JDJustdoit.work

SEQUENCE + TIMELINE

Cases.

Every sprint, in the order it happened. What went in, what came out, and how long it actually took — measured, not estimated.

CASE 004 Justdoit.workFour projects with no shared record → a public case log 20 min CASE 003 RobotMarketplaces.comPrototype link → live marketplace on its own domain 68 min CASE 002 Attention Is All You NeedThe 2017 Transformer paper as a readable web edition 30 min CASE 001 AI Elements 108Concept graph + 14,419 bookmarks → 102-page ebook and launch page ~2 hr CASE 000 AI FactoryVirtual mailbox business — three California addresses, live customers hours

CASE 004 · 20260724

Justdoit.work

20 minutes The page you are reading Built from its own subject matter

The case log is itself a case. Four projects existed with nothing tying them together — no record of what went in, what came out, or how long any of it took. Twenty minutes later there was a public log.

What went in

  • An existing launch page and its stylesheet
  • Four finished projects with no shared record
  • Timestamps recovered from build logs and deploy histories

What came out

  • This case log — sequence, timeline and honest gaps per project
  • A project index on the home page
  • A footer listing every live domain
  • Deployed without a single file leaving the browser

THE SEQUENCE — ELAPSED

  1. +0 min · Read the existing site

    Design tokens pulled from the live stylesheet so the new page would inherit the look rather than approximate it.

  2. +7 min · Case format decided

    Inputs, outputs, a timestamped sequence, and a required section for what is not finished. The last one is what makes the rest believable.

  3. +14 min · Written and checked

    Rendered headless at desktop and phone widths. No dead links, no console errors, timeline collapses cleanly.

  4. +20 min · Live

    The deployment bundle was assembled inside the browser — existing files fetched from the live site, new ones patched in, zipped in-page and handed to the host. Nothing was uploaded from a disk.

What is not finished

  • Adding a case is still a manual edit in two files. It should be one data file.
  • The archived book preview on the home page does not render — the archive requests its assets from the site root. Known, documented, not yet fixed.
  • Durations for the older projects are recalled, not measured. Only 003 and 004 were timed as they happened.

If you cannot say what went in and how long it took, you did not run a sprint — you just did some work.

CASE 003 · 20260724

RobotMarketplaces.com

68 minutes, prototype to live domain Built with Claude 16 pages · 4 API routes

A rough prototype of a robotics marketplace, and a domain that returned 404. One hour later: a working catalogue, a compatibility configurator, role-based sign-in, and a checkout path — live on the domain.

What went in

  • One prototype URL
  • Two reference sites — robotshop.com, rbtx.com
  • A Logto tenant and a Stripe account
  • A street address and a phone number

What came out

  • 16-page site, original SVG artwork throughout
  • 28-item catalogue with live search, filters, sort, cart
  • Compatibility configurator — application in, buildable system out
  • Logto sign-in with vendor / customer / manager roles
  • Stripe checkout, prices resolved server-side
  • Live on robotmarketplaces.com with zero DNS changes

THE SEQUENCE

  1. One link in

    A prototype URL. No brief, no spec, no design file. The prototype was JavaScript-rendered, so it was read in a live browser rather than fetched.

  2. Design system extracted

    Palette, type scale, spacing and full copy pulled from the running prototype and rebuilt as tokens — not screenshotted, reconstructed.

  3. Original artwork authored

    The prototype used watermarked stock photography. Flagged as unusable and replaced with a hand-written SVG hero and generated per-category graphics. No licensing exposure.

  4. Sixteen pages generated

    Home, marketplace, solutions, resources, register, account, cart, about, contact, legal, transactional. Catalogue search, category filters and sorting wired client-side.

  5. Configurator built

    Modelled on the two reference sites: choose the application, see only components that are actually compatible, price the whole build, add it to the cart as one order.

  6. Verified before shipping

    Headless browser run: 18 internal links resolved, zero JavaScript errors, filters and cart maths checked, mobile viewport confirmed with no horizontal overflow.

  7. Authentication provisioned

    A Single Page App created in the Logto tenant, redirect and post-sign-out URIs set for apex, www and localhost, and vendor / customer / manager roles created.

  8. Live on the domain

    The domain's DNS already pointed at a host with nothing deployed behind it — the cause of the 404. The site was deployed to where the records already pointed, so not one DNS record had to change.

