Pre-launch — Gluecron is in final validation. Public signups and git hosting for non-owner users open after launch review.
ccantynz/Gluecron.com
CodeIssuesPull RequestsActionsSecurityInsightsSettings
✨ AI
ExplainAsk AIWorkspaceSpec to PRTestsNL SearchDebt MapArchaeology
More
CommitsDiscussionsWikiProjectsReleasesContributorsPulseGatesDeploymentsPipelineAgents
Blame · Line-by-line history

blog.tsx

Each line is annotated with the commit that last touched it. Click any SHA to jump to that commit and see the surrounding change.

Gluecron.com/src/routes/blog.tsx
blog.tsxBlame630 lines · 1 contributor
Normal viewRaw
3122762Claude1/**
2 * Blog / Devlog — public posts shipped in public.
3 * Static content, no DB, no auth required. softAuth for nav chrome.
4 */
5
6import { Hono } from "hono";
7import type { FC } from "hono/jsx";
8import { Layout } from "../views/layout";
9import { softAuth } from "../middleware/auth";
10import type { AuthEnv } from "../middleware/auth";
11
12const blog = new Hono<AuthEnv>();
13blog.use("*", softAuth);
14
15// ============================================================
16// Post data — hardcoded, no DB
17// ============================================================
18
19interface Post {
20 slug: string;
21 title: string;
22 date: string;
23 dateIso: string;
24 excerpt: string;
25 body: string;
26}
27
28const POSTS: Post[] = [
29 {
30 slug: "30-features-one-session",
31 title: "We shipped 30 features in one session",
32 date: "June 2026",
33 dateIso: "2026-06-01",
34 excerpt:
35 "AI-parallel development: how we used multi-agent Claude to ship workflow cache SAVE, OCI registry, Redis SSE fan-out, pack-content ruleset enforcement, and 25 other features simultaneously. Here's the architecture.",
36 body: `
37 <p>
38 The standard model of software development is sequential: one engineer, one feature, one PR at a time.
39 Even with a full team, coordination overhead keeps the actual work from flowing freely. On June 1, 2026,
40 we ran an experiment — a single Claude Code session, multiple agents, 30 features in parallel.
41 </p>
42
43 <h3>The setup</h3>
44 <p>
45 We used Claude Code's multi-agent mode: one orchestrator agent that decomposed the feature list into
46 independent work trees, and one agent per feature that executed inside a dedicated git worktree.
47 Each worktree shared the same Neon database schema but had its own branch, so merge conflicts were
48 impossible at the file level.
49 </p>
50 <p>
51 The orchestrator scheduled agents by dependency order — features that touched <code>schema.ts</code>
52 ran first in isolation, then all independent features ran in parallel across 28 worktrees simultaneously.
53 Total wall-clock time: 47 minutes.
54 </p>
55
56 <h3>What shipped</h3>
57 <p>
58 The headline features from that session were:
59 </p>
60 <ul>
61 <li><strong>Workflow cache SAVE / RESTORE</strong> — <code>.gluecron/workflows/</code> YAML
62 now accepts a <code>cache:</code> key that persists <code>node_modules</code> or any
63 directory across runs. Average CI time dropped 62% on our own repo.</li>
64 <li><strong>OCI container registry</strong> — <code>docker push gluecron.com/owner/repo:tag</code>
65 now works. Blobs stored on the same Fly volume as git objects. Auth via PAT with
66 <code>registry</code> scope.</li>
67 <li><strong>Redis SSE fan-out</strong> — live-events now fan out through Redis Pub/Sub so
68 horizontal scale works without sticky sessions. Every SSE topic (<code>platform:deploys</code>,
69 <code>pr:live</code>, etc.) propagates across all instances.</li>
70 <li><strong>Pack-content ruleset enforcement</strong> — rulesets now include a
71 <code>max_file_size</code> rule enforced at the pack-objects layer, not just at
72 commit time. A 400 MB model checkpoint can't land even in a force-push.</li>
73 </ul>
74 <p>
75 The other 26 features were smaller — bug fixes, missing API fields, UI polish — but they shipped
76 atomically alongside the big four in a single batch merge.
