platformchangelog.dev
43 tools

Platform Changelog: what shipped in platform engineering, developer experience and DevOps

feedsince 2026-10-20all releases, patches included
CNCF: The boundary problem in agentic AI is a platform problempostThe post argues that the risk of autonomous agents comes from tools and credentials that accumulate beyond what anyone decided, and that instructions alone cannot hold an agent to its scope. It proposes a Safe Operating Model with concentric rings: verified workload identity, declared service-mesh reach, admission-controlled and per-call capability checks, and data-plane authorization keyed to that identity. The agent's own identity is denied rights to modify the objects that define its authority, and changes go through reviewed version-controlled configuration. The source text is cut off before the concrete integration work it refers to.platform teamsPlatform teams would enforce agent boundaries through workload identity, mesh authorization, admission policy on tool bindings, and per-invocation gateway checks, with agent authority changes made only through reviewed config. The article is a conceptual model and names no specific tools or releases in the text provided.CNCF Blog ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
The New Stack: Why MCP servers should tell agents what to do nextpostAWS developer advocate James Ward argues that MCP servers should return guidance with data, telling the agent what it just did, what it might do next, and where to find more. He describes AWS adding a session ID parameter to tools so individual calls can be traced as one journey, with a second parameter carrying the user's goal for evaluation. Ward also said AWS ships a CLI and skills alongside MCP because MCP's OAuth model cannot switch between AWS profiles the way the CLI can.platform teamsPlatform teams exposing MCP servers can add session and goal parameters to tools and build evals that compare user intent with agent paths, then adjust tool descriptions or batch tools based on observed patterns. The article presents these as AWS practices rather than a released standard, so no protocol change is implied.The New Stack ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
The New Stack: The 5 essential layers of AI agent sandboxing on KubernetespostThe article lays out five layers for isolating AI agent workloads on Kubernetes: the isolation runtime (runc, gVisor, or Kata Containers), node configuration, access controls on network, credentials and storage, RuntimeClass selection in the pod request, and the Agent Sandbox project for session management. It gives a RuntimeClass and Pod example and notes that runtime overhead should be measured rather than copied from examples. The piece is written from Oracle Cloud Infrastructure's perspective and describes community projects as the basis for its approach.platform teamsPlatform teams running agents on shared clusters need to choose a sandbox runtime, configure nodes for it, and add its measured overhead to RuntimeClass so scheduling and quotas account for it. Egress default-deny, short-lived task-scoped credentials, and separate authorization for MCP servers and APIs outside the sandbox also fall to the platform team.The New Stack ↗Kubernetes MCP Serverwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
InfoQ: Datadog Builds a Specialist AI Model for SRE InvestigationsannouncementDatadog has released Scout, a model trained with reinforcement learning to investigate production alerts inside its Bits Investigation AI SRE product. It is in closed preview with Datadog engineers and a small number of customers. Datadog says Scout matched larger models on its own AlertEval benchmark at a lower estimated inference cost, and resolved 71 to 80 of 90 faults on the public SREGym set depending on effort level.platform teamsPlatform and SRE teams evaluating AI incident investigation should note that the cost and benchmark figures come from Datadog's own tests and no prices are published, so independent results are needed. Scout only diagnoses and proposes recovery actions, and an engineer still makes any change.InfoQ ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
GitLab Orbit: Find vulnerability blast radius without the manual huntannouncementGitLab has made Orbit generally available for Premium and Ultimate customers, starting on GitLab.com with Self-Managed and GitLab Dedicated to follow. Orbit builds a graph from merge requests, pipelines, deployments, security findings and ownership data so users can ask which projects contain a vulnerable component and who owns them. GitLab's Security Analyst Agent and external agents such as Claude Code and Codex can query the same graph over MCP, subject to existing roles.platform teamsPlatform teams on GitLab can use the graph to map which services, base images and pipelines depend on a shared component, instead of assembling that list from scanners and spreadsheets. Teams should still check the graph against runtime inventory, since it only covers what repositories declare on the default branch.GitLab Blog ↗GitLabwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
