Dora Noda
Software Engineer
2307 posts · View all authors
The AI Rollback Paradox: 81% of the Best-Governed Teams Pulled Their Agents Anyway
Sinch's 2026 survey found 74% of enterprises rolled back a live AI agent — and the best-governed teams rolled back the most, at 81%. Here's what that means for building rollback into an MCP deploy surface as a cheap default instead of another approval gate.
AI Sandbox Pricing Audited: Northflank's $0.01667/vCPU-hr vs E2B/Daytona's $0.0504 vs Modal's $0.1419
Every AI-sandbox pricing comparison quotes Modal at 0.1419 dollars per vCPU-hour against Northflank's 0.01667 — but Modal bills per physical core, not per vCPU. Here's the corrected math and the exact utilization threshold where self-hosting on owned Hetzner hardware actually beats each vendor's metered rate.
What Cloud Native Buildpacks' Convention Buys a Git-Push PaaS Over a Hand-Rolled Build Script — and the One Escape Hatch It Still Needs
Cloud Native Buildpacks trades a hand-rolled detect-and-build script for community-maintained language coverage, reproducible image digests, and built-in SBOMs — but a git-push PaaS still needs a Dockerfile escape hatch for the repos convention can't detect.
CNCF Says Cloud Native Already Runs AI. Here's What Actually Transfers to a Git-Push PaaS
CNCF says the same cloud-native primitives already run web apps and AI workloads. Here's exactly what transfers to a git-push PaaS today — GitOps, containers, observability — and what still doesn't: GPU scheduling, disaggregated serving, and multi-agent orchestration.
Docker Cancels Hub Pull-Rate Charges Outright: What Walking Back a Pricing Threat (Not Just Postponing It) Signals About Registry Cost Risk on Owned Hardware
Docker didn't just delay its Hub pull-consumption pricing again — it cancelled the plan outright in April 2025. Here's what that reversal actually changed, what rate-limit risk survived it untouched, and what a pull-through cache costs to run on owned hardware versus metered cloud infrastructure.
Docker Engine v29 Raised Its API Floor to 1.44 — Here's Exactly Who Survived
Docker Engine v29 raised its minimum client API version to 1.44 and instantly broke every self-hosted tool with an older version hardcoded. A tool-by-tool accounting of who fixed it, who didn't, and why the difference wasn't luck.
Dokploy and Coolify Stopped Calling Themselves 'Heroku Alternatives.' Here's What That Actually Means
Dokploy and Coolify used to pitch themselves as Heroku replacements. In 2026 both lead with Vercel and Netlify instead — a rebrand backed by real pricing math, GitHub star counts, and a feature gap that says a lot about who's actually migrating.
Earthly Shut Down a Year Ago. Here's What Betting on a VC-Funded Build Tool Actually Costs
Earthly raised $8.6M, hit 12,000 GitHub stars, then shut down its cloud and open-source project within a year. A dated retrospective on what the arranged migration to Dagger actually cost users, and why picking commodity build specs over single-vendor DSLs is the real fix.
HAMi Becomes a CNCF Incubating Project: What Fractional GPU Sharing Actually Buys a Fleet With a Handful of Cards
HAMi just reached CNCF Incubating status, and its own case studies show why it matters: DaoCloud cut GPU costs roughly in half after adopting it. Here's the actual per-tenant cost math for fractional GPU sharing versus buying a whole card per AI-agent sandbox.