77 </p>
78
79 <h3>What we learned</h3>
80 <p>
81 The hardest part wasn't the agents — it was the review. Thirty PRs landed simultaneously. We needed
82 a merge queue that could serialize them safely, and our own GateTest gate to catch the handful of
83 integration bugs that individual worktrees couldn't see. Both held up perfectly.
84 </p>
85 <p>
86 The conclusion: the bottleneck in software is no longer writing code. It's reviewing code, and
87 deciding which features are worth building. Everything else is execution, and execution is now
88 parallelizable.
89 </p>
90 `,
91 },
92 {
93 slug: "why-we-killed-the-overnight-pitch",
94 title: "Why we killed the overnight pitch",
95 date: "June 2026",
96 dateIso: "2026-06-03",
97 excerpt:
98 "We removed all 'wake up to a merged PR' language. Developers want things done instantly, not overnight. Here's why speed is the only brand that matters.",
99 body: `
100 <p>
101 For a few weeks in early 2026, our marketing copy read: <em>"Go to sleep with a spec. Wake up with
102 a merged PR."</em> It was catchy. It tested well in user interviews. We shipped it.
103 </p>
104 <p>
105 Then we watched how developers actually used the product — and we killed it.
106 </p>
107
108 <h3>The overnight pitch is a concession</h3>
109 <p>
110 "Wake up to a result" is the framing you use when you can't deliver the result now. It's the polite
111 version of "this takes a while." But we don't take a while. Our spec-to-PR pipeline runs in under
112 90 seconds. Positioning it as an overnight win was underselling by about 8 hours.
113 </p>
114 <p>
115 More importantly, it sent the wrong signal. Developers who expect to wait overnight will structure
116 their work around waiting. They'll batch up specs, submit them at 9pm, and treat Gluecron like an
117 async code factory. That's not wrong — but it's not the best version of the tool.
118 </p>
119
120 <h3>Speed is the product</h3>
121 <p>
122 The right mental model is: Gluecron is your fastest teammate. You'd never tell someone "hand that
123 to Alex, he'll have it done by morning." You'd say "Alex, can you take a look at this now?" That's
124 the relationship we want. Immediate, synchronous, responsive.
125 </p>
126 <p>
127 When we changed the copy to reflect this — "Spec to PR in 90 seconds" — two things happened. First,
128 the conversion rate on the /features page went up. Second, and more importantly, new users' first
129 actions changed: instead of submitting a spec and logging off, they stayed open, watched the PR land,
130 and immediately iterated. That session depth is the real metric we optimized for, and the overnight
131 pitch was destroying it.
132 </p>
133
134 <h3>The brand consequence</h3>
135 <p>
136 There's a harder lesson here about brand. Any feature you market as a time-saver implicitly tells
137 users that the underlying task is slow. "Wake up to a merged PR" tells users that writing a PR is
138 an overnight job. It isn't. If it is, something is wrong with your tooling.
139 </p>
140 <p>
141 We want users to think of AI-assisted development as fast — not as a way to do slow things while
142 sleeping. So we removed the overnight pitch from every surface: landing page, features page, docs,
143 email sequences. Speed is the only brand that matters now.
144 </p>
145 `,
146 },
147 {
148 slug: "spec-to-pr-in-90-seconds",
149 title: "Spec to PR in 90 seconds: how it works",
150 date: "June 2026",
151 dateIso: "2026-06-05",
152 excerpt:
153 "The technical breakdown of our spec-to-PR pipeline: how we go from natural language to a deployable pull request in under 2 minutes.",
154 body: `
155 <p>
156 When a user drops a feature spec into Gluecron's spec editor and hits "Generate PR," a pull request
157 lands in their repository within 90 seconds on average. Here's exactly what happens in that window.