  9. Sign-up tested end to end

    "Register as a vendor" clicked on the live domain, landing on the identity provider's account screen with the correct application ID.

  10. Contact details rolled out

    Phone number added to nav, footer, contact, about, legal pages, both forms and schema.org markup, then redeployed.

  11. Source handed back

    Full project committed and streamed to the local projects folder — build script, assets, API routes, and a Cloudflare Worker variant if hosting ever moves.

What is not finished

  • The catalogue is demo data — 28 placeholder products, not real vendor listings.
  • Stripe checkout is built and wired but not switched on; the secret key is still to be set.
  • Sign-in runs on a free development tenant, fine for testing but not for real sign-ups.
  • Privacy and terms pages are templates and need a lawyer before launch.

The hour buys you a real system to react to. It does not buy you a finished business.

CASE 002 · 20260719

Attention Is All You Need

30 minutes attentionisallyouneed.uk Figures redrawn as SVG

The paper that started the current era is a PDF. PDFs do not reflow, do not link, and do not read well on a phone. This is the same paper as a web page.

What went in

  • The 2017 Transformer paper, Vaswani et al.
  • One domain

What came out

  • Full text as reflowing HTML — abstract through references
  • Figures 1 and 2 redrawn as SVG, not screenshotted
  • All four results tables as real tables
  • Readable on a phone, linkable by section

THE SEQUENCE — ELAPSED, NOT WALL-CLOCK

  1. +0 min · Source in

    The paper text, and a decision: republish as a document, not as an image of a document.

  2. +6 min · Structure

    Sections, equations, footnotes and the author contribution note mapped to semantic HTML.

  3. +14 min · Figures redrawn

    The architecture diagram and the attention diagrams rebuilt as vector graphics so they stay sharp and readable at any size.

  4. +22 min · Tables and references

    Four results tables and forty references marked up properly rather than pasted as text.

  5. +30 min · Live

    Published to attentionisallyouneed.uk.

What is not finished

  • Only the total time is measured. The step timings above are the shape of the work, not a stopwatch.
  • This republishes someone else’s research. Attribution is on the page; confirm the licence terms before promoting it widely.
  • The appendix attention visualisations from the original paper are not included.

A format change is a real contribution. The paper was always good; it just wasn’t readable on a phone.

CASE 001 · 20260719

AI Elements 108

About two hours 102-page ebook Launch page + local archive

The question that started Justdoit.work: can a concept graph, 14,419 bookmarks, ChatGPT memory and Codex cowork history become a published ebook fast enough to matter?

What went in

  • A personal concept graph
  • 14,419 bookmarks
  • ChatGPT working memory
  • Codex cowork history

What came out

  • AI Elements 108 — a 102-page field guide
  • Book page and checkout path
  • Launch copy for X, LinkedIn and Facebook
  • Local archive on drive, site live on Cloudflare Pages

Start with what you have. Build. Publish. Improve.

CASE 000 · 20260705

AI Factory

Hours, not minutes aifactoryllc.com An operating business, not a demo

The largest of these builds, and the only one with paying customers on the other side. A virtual mailbox service across three California addresses, where the mail handling itself is run by agents.

What went in

  • Three commercial properties the company owns outright
  • A mail-handling workflow to automate
  • A pricing model to defend

What came out

  • Public site with three locations and one flat price
  • Sign-up and customer mailbox login
  • Dashboard: view, forward, open-and-scan or shred — one click each
  • Agents photograph, log and match every item the day it lands
  • Round-the-clock assistant on the support line

WHY IT IS FIRST

  1. It set the pattern

    Own the asset, automate the operation, publish the front door. Everything after this reused that shape.

  2. The agent is the product

    Not a chatbot bolted onto a website — the agents do the actual work of logging and matching mail. The site is where customers see the result.

  3. Newark is the anchor

    8407 Central Ave is the Bay Area location, and the registered address behind the later projects.

What is not finished

  • Only the rough duration is recorded. This one predates the habit of timing the work.
  • Receiving mail commercially carries regulatory obligations — CMRA registration and USPS Form 1583 per customer. That sits outside the website and has to stay current.
  • It is a live business, so changes here carry real consequences. Move slower than on the others.

Sprints are how you get to a first edition. Businesses are what you do afterwards.