01
Google introduces Gemini agent for cross-platform enterprise work
Google Cloud announced a Gemini agent that connects organizational data and business tools for multi-step work. Google says it can create task-specific sub-agents and select models for individual tasks.
- Prioritize a bounded pilot across systems such as SAP, Salesforce or ServiceNow to test whether the agent preserves existing access restrictions.
- Require demonstrations of identity controls, action logs and approval steps before granting agents permission to change business records.
- Benchmark model-routing costs and confirm EU processing arrangements before comparing the service with existing enterprise assistant contracts.
02
SAP announces agents across core business processes
SAP announced AI assistants, agents and autonomous capabilities across finance, procurement, HR, customer experience and supply chain at SAP Connect 2026. It also introduced Joule Work as a unified interface for SAP AI agents.
- Map the announced agents to existing ERP automation plans before procuring a separate orchestration platform for the same processes.
- Verify Czech-language support, release dates, licensing and compatibility with your SAP deployment before allocating implementation budgets.
- Keep payment execution and other consequential actions behind explicit approvals and segregation-of-duties controls during pilots.
03
AWS previews managed agents and highlights multi-agent workflows
AWS reported that Amazon Bedrock Managed Agents powered by OpenAI had entered public preview with AWS-native identity and governance integration. Its weekly update also highlighted the AWS Well-Architected Agent and Kiro multi-agent workflows.
- Use sandbox AWS accounts to test agent permissions against existing identity policies before exposing production resources.
- Confirm supported EU regions, data handling and preview service terms before introducing customer or employee data.
- Evaluate architecture reviews and software-delivery tasks first, with human approval for infrastructure changes and code deployment.
04
Mistral previews Large 4, with model weights still pending
Mistral announced a public API preview of Large 4, a natively multimodal mixture-of-experts model with one trillion total parameters and 52 billion active parameters. The company plans to release weights by the end of October.
- Use the API preview to benchmark Czech-language documents, coding tasks and domain accuracy before committing to a migration.
- Defer private-deployment commitments until Mistral releases the weights, licence terms and operational requirements.
- Assess the planned European deployment through contractual processing locations, subprocessors and access controls rather than treating European operation as proof of GDPR compliance.
05
Google expands SynthID Detector globally in English
Google made SynthID Detector available globally in English to check images, video and audio for watermarks from Google or partner AI systems.
- Add the detector to communications and fraud-response checks for media that may originate from supported AI systems.
- Do not treat a negative result as proof of authenticity because the detector does not establish the provenance of all AI-generated media.
- Check upload and data-retention terms before submitting confidential corporate recordings, customer images or employee audio.