November 2026 Summaries
4 posts from Redpanda
Filter
Month:
Year:
Post Summaries
Back to Blog
No summary generated yet.
Nov 16, 2026
1,124 words in the original blog post.
Redpanda introduced R2, a new C++23 coroutine-based runtime intended to become the shared foundation for Redpanda Streaming, Redpanda SQL, and future products beginning with version 26.3.1. The company says R2 combines deterministic simulation testing, TLA+ concurrency specifications, Lean proofs for algorithms, explicit error handling, and hardware acceleration support to improve reliability while enabling faster development, reporting that these checks have prevented roughly 40 faults from reaching its main branch. Its architecture uses one reactor per CPU core, pluggable I/O completion mechanisms such as Linux io_uring and macOS kqueue, deterministic virtualized testing, and mimalloc-based memory management rather than statically partitioned per-core memory. Redpanda also stated that Streaming and its Operator will no longer be source-available from version 26.3.1, while a free Community Edition will remain available for organizations using up to 50 cores. For Redpanda SQL, formerly Oxla, the company reports that porting its query engine to R2 substantially reduced thread counts in benchmark tests and improved cold-scan performance, although some warm-query workloads showed limited gains and higher peak memory use.
Nov 12, 2026
1,676 words in the original blog post.
Redpanda Agentic Data Plane uses attribute-based access control and Cedar policies to help enterprises operate a shared agent platform while preserving team-level autonomy, security, and governance. The proposed model relies on resource tags such as team, stage, and shared, combined with identity-provider group membership, to define ownership, development and production status, and access to common LLM providers. Teams receive permissions based on personas such as operators, builders, and consumers, allowing builders to manage their own agents and MCP servers while restricting policy administration and cross-team access to the platform team. Default-deny authorization and targeted forbid rules prevent use of untagged agents, keep development agents private to their creators, and stop teams from changing resource ownership tags. This approach aims to remain granular across teams and environments, flexible enough to accommodate new agents through tag changes rather than policy rewrites, and scalable because policy growth depends on teams and personas rather than the number of agents or resources.
Nov 09, 2026
2,772 words in the original blog post.
Agentic AI governance is presented as a central requirement for moving enterprise agents from pilots into production, with the primary concern being whether organizations retain control beyond a vendor’s proprietary platform. The passage argues that vendor-specific governance can create lock-in when policies, evidence, and controls cannot operate across different clouds, frameworks, models, or acquired systems, and advocates open standards such as MCP, OAuth, A2A, and OpenTelemetry to support portability. It recommends that buyers assess whether governance applies across platforms, remains within private networks, carries policies and evidence through migrations, enforces controls independently of agent cooperation, and permits customers to retain governance data upon leaving a vendor. The discussion warns that credentials, prompt-based rules, and model-based guardrails may be insufficient because compromised or manipulated agents can bypass or influence them. Instead, it describes a production-ready governance platform as using out-of-band controls, gateways for tool, data, and model calls, per-agent identities and budgets, human approval for consequential actions, immutable audit evidence, and centralized visibility into cost, performance, governance, and incidents.
Nov 03, 2026
930 words in the original blog post.