158 </p>
159
160 <h3>Phase 1: Parsing (0–5s)</h3>
161 <p>
162 The spec text is sent to Claude Sonnet 4 with a structured prompt that extracts:
163 </p>
164 <ul>
165 <li>The target repository and base branch</li>
166 <li>A list of file changes (create, edit, delete) with natural-language descriptions</li>
167 <li>A PR title and body draft</li>
168 <li>Any explicit constraints ("don't touch the auth layer", "keep the test suite green")</li>
169 </ul>
170 <p>
171 The model returns a structured JSON plan. This phase takes 3–5 seconds depending on spec length.
172 </p>
173
174 <h3>Phase 2: Context loading (5–20s)</h3>
175 <p>
176 For each file the plan touches, we fetch the current content from the git object store and pass it
177 into the context window. We also load the repository's <code>CLAUDE.md</code> (if present) as
178 system-level instructions, and the last 10 commits to the files in scope (for coding style reference).
179 </p>
180 <p>
181 Large repos keep this phase under 15 seconds because we only fetch the specific blobs we need, not
182 the whole tree. The git object model is extremely efficient for point lookups.
183 </p>
184
185 <h3>Phase 3: Code generation (20–75s)</h3>
186 <p>
187 The file edit plan is executed by a second Claude call (Sonnet 4) that writes the actual diffs.
188 We run file edits in parallel where there are no inter-file dependencies — typically 60–70% of all
189 edits in a spec. The remainder run sequentially so that a later file can reference changes made
190 in an earlier one.
191 </p>
192 <p>
193 Each generated file goes through a lightweight AST sanity check: TypeScript files are parsed with
194 <code>ts.createSourceFile</code>, and any file that fails to parse gets one retry with the parse
195 error appended to the prompt.
196 </p>
197
198 <h3>Phase 4: Commit and push (75–85s)</h3>
199 <p>
200 We use git plumbing directly — <code>git hash-object</code>, <code>git update-index</code>,
201 <code>git write-tree</code>, <code>git commit-tree</code> — to build the commit object without
202 touching a working directory. The branch is created and the commit pushed atomically. This runs
203 in under 3 seconds even for large change sets.
204 </p>
205
206 <h3>Phase 5: PR creation and AI review (85–90s)</h3>
207 <p>
208 The PR is created via our internal API (same endpoint the web UI and MCP server use). Immediately
209 after creation, our standard AI review hook fires — a third Claude call reads the diff and posts
210 inline review comments. This is the same review that runs on every human-authored PR. Spec-generated
211 PRs get no special treatment.
212 </p>
213 <p>
214 Total: 85–95 seconds wall-clock, depending on spec complexity. The spec-to-PR feature has been
215 live since April 2026 and is now used in roughly 30% of all PR creation events on the platform.
216 </p>
217
218 <h3>What we don't do</h3>
219 <p>
220 We don't run the generated code. We don't check out a working copy. We don't call any external
221 tool-execution API. Everything happens in-process using git plumbing, Bun, and Claude. The
222 simplicity is what makes it fast.
223 </p>
224 `,
225 },
226];
227
228// ============================================================
229// Index — /blog
230// ============================================================
231
232blog.get("/blog", (c) => {
233 const user = c.get("user");
234 return c.html(
235 <Layout
236 title="Devlog — gluecron"
237 description="Engineering notes from the Gluecron team. We ship in public."
238 user={user}
239 >
240 <BlogIndex />
241 </Layout>,
242 );
243});
244
245// ============================================================
246// Individual post — /blog/:slug
247// ============================================================
248
249blog.get("/blog/:slug", (c) => {
250 const user = c.get("user");
251 const slug = c.req.param("slug");
252 const post = POSTS.find((p) => p.slug === slug);
253 if (!post) {
254 return c.html(
255 <Layout title="Post not found — gluecron" user={user}>
256 <div style="max-width:720px;margin:80px auto;padding:0 24px;text-align:center">
257 <p style="font-family:var(--font-mono);font-size:11px;text-transform:uppercase;letter-spacing:0.14em;color:var(--accent);margin-bottom:12px">
258 404
259 </p>
260 <h1 style="font-size:clamp(24px,4vw,36px);margin-bottom:16px">Post not found</h1>
