Public edition“All the posts worth keeping”Collected & filed
All the posts worth keeping
Section

Engineering & CI

20 stories · Engineering / CI, Engineering / CI & deploys, Engineering / Docker & containers, Engineering / PRs & agents, Engineering / code review, GitHub / agents & PRs, PRs & agents
Engineering / CI
Moved some GitHub Actions runners off Blacksmith ($400+/mo of compute) to an M5 Ultra 96GB Mac Studio at…
Engineering / CI — via @JPoehnelt

Moved some GitHub Actions runners off Blacksmith ($400+/mo of compute) to an M5 Ultra 96GB Mac Studio at…

Moved some GitHub Actions runners off Blacksmith ($400+/mo of compute) to an M5 Ultra 96GB Mac Studio at ~$110/mo with compute to spare. Thread context: runners managed with Tailnet (Tailscale) + k3s + actions-runner-controller (ARC); Mac Studio bought from Apple via Klarna financing (~$110/mo), not a hosting lease; also runs runners on an old Dell R630; staging/prod jobs stay in the cloud; on Starlink. A replier (joearroyo) also lets his idle M5 Ultra serve inference on Darkbloom (~34.5k jobs, ~$20).

14,717 views · 108 likes · 90 bookmarks · 13 replies
Engineering / CI
Thread reply

Thread reply

Thread reply: a homelab makes sense for him at $0.1156/kWh with solar panels; attached NLR degree-day projections chart for La Plata County, CO (heating-dominant climate).

1,178 views · 3 likes · 2 bookmarks · 2 replies
Engineering / CI
Answer to Sawyer Hood ('what do you use to self run?

Answer to Sawyer Hood ('what do you use to self run?

Answer to Sawyer Hood ('what do you use to self run? buildkite?'): Tailnet + k3s + GitHub actions-runner-controller.

841 views · 10 likes · 37 bookmarks
Engineering / code review
Will Larson (Imprint) on 'small-change'

Will Larson (Imprint) on 'small-change'

Will Larson (Imprint) on 'small-change': with PRs/eng/week up 4x avg (10x on some teams) and p50 review time 33 to 77 min, CI auto-marks a PR small-change if it's under 1,000 LOC, has no p1/p2 ai-code-review findings, the author has 10+ merged PRs in that service, it touches few services, and it skips shared libs (per-repo .small_change file). A review-small-change skill only approves or ignores (never comments, to avoid 'meat proxy review'), and review-pr-cluster reviews batches for themes. About 10% of PRs qualify so far. Incidents are down 42% and critical incidents down 78% half-over-half.

46,985 views · 536 likes · 821 bookmarks · 26 replies
Engineering / Docker & containers
Grok (replying to henkis in the schotek thread) compares Apple Containers vs OrbStack

Grok (replying to henkis in the schotek thread) compares Apple Containers vs OrbStack

Grok (replying to henkis in the schotek thread) compares Apple Containers vs OrbStack. Apple Containers: free/open source, per-container micro-VMs, near-zero idle, CLI-only (GUIs like Davit), limited Compose/K8s, macOS 26 + Apple Silicon only. OrbStack: full Docker drop-in (Compose, K8s, socket), polished GUI, 2-7x faster file reads, ~0.1% idle CPU; free personal, paid commercial. OrbStack wins for most workflows.

40 views · 1 replies
Engineering / Docker & containers
(Czech) schotek uninstalled Docker Desktop and now uses Apple's native Containers (container CLI 1.3.1)…

(Czech) schotek uninstalled Docker Desktop and now uses Apple's native Containers (container CLI 1.3.1)…

(Czech) schotek uninstalled Docker Desktop and now uses Apple's native Containers (container CLI 1.3.1) for Linux dev on the Mac: memory use went down and disk I/O is far better. Recommends a GUI dashboard app for managing containers/images/volumes/networks (name truncated in the post); screenshot shows it.

156,718 views · 1,326 likes · 1,749 bookmarks · 51 replies
Engineering / CI
Vercel Labs' tsc-ts

Vercel Labs' tsc-ts

Vercel Labs' tsc-ts: TypeScript's compiler (originally TS, rewritten in Go as tsgo) ported back to TypeScript and compiled natively with scriptc; claimed 2-4x faster than tsgo.

86,314 views · 1,178 likes · 290 bookmarks · 111 replies
PRs & agents
Claude Code Review is GA for Claude Team and Enterprise plans

Claude Code Review is GA for Claude Team and Enterprise plans

Claude Code Review is GA for Claude Team and Enterprise plans: faster reviews catching more important bugs, and admins choose Quick, Standard, or Deep review level per repo.

333,297 views · 4,128 likes · 1,762 bookmarks · 146 replies
Engineering / CI
ts-rust v0.6.0 (Theo's Rust port of TypeScript type checking)

ts-rust v0.6.0 (Theo's Rust port of TypeScript type checking)

ts-rust v0.6.0 (Theo's Rust port of TypeScript type checking): 30% faster, up to 70% faster with --fast (uses more RAM). Mostly closes the gap with `bun check` while keeping 100% compatibility with upstream TypeScript, which Bun lacks. Quotes his earlier video on why he built it (not just faster type checks).