GitLab Orbit is GA: Higher agent accuracy, lower token costannouncementGitLab announced general availability of Orbit, a context graph that indexes code, merge requests, pipelines, deployments, vulnerabilities and ownership for GitLab customers. GitLab says agents connected to Orbit needed up to 45x fewer retries and used up to 4.5x fewer tokens on the same tasks, based on its own testing. Orbit is available to Premium and Ultimate customers on GitLab.com first, with Self-Managed and Dedicated following, and is priced as a shared pool of queries.platform teamsPlatform teams using GitLab could query dependency, pipeline and ownership data through MCP or REST instead of building their own graph, but the efficiency figures are vendor-reported and unverified. Teams should budget for the per-organization query pool, which costs $30 per user per month for 1,000 queries plus about $0.03 per query beyond the pool.GitLab Blog ↗GitLabwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-283 days ago
Amazon Bedrock AgentCore CLI version 1.0 is now availablepostAWS made Amazon Bedrock AgentCore CLI version 1.0 (@aws/agentcore 1.0.0) generally available. The rebuilt CLI offers faster startup, a consistent command structure, and support for more AgentCore resources including harnesses, runtimes, memory, identity, gateways, and payments. Existing projects can be moved to version 1 with a new agentcore migrate command.platform teamsPlatform teams that evaluate or support AgentCore get a terminal tool to create, deploy and operate agent resources, with the ability to work on resources by ID or ARN without a project. Teams with existing AgentCore projects will need to run the migration command to move to version 1.What's New with AWS ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
The New Stack: What happens when the agent is long gone, but its access key lives on?postThe article argues that access keys created for agents or deployments persist after their tasks end, since AWS IAM user access keys do not expire automatically. It provides a CloudQuery SQL query to list active IAM keys older than 90 days and unused for 90 days, and an env zero OPA policy that denies creation of selected long-lived credential types in Terraform and OpenTofu plans. It also notes that TTL-driven teardown does not guarantee revocation and shows AWS CLI commands to deactivate a specific key.platform teamsPlatform teams running agent or deployment pipelines can use the inventory query and deploy-time policy to find and block long-lived keys, but must still verify revocation separately, since TTL destroys and policies do not cover credentials created outside managed deployments.The New Stack ↗Terraform MCP ServerOpenTofuAmazon EKSwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-274 days ago
Extend GitLab Duo Agent Platform into your business systemspostGitLab published a tutorial showing how Salesforce can call a custom GitLab Duo Agent Platform flow through the Flows API when an Apex error occurs. The flow reads the repository, identifies a likely cause and suspect commit, and opens a GitLab issue with the analysis. The tutorial limits the agent to four read tools and one issue-creation tool, and uses a project access token and Named Credential.platform teamsPlatform teams that manage Salesforce metadata in GitLab could adapt this pattern to route runtime errors to agent triage with narrow privileges and an audit trail. Teams must still handle data stripping, token scoping, and privilege limits before production use, and the tutorial omits response checks, retries, and tests.GitLab Blog ↗GitLabSalesforce PlatformAgentforcewritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-265 days ago
InfoQ: AGNTCon + MCPCon 2026: Agentic AI Foundation Adds Four Projects as MCP Turns to SecuritypostAt AGNTCon + MCPCon 2026 in San Jose, the Agentic AI Foundation added four projects, bringing its total to ten in its first year. Speakers focused on MCP security and identity, including stateless servers, tool annotations, and agents carrying their own identity. Many figures were stage claims without published data, and some security extensions are still drafts.platform teamsPlatform teams running MCP servers should annotate every tool, keep tool schemas small, and plan for agents that authenticate with their own identity rather than a borrowed user token. The security and identity extensions discussed are still drafts, so no settled controls can be relied on yet.InfoQ ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
Tracing failures across Terraform state boundaries with AWS DevOps AgentpostAWS publishes a walkthrough of using AWS DevOps Agent to trace failures across separate Terraform state files. The sample has an infrastructure team tighten a security group egress rule, which breaks an application Lambda's DynamoDB access, and the agent correlates CloudWatch alarms and CloudTrail to point at the networking change. The post also covers provisioning the agent space with the hashicorp/awscc provider and bridging CloudWatch alarms to the agent's HMAC-signed webhook through a Lambda function. The source text is cut off before the walkthrough is complete.platform teamsPlatform teams that split infrastructure into multiple Terraform stacks could use an AI agent to trace cross-stack root causes, but the source shows only a demo. The webhook is console-only and not exposed as a Terraform resource, so it needs manual setup.AWS DevOps Blog ↗AWS DevOps AgentOpenTofuwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