261 <p style="color:var(--text-muted);margin-bottom:32px">
262 This post doesn't exist. Check the <a href="/blog">devlog index</a> for all posts.
263 </p>
264 </div>
265 </Layout>,
266 404,
267 );
268 }
269 return c.html(
270 <Layout
271 title={`${post.title} — gluecron devlog`}
272 description={post.excerpt}
273 user={user}
274 >
275 <BlogPost post={post} />
276 </Layout>,
277 );
278});
279
280// ============================================================
281// Components
282// ============================================================
283
284const BlogIndex: FC = () => (
285 <>
286 <style dangerouslySetInnerHTML={{ __html: blogCss }} />
287 <div class="blog-root">
288 <header class="blog-hero">
289 <div class="blog-hero-orb" aria-hidden="true" />
290 <div class="blog-hero-inner">
291 <div class="blog-eyebrow">
292 <span class="blog-eyebrow-pill" aria-hidden="true">
293 <svg width="11" height="11" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">
294 <path d="M12 20h9" />
295 <path d="M16.5 3.5a2.121 2.121 0 0 1 3 3L7 19l-4 1 1-4L16.5 3.5z" />
296 </svg>
297 </span>
298 Devlog
299 </div>
300 <h1 class="blog-hero-title">
301 Gluecron Devlog.{" "}
302 <span class="blog-hero-grad">We ship in public.</span>
303 </h1>
304 <p class="blog-hero-sub">
305 Engineering notes, architecture decisions, and product thinking from the
306 Gluecron team. No polish — just what we built and why.
307 </p>
308 </div>
309 </header>
310
311 <section class="blog-posts">
312 {POSTS.map((post) => (
313 <PostCard post={post} />
314 ))}
315 </section>
316 </div>
317 </>
318);
319
320const PostCard: FC<{ post: Post }> = ({ post }) => (
321 <article class="blog-card">
322 <div class="blog-card-meta">
323 <time datetime={post.dateIso} class="blog-card-date">{post.date}</time>
324 </div>
325 <h2 class="blog-card-title">
326 <a href={`/blog/${post.slug}`}>{post.title}</a>
327 </h2>
328 <p class="blog-card-excerpt">{post.excerpt}</p>
329 <a href={`/blog/${post.slug}`} class="blog-card-read" aria-label={`Read: ${post.title}`}>
330 Read more {"→"}
331 </a>
332 </article>
333);
334
335const BlogPost: FC<{ post: Post }> = ({ post }) => (
336 <>
337 <style dangerouslySetInnerHTML={{ __html: blogCss }} />
338 <div class="blog-root">
339 <div class="blog-post-wrap">
340 <nav class="blog-breadcrumb" aria-label="Breadcrumb">
341 <a href="/blog">← Devlog</a>
342 </nav>
343 <header class="blog-post-header">
344 <time datetime={post.dateIso} class="blog-post-date">{post.date}</time>
345 <h1 class="blog-post-title">{post.title}</h1>
346 <p class="blog-post-excerpt">{post.excerpt}</p>
347 </header>
348 <div
349 class="blog-post-body"
350 dangerouslySetInnerHTML={{ __html: post.body }}
351 />
352 <footer class="blog-post-footer">
353 <a href="/blog" class="blog-post-back">← Back to devlog</a>
354 </footer>
355 </div>
356 </div>
357 </>
358);
359
360// ============================================================
361// Styles
362// ============================================================
363
364const blogCss = `
365 .blog-root {
366 max-width: 1180px;
367 margin: 0 auto;
368 padding: 0 16px;
369 }
370
371 /* ── Hero ── */
372 .blog-hero {
373 position: relative;
374 text-align: center;
375 margin: var(--s-10) auto var(--s-12);
376 max-width: 820px;
377 padding: clamp(28px, 4vw, 52px) clamp(24px, 4vw, 48px);
378 background: var(--bg-elevated);
379 border: 1px solid var(--border);
