks in the future. Features such as the generation of Compose Previews or ‘Journeys’ (test automation via natural language) can optimise development cycles.
• Gemini Code Assist – Code support more widely available: Code Assist (generally for individuals/GitHub), based on Gemini 2.5, extends pure code generation with a ‘mixture of agents’ approach. The large context window with a capacity of one million tokens enables a deep understanding of extensive code bases.

A personal highlight: the next generation of AI video production with Veo 3 and Flow

There was particular excitement about the introduction of Veo 3, Google's most powerful model for video creation, and the AI filmmaking tool Flow.

Veo 3 sets new standards in video generation by not only delivering visually stunning results, but also generating audio natively within the model (sound effects, dialogue). This enriches immersion and storytelling. Flow is a platform for creatives that enables intuitive exploration of ideas with Veo, Imagen and Gemini. It offers professional features such as precise camera control, a scene builder for consistent character editing and asset management.

Facts and figures from Google I/O 2025

The figures presented at Google I/O 2025 provide insight into the current scale and pace of development in the AI field at Google:

• Context processing: Gemini 2.5 Pro and Flash with a context window of one million tokens.
• Use of the Gemini API: Over seven million developers.
• Growth on Vertex AI: 40-fold increase in Gemini usage compared to the previous year.
• Distribution in applications: Google Gemini app: More than 400 million MAU; AI Overviews: over 1.5 billion monthly users.
• Number of tokens processed: 480 trillion tokens per month (~50-fold increase per year).

The significance of these innovations for your projects

Google I/O 2025 shows that AI, especially Gemini and smart developer tools, is becoming the core of modern software development. These innovations accelerate processes, enable new user experiences and solve complex challenges. For companies, engaging with these technologies is therefore crucial for their competitiveness and the development of future digital solutions. It is no longer a question of whether, but how to use them effectively to create real added value.

From announcement to application: adesso as your partner

The wealth of innovations unveiled at Google I/O 2025 opens up impressive potential. But how can you successfully transfer these innovations into your projects without getting lost in the complexity? Evaluating, integrating and utilising new AI and cloud technologies requires expertise and experience.

As a Google Cloud Premier Partner with a dedicated Competence Centre for Google Cloud, we support you in:

• evaluating the potential of the latest Gemini models for your specific use cases and developing prototypes, as well as
• designing, developing and operating complex AI-powered solutions on the Google Cloud platform.