32,620 views · 632 likes · 71 bookmarks · 39 replies
Engineering / CI
Another CI speed tip

Another CI speed tip

Another CI speed tip: use Oxlint instead of ESLint; it was 32x faster on Orca's repo. Quotes her earlier GitHub Actions parallel-steps tip.

16,990 views · 281 likes · 222 bookmarks · 11 replies
Engineering / CI
You can roughly 2x GitHub CI speed for free

You can roughly 2x GitHub CI speed for free

You can roughly 2x GitHub CI speed for free: GitHub Actions now runs steps in parallel (background steps, changelog 2026-06-25). AI tools often miss it because it's new; ask an agent to apply it and benchmark. https://github.blog/changelog/2026-06-25-actions-steps-can-now-be-run-in-parallel/

111,937 views · 888 likes · 1,241 bookmarks · 31 replies
Engineering / CI & deploys
Credits pstack and Opus 5.5 with cutting production deploy time by ~30% after giving the agent full…

Credits pstack and Opus 5.5 with cutting production deploy time by ~30% after giving the agent full…

Credits pstack and Opus 5.5 with cutting production deploy time by ~30% after giving the agent full access to deploy logs and source code. Table: whole deploy 5:27–6:28 → 4:10/3:57; pre-deploy 79–100s → 51/48s; rollout 84–142s → 52/63s; compress + upload build ~42s → ~25s.

48,201 views · 439 likes · 237 bookmarks · 28 replies
Engineering / CI
Follow-up to his Mac Mini self-hosted runner post

Follow-up to his Mac Mini self-hosted runner post

Follow-up to his Mac Mini self-hosted runner post: "No more of this" with a screenshot of GitHub's included usage maxed out (2,000 of 2,000 Actions minutes used, resets in 28 days).

6,678 views · 47 likes · 2 bookmarks · 1 replies
Engineering / CI
Switching from GitHub-hosted runners to a self-hosted runner on a 2023 Mac Mini cut CI workflows from 35…

Switching from GitHub-hosted runners to a self-hosted runner on a 2023 Mac Mini cut CI workflows from 35…

Switching from GitHub-hosted runners to a self-hosted runner on a 2023 Mac Mini cut CI workflows from 35 minutes to 5 minutes.

43,947 views · 682 likes · 375 bookmarks · 56 replies
Engineering / CI
After exhausting the GitHub Actions free tier, built self-hosted runners on Cloudflare Containers

After exhausting the GitHub Actions free tier, built self-hosted runners on Cloudflare Containers

After exhausting the GitHub Actions free tier, built self-hosted runners on Cloudflare Containers. If you switch to self-hosted only after the free quota is used, and you already pay for Workers Paid, the overage looks cheaper than GitHub.

83,339 views · 297 likes · 276 bookmarks
Engineering / CI

GitHub CI bill went from $0 to $380/month and was heading toward $800 in October

GitHub CI bill went from $0 to $380/month and was heading toward $800 in October. Cause: coding agents push ~300 times a day and the test suite grew 5x since August. Fix: no CI on PRs; dev only builds and deploys staging; full tests run nightly and on the release PR. New bill: ~$0.

148,783 views · 1,625 likes · 1,078 bookmarks · 141 replies
Engineering / PRs & agents

Responding to "slop grenade" chatter

Responding to "slop grenade" chatter: great teams treat huge exploratory PRs as throwaways — assume they'll be closed, not merged. Use them to explore architecture, blast radius, product feedback, and APIs; the mergeable PR can (and often should) be rebuilt separately. Goes further: non-maintainer PRs should be closed by default and kept only as references for future humans and agents.

28,884 views · 255 likes · 97 bookmarks · 29 replies
GitHub / agents & PRs
GitHub Product (Stacked PRs) announces the async merge API is generally available

GitHub Product (Stacked PRs) announces the async merge API is generally available

GitHub Product (Stacked PRs) announces the async merge API is generally available — for programmatic merges by humans or agents. Can merge individual/stacked PRs, add to a merge queue, or merge directly. Changelog: https://github.blog/changelog/2026-10-01-github-async-merge-api-generally-available/

64,001 views · 461 likes · 444 bookmarks · 15 replies
Engineering / CI

Tongue-in-cheek reply to Flo Crivello (@Altimor)

Tongue-in-cheek reply to Flo Crivello (@Altimor): when CI is the top bottleneck and spend is "stratospheric," kache's suggestion is simply "don't run CI." Quotes Altimor's post about CI bottlenecking engineering teams including Lindy.

88,056 views · 599 likes · 182 bookmarks · 35 replies
Engineering / CI

Reply

Reply: move many GitHub Actions checks into pre-commit and pre-push hooks so CI does less; agents can run those hooks, fix issues, and uphold codebase rules. Pair with Turborepo remote caching to cut cost materially.

346 views · 3 bookmarks