380 border-radius: 22px;
381 overflow: hidden;
382 box-shadow: 0 1px 0 rgba(255,255,255,0.04), 0 22px 56px -20px rgba(0,0,0,0.45);
383 }
384 .blog-hero::before {
385 content: '';
386 position: absolute;
387 top: 0; left: 0; right: 0;
388 height: 2px;
6fd5915Claude389 background: linear-gradient(90deg, transparent 0%, #5b6ee8 30%, #5f8fa0 70%, transparent 100%);
3122762Claude390 opacity: 0.78;
391 pointer-events: none;
392 }
393 .blog-hero-orb {
394 position: absolute;
395 inset: -28% -10% auto auto;
396 width: 520px; height: 520px;
6fd5915Claude397 background: radial-gradient(circle, rgba(91,110,232,0.22), rgba(95,143,160,0.10) 45%, transparent 70%);
3122762Claude398 filter: blur(80px);
399 opacity: 0.75;
400 pointer-events: none;
401 }
402 .blog-hero-inner { position: relative; z-index: 1; }
403 .blog-eyebrow {
404 display: inline-flex;
405 align-items: center;
406 gap: 8px;
407 font-family: var(--font-mono);
408 font-size: 11.5px;
409 text-transform: uppercase;
410 letter-spacing: 0.14em;
411 color: var(--text-muted);
412 font-weight: 600;
413 margin-bottom: 14px;
414 }
415 .blog-eyebrow-pill {
416 display: inline-flex;
417 align-items: center;
418 justify-content: center;
419 width: 18px; height: 18px;
420 border-radius: 6px;
6fd5915Claude421 background: rgba(91,110,232,0.14);
422 color: #5b6ee8;
423 box-shadow: inset 0 0 0 1px rgba(91,110,232,0.35);
3122762Claude424 }
425 .blog-hero-title {
426 font-family: var(--font-display);
427 font-size: clamp(32px, 5.5vw, 64px);
428 line-height: 1.04;
429 letter-spacing: -0.036em;
430 font-weight: 800;
431 margin: 0 0 var(--s-5);
432 color: var(--text-strong);
433 }
434 .blog-hero-grad {
6fd5915Claude435 background-image: linear-gradient(135deg, #5b6ee8 0%, #5b6ee8 50%, #5f8fa0 100%);
3122762Claude436 -webkit-background-clip: text;
437 background-clip: text;
438 -webkit-text-fill-color: transparent;
439 color: transparent;
440 }
441 .blog-hero-sub {
442 font-size: clamp(15px, 1.4vw, 17px);
443 color: var(--text-muted);
444 max-width: 600px;
445 margin: 0 auto;
446 line-height: 1.6;
447 }
448
449 /* ── Post list ── */
450 .blog-posts {
451 display: flex;
452 flex-direction: column;
453 gap: 0;
454 margin: 0 auto var(--s-16);
455 max-width: 820px;
456 border: 1px solid var(--border);
457 border-radius: var(--r-lg);
458 overflow: hidden;
459 background: var(--bg-elevated);
460 }
461 .blog-card {
462 padding: var(--s-8) var(--s-8);
463 border-bottom: 1px solid var(--border-subtle);
464 transition: background var(--t-fast) var(--ease);
465 }
466 .blog-card:last-child { border-bottom: none; }
467 .blog-card:hover { background: var(--bg-hover); }
468 .blog-card-meta {
469 margin-bottom: var(--s-2);
470 }
471 .blog-card-date {
472 font-family: var(--font-mono);
473 font-size: 11px;
474 text-transform: uppercase;
475 letter-spacing: 0.12em;
476 color: var(--accent);
477 font-weight: 500;
478 }
479 .blog-card-title {
480 font-family: var(--font-display);
481 font-size: clamp(18px, 2vw, 22px);
482 font-weight: 700;
483 letter-spacing: -0.022em;
484 line-height: 1.2;
485 margin: 0 0 var(--s-3);
486 color: var(--text-strong);
487 }
488 .blog-card-title a {
489 color: inherit;
490 text-decoration: none;
491 }
492 .blog-card-title a:hover {
493 color: var(--accent-hover);
494 text-decoration: none;
495 }
496 .blog-card-excerpt {
497 font-size: var(--t-sm);
498 color: var(--text-muted);
