Two announcements from the same day point in the same direction: development environments are becoming workplaces for AI agents, and the model providers want to supply not just tools but the infrastructure beneath them.Google is opening Android Studio to third-party agents, while OpenAI is promoting in Berlin the idea of building products directly on its own platform.

Android Studio opens up to third-party agents

Under the name Bring Your Own Agent, more than 40 coding agents can be run in Android Studio – Google's official development environment for Android apps.A coding agent here is an AI system that not only suggests code but executes multi-step tasks itself: changing files, running commands in the terminal, checking results.

The selection is notable: alongside Google's own tools, Anthropic's Claude Agent and OpenAI's Codex also run in the IDE.Google is thus giving up control over the model while retaining control over the environment.For teams, this means that choosing the agent no longer necessarily dictates the choice of development environment.

On top of that come new terminal tools through which agents complete tasks at the user interface 77 percent faster, according to Google, with up to three times lower token consumption.Tokens are the billing unit for language models; fewer tokens per task means directly lower costs and usually shorter runtimes too.These figures come from the provider itself and relate to a specific type of task – they serve as an indication of direction, not as a guarantee for arbitrary projects.

Google itself sums up the core message of the Android Dev Summit concisely: Android development, too, is becoming agent-centric.Alongside this, new Gemini APIs were presented – interfaces through which apps can integrate Google's models directly.

OpenAI wants to be a platform, not just a model provider

At the DevDay Exchange in Berlin, according to press coverage, OpenAI mainly repeated familiar material but provided a few new details.The AI assistant Dots is to launch more broadly in Europe "in a few weeks" – a phrasing that deliberately leaves open exactly when and to what extent.

The figure more interesting from a client's perspective: weekly Codex users in Germany have increased twentyfold since January.Codex is OpenAI's coding agent which – see above – now also runs in Android Studio.A twentyfold increase says little about absolute adoption without a baseline, but it shows that agentic programming in German development teams is currently moving from a niche into everyday practice.

The actual message of the event was strategic: developers should build their business directly on OpenAI's infrastructure.OpenAI is thus no longer competing only for the best model, but for the role of the base platform – the layer on which products are built and from which they can later be detached only with effort.

Where the two reports fit together – and where they do not

Both developments show the same movement from different directions.Google is opening the tool layer and making agents interchangeable.OpenAI is working to ensure that as much application logic as possible sits on its own infrastructure.For companies, this creates a useful tension: the more open the IDE, the easier it is to switch an agent; the more deeply a product is integrated into a provider platform, the harder.

It should also be noted that the concrete performance figures come only from Google's announcement and relate to UI tasks in Android Studio.For OpenAI there is no comparable efficiency figure, only a usage figure.The two values therefore cannot be read against each other.

What clients should clarify now

  • Separate tool choices: Development environment and agent should be decidable separately.An open IDE preserves this option.
  • Plan for token costs: Agentic workflows generate ongoing costs per task.Efficiency gains in tokens have a direct effect on the budget and belong in project calculations.
  • Assess dependencies: Those who build on a provider platform gain speed and lose freedom to switch.This trade-off should be made consciously and documented, not on the side.
  • Check data flows: When using external agents, source code potentially leaves your own environment.Contractual and technical framework conditions should be clarified before the first pilots.
  • Adapt review processes: Agents generate more code in less time.Review capacity in the team thus becomes the new bottleneck.

Assessment for web and software projects

For web and software projects, the bottleneck is shifting: it is not writing code that becomes expensive, but reviewing it, architectural discipline and the question of whose infrastructure the result runs on.Anyone commissioning development today should find agent use, token budget and platform lock-in explicitly in the proposal rather than as a tacit assumption.

The open approaches – interchangeable agents in a neutral environment – are the lower-risk starting position, because they keep later corrections cheap.

Sources