TLDR DevOps 2026-02-06

Agentic AI is becoming a buying decision, not a feature. In a survey of 1,000+ developers and product leaders, we found that agentic AI is already acting as a switching event across the modern tech stack.

The Developer's Guide to Agentic AI 2026 report breaks down:

• The timeline: Why 85% say agentic AI will be table stakes within three years
• Production over hype: How 67% of teams are already shipping today (mostly internal ops)
• The bottlenecks: Why security approvals (50%) and integration complexity (44%) are the biggest blockers to scale

Read the full report →

Container security risk concentrates in long-tail images, not popular bases, with 98% of CVEs outside the top 20 images. Fast remediation, minimal images, and compliance-driven hardening are critical to managing widespread open-source dependency risk.

Pulumi's Cloud REST API is now described by an OpenAPI 3.0 specification, a highly requested feature that provides better tooling and a predictable API experience for developers. The OpenAPI 3.0.3 document, which powers the Pulumi CLI and Console, is generated directly from the API definition to ensure it stays in sync with the service.

Databricks has unveiled an architecture for continuously improving AI agents, exemplified by an American Football Defensive Coordinator Assistant, which leverages MLflow's `align()` and `optimize_prompts()` functions to integrate domain expert feedback for automated prompt refinement and enhanced performance. This system encodes human expertise directly into the agent, providing situation-aware insights for coaches and a reusable framework for other domain-specific AI applications.

GitHub Actions wins by convenience, not quality, and its slow UI, brittle YAML DSL, opaque permissions, untrustworthy marketplace, and rented compute quietly drain engineering time and morale. Buildkite, by contrast, keeps config simple, puts real logic in real code, lets teams own fast, debuggable infrastructure, and feels like a CI tool built by people who actually suffer through CI every day.

Comma.ai runs all of its ML training, storage, and metrics on a self-owned, in-office data center, which it estimates cost about $5M to build and operate versus $25M+ for equivalent cloud usage. The data center uses self-built GPU servers, non-redundant high-throughput SSD storage, Slurm-managed compute, PyTorch distributed training, and custom lightweight infrastructure to efficiently support large-scale model training and experimentation.

Concourse, a container-based automation system written in Go, is primarily utilized for CI/CD, offering a scalable solution for complex pipelines with strong opinions on idempotency and reproducible builds.

kagent, a Kubernetes native framework, was developed to simplify the building, deployment, and management of AI agents within the Kubernetes orchestration platform. This Cloud Native Computing Foundation project is currently in active development.

Codefresh's Pipeline Performance Profiling is an OpenTelemetry and Prometheus-based capability that exposes step-level timing, resource usage, and cache metrics to help teams identify CI/CD bottlenecks, right-size resources, and optimize pipeline speed and cost using existing observability tools.

Scalable DevOps requires standardized, automated best practices, and AWS Well-Architected provides broad guidance relevant to CI/CD. The Octopus AI Assistant uses LLMs to analyze Octopus projects and generate one click AWS Well-Architected reports for fast, consistent alignment.

AWS Lambda added CloudWatch Logs and metrics for Kafka event source mappings, improving visibility into polling, scaling, and event processing for Amazon MSK and self-managed Kafka.

Amazon S3 now supports changing server-side encryption for encrypted objects without data movement using the UpdateObjectEncryption API.

AWS launched new Amazon EC2 C8id, M8id, and R8id instances, delivering up to 43% higher compute performance and 3.3 times more memory bandwidth than previous generations.