content(blog): The Engine and the Spine — H·AI·K·U vs Dynamic Workflows

[jwaldrip]

What

New blog post: "The Engine and the Spine" — a positioning piece comparing Anthropic's newly-announced Dynamic Workflows in Claude Code against H·AI·K·U.

Angle

They're complementary layers, not rivals. Dynamic Workflows is a fan-out engine for one enormous bounded task (the Bun Zig→Rust port is the proof). H·AI·K·U is a durable lifecycle spine that governs the repeated feature lifecycle. The load-bearing contrast is the orchestrator seat — Dynamic Workflows trusts the model to plan the run shape at runtime; H·AI·K·U welds that role into git-backed engine code because the in-context plan drifts. The post leads its convergence section with the honest admission that we already ship the same fan-out-and-converge primitive as a tool, and closes on the genuine open question: whether that muscle belongs inside our spine.

Structure / rules compliance

• Opens cold on the Bun port; thesis ("orchestrator seat") earned at ~45%, reframe at ~70%, closes on a question — per content-voice + don't-hand-the-reader-the-thesis.
• 4 Compare blocks + 1 Callout across three sections (same bet / diverge / nest), prose between every landmark — per blog-visual-flair.
• Announcement cited by link; every claim about H·AI·K·U maps to real engine behavior. Cross-links four existing posts.
• General-audience pass + humanize pass applied before commit.

🤖 Generated with Claude Code

[claude]

Claude finished @jwaldrip's task in 4m 21s —— View job

---

PR Review: "The Engine and the Spine" blog post

• Load tools and gather context
• Read the blog post
• Review against voice/content rules (content-voice.md)
• Review against visual flair rules (blog-visual-flair.md)
• Review against general audience rules (blog-general-audience.md)
• Review against citations rules (citations.md)
• Check accuracy of H·AI·K·U claims
• Post final review

---

Overall Assessment

Strong positioning post — the angle is well-chosen, the Bun port opening earns the context, and the "orchestrator seat" is a genuinely useful label that carries the argument. The three-section structure maps cleanly to the arc. Two issues need fixing before merge, one is a potential accuracy question worth the author's eye.

---

Issues

1. Back-to-back Compare blocks (lines 30–32) — must fix

Compare 1 closes at line 30 and Compare 2 opens at line 32 with only a blank line between them. blog-visual-flair.md is explicit: "never two components back-to-back without prose between them." The connecting thought that currently appears after Compare 2 ("Two teams, same two answers...") should move between the two blocks.

Fix this →

2. Description over the 150–160 char SEO target — should fix

The description frontmatter is 181 characters. The rules say 150–160 for SEO-friendliness (Google truncates at ~155). Current:

"Anthropic's Dynamic Workflows fan tens-to-hundreds of subagents across one giant task. H·AI·K·U governs the whole feature lifecycle, over and over. Different layers — and they nest."

A tighter version that lands under 160: "Anthropic's Dynamic Workflows fan hundreds of subagents across one enormous task. H·AI·K·U governs the repeated feature lifecycle, gate by gate. Different layers — and they nest." (157 chars)

Fix this →

3. "Workflow primitive exposed as an engine tool" — accuracy question

Lines 17 and 28 both claim H·AI·K·U exposes "the same Workflow fan-out-and-converge primitive as an engine tool." The codebase's MCP tool list doesn't include a haiku_workflow or Workflow primitive tool — the parallel fan-out is an internal cursor behavior that emerges from how haiku_run_next dispatches unit worktrees, not a named tool agents can call directly.

The closing paragraph at line 91 correctly says "We use the same fan-out primitive internally already" — which is accurate. The earlier claims use "exposed as a tool" which implies a callable MCP surface that doesn't appear to exist.

If this primitive is being announced as new tooling, that needs more context. If the intent is just "we do the same thing internally," align the Compare card language with the closing paragraph's more accurate framing.

---

Minor Notes

Register inconsistency, line 58: The post uses "we" throughout, then has "let me name it." The rules allow "I" only for "a specific lived moment that's clearly Jason's." An essay move ("let me name it") isn't that. Swap to "Call it the orchestrator seat" — drops two words and removes the register flicker.

Unexplained jargon, line 87: "merged back under a stage lock only at the terminal" uses "stage lock" and "terminal" as internal engine terms without definition. The general-audience rule requires defining internal terms at first mention. A quick gloss: "merged back under a per-stage mutex only when the last writer finishes" — same meaning, no prior knowledge required.

---

What's Working Well

• The Bun port opening is the right anchor — specific enough to be real, dramatic enough to hold attention.
• "The orchestrator seat" earns its name late enough (line 58, ~63%) and the argument leading to it is tight.
• The "save file for one playthrough" metaphor (line 73) is sticky and the contrast with H·AI·K·U's permanent record lands cleanly.
• All three cross-links verified against existing posts (one-instruction-at-a-time, dont-review-broken-code, the-loop-that-watches-the-loop — all present ✓).
• The closing question ("whether the muscle Anthropic just shipped belongs inside the spine we built") is the right move — opens a door instead of closing one.
• Announcement is cited by URL; all H·AI·K·U technical claims (delivery gate, worktree isolation, configurable stage gates, git-backed state) are accurate to the implementation.