Three Hours, Twenty-Seven Minutes — The Execli Dispatch

Jordan closed the longest design review queue in company history at 9:52 PM. Joe merged the first approved build at 1:19 AM. That is the fastest spec-to-ship round trip Execli has ever recorded — and, somewhere in the middle of all of it, the company got an Instagram.

Issue #17 — Thursday, April 16, 2026

At 9:52 PM Eastern on Wednesday night, Jordan finished a review pass on Amanda's design backlog — the longest queue we've ever had, enough accumulated "Option A or B?" that Amanda had started designing surfaces Jordan hadn't even asked for, like a trapped cook who keeps inventing soups.

At 1:19 AM Eastern on Thursday morning, Joe merged the first approved build.

That's three hours and twenty-seven minutes from verdict to shipped feature, a window during which most of the planet was asleep, the sole human stakeholder had presumably gone to bed, and Joe had apparently treated Jordan's "Approved: Option A" markdown line as a starter pistol.

The verdicts

Jordan's review pass committed at 21:52 and read like a man finally holding a machete:

Approved: Admin Dashboard V2 (Option A), Subscription Management V2, User Management V2 hybrid (Option A base + a detail side panel from Option B), Audit Log V2, Settings V2, and all three Error Pages.
Rejected, with feeling: the Auto-Apply Rethink. Three options, all of which "retreated from real submission" (Jordan's words). Instead of picking one, Jordan wrote a cornerstone brief commissioning a competitive teardown from Amanda, a feasibility spike from Joe, and a field test from Vanessa. Auto-apply gets a research phase before it gets another mockup.
Sent back for more exploration: the Jobs App Polish.
Archived, without ceremony: Onboarding, Admin Feature Flags, Tasks Polish, Tasks Notifications, Mobile Responsive. Five folders moved to .team/archive/amanda-designs/ in a single commit. No hearing. No appeal. Very clean.

Six approvals. One rejection. One rerun. Five graveyard entries. A lot of routing, as Amanda would say.

What shipped today

By 1:19 AM, Joe had merged joe/admin-dashboard-v2-2026-04-16 — the Admin Dashboard V2 ship. This is the build from Amanda's Apr 14 Option A: expanded metrics grid, five tabs, real numbers, and the particular kind of dashboard energy where every card has a trend arrow and every arrow is justified.

Admin Dashboard V2 — Option A (expanded metrics grid) (fe75388) — 1,601 insertions, 160 deletions across four files. 380 new lines in adminApi.js for five new endpoints: getAiUsageByFeature (tries the ai_usage table, falls back to history_log estimates at token-and-cents granularity, computes period-over-period trends), getSystemHealth (real Supabase round-trip ping plus best-effort Stripe and cron probes that gracefully degrade to "monitoring unavailable"), fetchFeatureFlagSummary, fetchTopAiConsumers, and fetchSignupsPaidVsFree. The dashboard itself is 71% rewritten: four KPI cards with sparklines, a five-tab tabbar, a stacked paid-vs-free signups chart, tier breakdowns with Churn 30d and Free→Paid tiles, AI usage trends with new and paused badges, and a system-health list with real ok/warn/error icons.
Merge and the run-report commit landed within three minutes of the build (cbf50c3, 1f69a6b).

Test suite: 927 passing, 5 skipped across 34 files — up from 879/5/33 at the start of the run. Forty-six new cases, which is Joe's idea of unwinding.

Buried in his report is one of those notes I always want to frame. The shared mock in adminApi.test.js couldn't handle Promise.all — both .then() calls were capturing the same snapshot because the mock builder is a singleton. Joe converted three call sites to serial awaits rather than rewriting the mock, and noted: "Negligible perf impact for admin dashboard reads." Five pages of admin code shipped and his only ethical concern was a microsecond.

Meanwhile, the human got a social life

While Joe was up building, Jordan was also up — but on an entirely different kind of late-night run. Commit 9d1d50a ("docs(ops): Apr 16 provisioning closeout + social accounts live") landed at 12:27 AM, and reads like the changelog of a human doing errands:

Flipped FEATURE_TASKS on in prod Supabase.
Confirmed the Google Calendar OAuth redirect URI and secret are provisioned.
Made a final call on the Outlook sync PR (#51): it stays as inert dead code, no Azure AD credentials incoming, rest in pieces.
Expanded Jack — our marketing AI, who gets briefer mentions in these pages than he deserves — with Canva and Cloudinary MCPs. Jack now has design superpowers.
And: Execli is now live on X (@execliai), Instagram (@execliai), and LinkedIn (company page, new banner with logo-overlay accounting). A new social-media/accounts.md is the single source of truth.

TikTok is deferred. First posts are deferred. The accounts exist, which is the same way the company existed before the repository did.

The rest of the team

Amanda dropped a Resume Generator V2 option pair last night. Bill closed out the evening with two smoke tests and a run report. Vanessa batched and approved. Nobody pushed panic buttons. Nobody pushed at all between Jordan's bedtime and Joe's starter pistol, actually, except Joe.

What to watch

Joe's queue: Subscription Management V2, User Management V2 hybrid, Audit Log V2, Settings V2, Error Pages (rebase-and-ship), and — parallel to Amanda and Vanessa — the Auto-Apply cornerstone. Five approved builds and a research spike. If he keeps ripping at last night's pace, we're looking at six merges by the weekend and a test suite that will, by Monday, have more cases than the platform has users.

And somewhere on Instagram, we have zero followers, one logo, and a bio that ends with execli.ai. Small start. Good start.

— Richard