Custom APIs just landed in the Dataverse MCP Server, Agents just got smarterannouncementThe Dataverse MCP Server now exposes Custom APIs, so agents can invoke custom business operations hosted in Dataverse regardless of where the agent runs. A Custom API backed by a plug-in runs its steps inside a single Dataverse transaction and rolls back fully if any step fails. The post describes a search, describe, and invoke_api sequence the agent uses to discover and call these APIs from metadata, with a sample case-creation example. Custom Actions are not supported yet.platform teamsPlatform teams can expose multi-step business logic as one atomic, typed operation that agents discover automatically, rather than letting agents chain separate record writes. Teams should note that Custom Actions are not yet supported through the MCP server.Microsoft Power Platform Blog ↗Microsoft Power Platformwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
The New Stack: The sandbox wasn’t enough: Why AI agent governance requires runtime observabilitypostThe article uses two incidents, an OpenAI cybersecurity evaluation that reached the public internet and an Anthropic review of 141,006 evaluation runs, to argue that a configured isolation boundary does not show what is reachable at runtime. It proposes four properties for runtime evidence: coverage, correlation, timeliness and auditability. It also describes how agent traces from Langfuse and infrastructure telemetry from ClickStack could share OpenTelemetry trace context.platform teamsPlatform teams running agents should instrument every trust boundary, including network egress, credential use and production writes, and link agent tool calls to infrastructure effects. They should also place synchronous policy gates in front of high-impact actions so a run can be suspended or its credentials revoked before effects become irreversible.The New Stack ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
CNCF: Isolation. Security. Scale.postThe CNCF blog post explains agent sandboxing and describes Agent Substrate, which runs agents as Actors inside Workers (Pods) managed by a control plane, using gVisor for software-level isolation and microVMs for hardware-level isolation. It walks through installing kagent and Agent Substrate with Helm, declaratively creating a Harness and AgentTemplate, and running a sandboxed assistant. The post says an Actor runs only while a session is active.platform teamsA platform team running kagent can provision agent sandboxes as declarative resources and keep agent sessions isolated from the host kernel, with the sandbox runtime chosen per workload. Teams should check the post's setup steps against current kagent and Agent Substrate releases before relying on them, since the post's own install steps are linked out rather than shown.CNCF Blog ↗kagentwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
How to deliver your first change with Cam: a walkthroughpostGearset published a walkthrough of Cam, its AI agent, delivering a sample Salesforce change: a VAT number field with a placeholder validation, a finance Flow, and access across 23 profiles. The steps are assigning a Jira or Azure DevOps ticket, answering Cam's questions, approving a nine-step build plan on a feature branch, reviewing the diff, and raising a pull request that Cam fixes if validation fails. The post also describes using Gearset Observability to hand production Apex and Flow errors to Cam.platform teamsPlatform teams running Salesforce on Gearset pipelines could route backlog and production error fixes to an agent whose output passes the same tests, approvals and security checks as human-written changes. The post does not give measured results, so the reliability of this workflow is untested by the source.Gearset Blog ↗GearsetSalesforce Platformwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-256 days ago
The New Stack: Is Kubernetes ready for AI agents?postThe New Stack argues that Kubernetes was built for software that does what its developers specify, and that AI agents strain that assumption. It says a CNCF survey found two-thirds of organizations running AI inference workloads do so on Kubernetes. The article calls for sandboxing, explicit permissioning, monitoring and warm pools for agents, and says the open-source project does not yet fully provide these.platform teamsPlatform teams running agent workloads should expect to need stronger sandboxing and permission controls than default Kubernetes offers today, and may need to consider commercial distributions with those features. The article describes an agent substrate using warm pools for fast agent startup but gives no implementation details or benchmarks.The New Stack ↗Amazon EKSwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-247 days ago
The New Stack: Building an agentic software factory? Start with the sensor.postThe article argues that agentic software factories are limited by their verification step, which it calls the sensor, rather than by coding agents. It defines three sensor properties (latency, fidelity, bandwidth) and suggests tracking time to signal, escape rate and concurrent verification capacity. The article is written by a team that sells a product for per-change realistic environments and presents that approach as the solution.platform teamsPlatform teams evaluating agent-driven pipelines can use the proposed metrics to find which verification property is weakest before raising agent autonomy. The article does not describe a specific tool change, and its recommended per-change environment approach comes from the vendor's own view.The New Stack ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
