ctx: tickets → self-service · SDLC → ADLC
Platform engineering and AI tool reviews — because ninety tickets
to get a namespace and every deploy blocked on a human clicking
apply is not a process, it's a queue.
How we judge
Tools — we review the platforms, pipelines, and agents that replace ticket queues with self-service golden paths: IDPs, GitOps, observability, IaC, and the AI agents now doing the work humans used to wait on.
Architecture — we cover the patterns that tie them together: how control planes, platform APIs, and reconciliation loops compose into systems that run without a human in the loop.
The bar — we rate everything on how far it moves teams from a
human-paced SDLC to an agent-driven ADLC, not on
feature checklists.
API keys break at agent scale. OAuth token exchange, MCP authorization, SPIFFE/SPIRE identity, A2A security — with honest weaknesses and skip verdicts.
Karpenter 1.15.0 turns node auto-repair into a voluntary disruption method: budgets, per-condition drain bounds, XID-aware GPU policies, in-place reboots.
AI coding agent security tools compared: Shoofly, ClawMoat, Armorer, AgentPort, Lakera, DIY — live-tested detection gaps and October 2026 pricing.
Envoy Proxy 1.40.0 hands-on: stable H3/QUIC, PQC-by-default TLS, the MCP router auth-header breaking change, and the bzlmod migration tax.
Hands-on: OpenCode 2.0's always-on service daemon, Code Mode interpreter, and the /etc/opencode managed-config regression that broke enterprise lockdown.
Agent37 sells agent sandboxes at $0.37/mo. We took apart the gVisor, Hetzner, and oversubscription math — and the failure modes nobody prices in.
Enterprise Terraform alternatives compared: Pulumi vs OpenTofu vs Crossplane, the CDKTF shutdown exit plan, and control-plane pricing as of Oct 2026.
From a 'conductor' codename and a WordPress demo to CNCF graduation — how Crossplane shipped, deleted, and rebuilt its way into the control plane era.
Anthropic moved off git at 30,000 merges a day. The walls are real, the fix is probably not yours to build — here is the ladder.
Model caching on GKE compared: GCS FUSE Profiles, Hyperdisk ML, Rapid Cache, and Kubernetes Image Volumes — cost math and when NOT to use each.