The Week in Cloud & AI: The Race for AI Context Begins

The Week in Cloud & AI

This week, the AI landscape shifted from raw model capability to practical integration. Databricks introduced a unified context layer to connect AI with business knowledge, while Google advanced its AI agents with a new memory feature. In foundational tech, Google Cloud added post-quantum cryptography to its Key Management Service, and AWS automated log storage optimization to reduce costs.

Top Stories

Databricks Introduces Unified Context Layer for Enterprise AI

Databricks unveiled Genie One, an "AI coworker," and Genie Ontology, a framework that provides a unified context layer for enterprise AI. The Ontology maps business terms, metrics, and relationships, allowing AI agents to understand business concepts and operate consistently, tackling the common problem of AI assistants lacking business-specific knowledge.

Why it matters: This directly addresses a major blocker to AI adoption: the lack of business context. By creating a semantic layer between data and AI, Databricks is betting that the next wave of value will come from deep integration, not just powerful models, positioning the data platform as the central nervous system for enterprise AI.

Key Takeaways:

• Genie One is an AI assistant that uses a new unified context layer.
• Genie Ontology maps business logic, metrics, and relationships for the AI to use.
• The goal is to provide grounded, trustworthy answers and enable AI-driven actions.
• This approach aims to make AI agents more reliable and aware of business specifics.

Who should care? AI Engineers, Data Engineers, Engineering Managers, Technical Architects Impact: High

Gemini Enterprise Agents Get a Memory Boost

Google Cloud's Feature Memory Bank for Gemini Enterprise Agents is now generally available. It provides agents with a structured, low-latency memory system to recall information without costly search operations, which is crucial for building stateful agents that maintain context over complex tasks.

Why it matters: Agent memory is foundational for moving beyond chatbots to sophisticated AI assistants. A dedicated, structured memory system is a major improvement over stuffing context into a prompt, enabling developers to build more capable, persistent, and efficient AI agents for real-world business processes.

Key Takeaways:

• Feature Memory Bank provides structured, low-latency memory for AI agents.
• It allows agents to access evolving data without repeated, costly search operations.
• The feature is now generally available for production use cases.
• This is a key building block for creating more stateful and capable AI agents.

Who should care? AI Engineers, Software Engineers Impact: High

Google Cloud KMS Prepares for a Post-Quantum World

Google Cloud Key Management Service (KMS) now supports post-quantum cryptography (PQC) signing algorithms, making them generally available. This adds support for algorithms like SLH-DSA-SHA2 and ML-DSA, designed to resist attacks from future quantum computers and ensure long-term data integrity.

Why it matters: The "harvest now, decrypt later" threat is real: adversaries can capture encrypted data today to decrypt with future quantum computers. By making PQC algorithms available, Google Cloud lets organizations proactively protect high-value, long-lifespan data-a foundational security measure for engineering and security teams.

Key Takeaways:

• Google Cloud KMS now supports post-quantum cryptography (PQC) signing algorithms.
• This helps protect against future threats from quantum computers.
• The feature is now generally available for all KMS users.
• This is a critical step for long-term data security and integrity.

Who should care? Security Engineers, Platform Engineers, Technical Architects Impact: High

AWS Automates Log Storage with CloudWatch Intelligent Tiering

Amazon CloudWatch Logs now offers intelligent tiering, automatically moving log data between storage tiers (Standard, Infrequent Access, etc.) to optimize costs. It automates cost management for long-term retention while ensuring all logs remain queryable through a single interface, regardless of tier.

Why it matters: Managing observability data costs is a significant operational burden. Intelligent tiering removes this toil by offering a "set and forget" solution for cost savings with no performance trade-off. It simplifies compliance and forensics by making it affordable to retain logs longer with less manual overhead.

Key Takeaways:

• CloudWatch Logs automatically moves data to cost-effective tiers based on access.
• It aims to reduce storage costs for long-term log retention.
• All logs remain queryable from a unified view, regardless of tier.
• The feature can be enabled at the account level, simplifying configuration.

Who should care? Platform Engineers, DevOps Engineers, Security Engineers, Engineering Managers Impact: High

Breaking Changes

• Platform: Databricks SDKs (Go, Java, Python)

• Breaking change: Various Databricks SDKs introduced breaking changes. The Java and Go SDKs changed the InternalId field for IAM entities to a string. The Go and Java SDKs also removed the CodeSourcePath field from AiRuntimeTask in the jobs service.

• Who is affected: Developers using these specific fields in the Databricks Go, Java, or Python SDKs.

• Migration recommendation: Review the SDK release notes and update code to accommodate the type changes and field removals.

• Potential risk: Medium. Failure to update will result in compilation errors or runtime failures when interacting with Databricks APIs.

• Platform: GCP

• Breaking change: The gcloud storage rsync command now decompresses gzip files by default during synchronization. Previously, it transferred them as-is.

• Who is affected: Users of gcloud storage rsync who rely on the previous behavior of not decompressing gzipped files.

• Migration recommendation: If the previous behavior is desired, users must now explicitly use a new flag to disable decompression.

• Potential risk: Medium. Scripts or workflows could fail if they expect gzipped files at the destination.

Attention Required

• OpenAI Model Deprecations: Several OpenAI models, including gpt-5-chat-latest and gpt-5-codex, are scheduled for end-of-life on July 23, 2026. Users should migrate to newer models. (Source)

• Gemini Model Deprecations: The gemini-embedding-001 model reached its end-of-life on July 14, 2026. Additionally, the gemini-2.5-pro, gemini-2.5-flash, and gemini-2.5-flash-lite models are scheduled for end-of-life on October 16, 2026. (Source)

• GitLab Version EOL: GitLab version 18.11 reached its end-of-life on July 16, 2026. Users on this version will no longer receive security updates or fixes and should upgrade immediately. (Source)

ReleaseBytes Insights

The enterprise AI narrative is maturing from powerful models to reliable systems. This week's announcements-Databricks' context layer and Gemini's agent memory-show that a model alone isn't a product. An effective AI system requires business context and persistent memory. This shift positions data platforms as essential context providers, while cloud providers race to offer the building blocks to construct these systems. The conversation is shifting from "which model is best?" to "which platform provides the best ecosystem for building, governing, and securing production-grade AI agents?"

If You Only Read One Thing...

Read about Databricks' introduction of a unified context layer for AI. It's the week's most significant move, shifting focus from model intelligence to business awareness. By creating a central "ontology" to map business logic for AI agents, Databricks is framing the core problem enterprises face: making AI understand the business. This could define a new architectural layer in the enterprise AI stack.

By the Numbers

• 264 releases analysed (Jul 13 - Jul 19, 2026)
• 38 general-availability releases
• 24 deprecations / retirements
• 13 security updates
• 6 breaking changes

---

Thank you for reading this edition of the ReleaseBytes Weekly Newsletter. To stay up-to-date with all 264 releases from this week, visit the ReleaseBytes feed. Share this newsletter with your colleagues and encourage them to subscribe.