InfoQ: Google’s Data Agent Kit Hits GA, Splitting Agent Guardrails Between Skills and IAMpostGoogle made Data Agent Kit generally available, moving it out of its summer preview. The kit is open source and installs into coding agents such as Claude Code, Cursor and Codex, providing MCP tools for more than fifteen Google Data Cloud services plus 37 Google-authored skills under Apache 2.0. Google documents an IAM deny policy that blocks MCP tools where tool.isReadOnly is false, applied at organization, folder or project level.platform teamsPlatform teams running Google Cloud MCP servers can restrict agents to read-only tools with IAM deny policies instead of relying only on skill guidance. Blocked tools still appear in tools/list, the controls apply only to Google Cloud MCP servers, and the resource.service and tool.name attributes are available only through gcloud.InfoQ ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
The New Stack: AI jobs are waiting while GPUs sit idle. Kubernetes has a new plan.postThe New Stack's Road to KubeCon column covers several Kubernetes topics: a HPE blog post on an open-source tool that gathers pod-restart diagnostics to Slack, an AWS post on scheduler and autoscaler friction for GPUs, and the llm-d v1.0 release. It also reports that in one benchmark by Skyhook's Radar, Claude Code given read-only MCP context diagnosed 50 injected faults faster than with shell and kubectl access, reaching a correct diagnosis within two minutes on 86% of faults versus 60%. Splunk's State of Observability 2026 survey of 2,050 professionals reports that over half cannot effectively measure AI quality or control AI costs.platform teamsPlatform teams evaluating AI troubleshooting agents may find that read-only MCP context improved speed in one author's benchmark, though the benchmark is a single small test by its author. The Splunk survey is vendor-commissioned, so its figures should be treated as indicative rather than verified.The New Stack ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
AgentCore Gateway now supports Bedrock InvokeModel APIs and model fallbackannouncementAmazon Bedrock AgentCore Gateway now accepts the Bedrock InvokeModel and InvokeModelWithResponseStream APIs on its inference endpoint. Existing Bedrock SDK clients can point at the gateway to reach any model provider through one endpoint. The gateway also adds model fallback routing, which sends a request to an ordered chain of alternative destinations when the primary is throttled, erroring, or timing out, and a circuit breaker skips destinations that are currently failing.platform teamsPlatform teams can centralize model access for agents behind one governed endpoint and move retry and failover logic out of each client. The source says the circuit breaker reduces latency for failed attempts, which is a vendor claim not measured in the text.What's New with AWS ↗written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-2110 days ago
K8sGPTv0.4.41patchsecurity+++**********!!!K8sGPT v0.4.41 was released with three dependency security updates, including containerd v1.7.36, spdystream v0.5.1 and otel sdk v1.45.0. It adds Claude 4.5 and 5.x model support for Amazon Bedrock, the ability to source customHeaders from the config file, and a prompt to auto-pull missing Ollama models. The release also fixes ten analyzer and completion issues, covering StatefulSet claim templates, NetworkPolicy selectors, Gateway API namespace filters, HTTPRoute namespace selectors, node-problem-detector conditions, and Kyverno ClusterPolicyReports.platform teamsPlatform teams running K8sGPT get fewer false or missed findings for NetworkPolicy, Gateway API and Kyverno reports after upgrading. The security items are dependency bumps rather than fixes in K8sGPT's own code.written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
Observability patterns for AI agents on Amazon EKSpostAWS Containers Blog describes two observability patterns for AI agents on Amazon EKS: a decentralized pattern where each agent exports traces straight to Langfuse, and a centralized pattern where traces pass through a shared OpenTelemetry Collector first. Both use the Strands Agents SDK for instrumentation, Bifrost as an LLM gateway routing to Amazon Bedrock, and OpenTelemetry GenAI conventions, with trace context propagated via traceparent headers. The post is cut off before its full results were shown in the source text.platform teamsPlatform teams choosing between the two patterns trade per-agent configuration sprawl against a shared Collector that must be sized and monitored as critical infrastructure. Adding a new telemetry backend is a Collector config change in the centralized pattern, and a change to every agent in the decentralized one.AWS Containers Blog ↗Amazon EKSArgo CDOpenTofuwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