499 line-height: 1.65;
500 margin: 0 0 var(--s-4);
501 max-width: 680px;
502 }
503 .blog-card-read {
504 display: inline-flex;
505 align-items: center;
506 font-size: 13px;
507 font-weight: 600;
508 color: var(--accent);
509 text-decoration: none;
510 transition: color var(--t-fast) var(--ease), gap var(--t-fast) var(--ease);
511 gap: 4px;
512 }
513 .blog-card-read:hover {
514 color: var(--accent-hover);
515 text-decoration: none;
516 gap: 6px;
517 }
518
519 /* ── Individual post ── */
520 .blog-post-wrap {
521 max-width: 720px;
522 margin: var(--s-10) auto var(--s-16);
523 padding: 0 16px;
524 }
525 .blog-breadcrumb {
526 margin-bottom: var(--s-7);
527 }
528 .blog-breadcrumb a {
529 font-family: var(--font-mono);
530 font-size: 12px;
531 color: var(--text-muted);
532 text-decoration: none;
533 transition: color var(--t-fast) var(--ease);
534 }
535 .blog-breadcrumb a:hover { color: var(--accent); text-decoration: none; }
536 .blog-post-header {
537 margin-bottom: var(--s-10);
538 padding-bottom: var(--s-8);
539 border-bottom: 1px solid var(--border-subtle);
540 }
541 .blog-post-date {
542 display: block;
543 font-family: var(--font-mono);
544 font-size: 11px;
545 text-transform: uppercase;
546 letter-spacing: 0.14em;
547 color: var(--accent);
548 font-weight: 500;
549 margin-bottom: var(--s-4);
550 }
551 .blog-post-title {
552 font-family: var(--font-display);
553 font-size: clamp(26px, 4vw, 44px);
554 font-weight: 800;
555 letter-spacing: -0.034em;
556 line-height: 1.08;
557 margin: 0 0 var(--s-5);
558 color: var(--text-strong);
559 }
560 .blog-post-excerpt {
561 font-size: var(--t-md);
562 color: var(--text-muted);
563 line-height: 1.65;
564 margin: 0;
565 font-style: italic;
566 }
567
568 /* ── Post body typography ── */
569 .blog-post-body {
570 font-size: 16px;
571 line-height: 1.75;
572 color: var(--text);
573 }
574 .blog-post-body p {
575 margin: 0 0 var(--s-5);
576 }
577 .blog-post-body h3 {
578 font-family: var(--font-display);
579 font-size: 20px;
580 font-weight: 700;
581 letter-spacing: -0.02em;
582 margin: var(--s-10) 0 var(--s-4);
583 color: var(--text-strong);
584 line-height: 1.2;
585 }
586 .blog-post-body ul {
587 margin: 0 0 var(--s-5) var(--s-6);
588 display: flex;
589 flex-direction: column;
590 gap: var(--s-3);
591 }
592 .blog-post-body li {
593 line-height: 1.65;
594 color: var(--text);
595 }
596 .blog-post-body li strong { color: var(--text-strong); font-weight: 600; }
597 .blog-post-body code {
598 font-family: var(--font-mono);
599 font-size: 0.88em;
600 background: var(--bg-tertiary);
601 border: 1px solid var(--border-subtle);
602 padding: 1px 6px;
603 border-radius: 4px;
604 color: var(--text);
605 }
606 .blog-post-body em { color: var(--text-muted); }
607
608 /* ── Post footer ── */
609 .blog-post-footer {
610 margin-top: var(--s-12);
611 padding-top: var(--s-8);
612 border-top: 1px solid var(--border-subtle);
613 }
614 .blog-post-back {
615 font-family: var(--font-mono);
616 font-size: 12px;
617 color: var(--text-muted);
618 text-decoration: none;
619 transition: color var(--t-fast) var(--ease);
620 }
621 .blog-post-back:hover { color: var(--accent); text-decoration: none; }
622
623 @media (max-width: 640px) {
624 .blog-hero { padding: var(--s-8) var(--s-5); }
625 .blog-card { padding: var(--s-6) var(--s-5); }
626 .blog-post-title { font-size: clamp(22px, 6vw, 32px); }
627 }
628`;
629
630export default blog;