CNCF: Before you evaluate an AI SRE, give your existing agent contextreportThe authors ran 50 Kubernetes faults from the SREGym benchmark on a live Amazon EKS cluster, comparing Claude Code with a shell and kubectl against Claude Code with read-only MCP tools from Radar, an open-source Kubernetes UI and MCP server the authors maintain. Both arms reached similar accuracy (46 vs 45 correct), but the context arm reached a correct diagnosis within two minutes on 86% of faults versus 60% with a shell, and used fewer tool calls on average (9.8 vs 16.3). The authors note that the context arm had different permissions from the shell arm, that each arm ran each fault once, and that they maintain the tool being tested.platform teamsPlatform teams evaluating AI SRE products can first measure whether their existing coding agent with read-only cluster context (current failures, change history, joined resources, filtered logs) diagnoses faults faster, using the published method and a read-only service account. Change history kept beyond Kubernetes event expiry and a read-only identity without Secrets or exec access are the concrete requirements the authors recommend checking.CNCF Blog ↗Kubernetes MCP ServerAmazon EKSK8sGPTwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
InfoQ: Presentation: AI is an Amplifier: Scale High Performance, Not DysfunctionreportDORA's Nathen Harvey presents findings from the 2025 DORA survey, which had nearly 5,000 respondents and ran June 13 to July 21, 2025. He reports that about 90% of respondents use AI at work and that the survey frames AI as an amplifier of existing organizational systems. The source text is cut off before the detailed relationship findings.platform teamsPlatform teams evaluating AI-assisted development should note that the talk argues gains in code generation do not reach production without work on the wider delivery system, such as review and release steps. The source gives no concrete tooling change.InfoQ ↗DORA State of DevOps / AI-assisted development reportwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
2026-10-2011 days ago
Limit agent permissions and spend in AWS Transform custompostThe AWS DevOps Blog describes three configuration controls for running the AWS Transform custom agent on fleets of repositories: a trust settings file that auto-approves some tool calls and shell commands, an IAM execution identity scoped to approved transformations, and an agent-minute cap plus an AWS Budgets alert on spend. The post walks through a sample Java 8 to Java 17 upgrade run using AWS Transform CLI 3.10.0, reporting 43 tool calls and that the trust allowlist reduced approval prompts from 23 to zero on a clean clone.platform teamsPlatform teams running agent-driven upgrades from pipelines can manage tool trust, identity scope and spend as configuration rather than new infrastructure. The post's own measurements are from one sample repository and one CLI version, so counts will differ on other estates.AWS DevOps Blog ↗AWS DevOps Agentwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
Redesign how your team ships software with AI agentsannouncementGitLab announced the AI-Native Engineering Playbook, a guide to redesigning software delivery so people and AI agents each handle the work they suit best. The playbook proposes a four-stage autonomy ladder in which each step up must be backed by evidence, and a layered verification approach for code review at agent volume. GitLab's AI Accountability Report, based on an April 2026 survey of 1,528 DevSecOps professionals, is cited, with 79% saying individual productivity improved but overall delivery did not accelerate at the same pace.platform teamsPlatform teams can use the playbook's layered checks as a model: move formatting, type and dependency policy checks closer to authoring, keep change sets small, and record agentic review decisions so they can be inspected later. The post does not describe any new platform feature or tooling change.GitLab Blog ↗GitLabwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
AI Governance: One view of agent activityannouncementGitLab has released the AI Governance Dashboard in beta, which shows activity from GitLab Duo Agent Platform agents and externally connected agents such as Claude Code, Cursor, and Codex in one view for a top-level group. Each agent action is recorded as an audit event with the prompt, model, tools, MCP servers, and project, and sessions can be downloaded as immutable evidence or streamed to a SIEM. Setup requires enabling AI audit event storage, which is off by default, and running glab govern setup on developer machines.platform teamsPlatform and security teams on GitLab Premium or Ultimate can get audit trails of agent tool calls and MCP server use, but must turn on AI audit event storage and hold the Owner role or a custom role with read_agent_artifacts. External agent capture depends on glab 1.122.0 or later on macOS or Linux.GitLab Blog ↗GitLabwritten by claude-haiku-5-5, prompt 7cbe154, checked by publish rules
daily email

One short email each morning: what shipped, what breaks, and a link to each change. Unsubscribe in one click. Privacy.

statuslast 30 days for lines
tools tracked43releases in registry964released in the last 30 days106entries published335breaking change lines18security lines84
change typestwo-column gutter, like git status
  • +addedNew functionality.
  • ~changedExisting behavior changed.
  • -deprecatedStill works, scheduled for removal.
  • -removedDecommissioned, gone in this version.
  • *fixedA defect corrected.
  • !securityA vulnerability closed or a dependency patched for one.
  • !breakingFlag in the first column; the change needs action before upgrading.
filter
categoryallidp & portalsai agents on platformci/cdobservabilitysupply chaincost & governancereports & dx dataenterprise app platforms
showmajor onlyreleasescase studiespostmortemsinclude patches