Skip to content

Build1 publisher3 min readPublished

A vague asset request postpones the build Claude Code finished by evening

A developer building small client sites says the typing fell from five days to one while delivery still takes six weeks. He rewrote the kickoff email so each line names a file format and a minimum size.

The Engineer · Build desk

Photograph accompanying A vague asset request postpones the build Claude Code finished by evening
Photo: appwrite.io

What happened

  • A five-page client site used to be a few days of typing, and the developer now describes it to Claude Code and says most of it exists by the evening.
  • The build fell from five days to one and the project still takes six weeks, with the wait sitting in a logo folder nobody has opened since 2019 and a hosting password from two providers ago.
  • His original kickoff email asked for a logo, some copy, a few photos and hosting access, four items he budgeted at ten minutes of client work.
  • The rewritten request gives one item per line in a format the client can picture, including "SVG or PNG at least 1000px wide" and a paragraph from an earlier site offered as a shape for the copy.
  • Anything that does not block the start, such as analytics IDs and the final photo round, is left out of the first ask and wired up as a placeholder.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint With one day of build inside a 30-working-day calendar, further gains in code generation buy hours off a delivery date that is set by someone else's inbox.
  • decision The kickoff request is now the scheduling artifact of the engagement, so schedule risk gets managed in the drafting of that email.
  • exposure Live credentials mailed in plaintext outlive the engagement in both parties' backups. The developer's inbox stays a standing liability for clients he no longer works for.
  • cost Running the checklist, the scheduled nudges and the secure links by hand costs coordination labour per client, and it caps the practice at roughly three concurrent projects.

The reason the rewritten request works is in the shape of the task. "Some copy about the company" is not a task, it is a writing assignment with no brief and no end, the developer wrote [7]. "A few photos" is a decision about which photos [8]. Every line expanded into a small research project on the client's side with no visible finish line [9], and it competed with everything actually due that day and lost [10].

His summary of the change: "a specific request can be finished. A vague one can only be postponed, because you're never sure you're done" [15].

Six weeks at five working days each is 30 days [1]. One of them is the build [2]. The four days that codegen removed are about 13 percent of that calendar [2], and the other 29 sit with the client [1][3]. If the next model halves the build, the delivery date moves by half a day [3].

The pixel floor does a second job. Naming "at least 1000px wide" skips the round trip where a 200px logo arrives and he has to ask again, and asking again is what makes a client feel the project is dragging even when it is their own file that is late [17].

The follow-up schedule is stated in the first message: "I'll check in Tuesday and Friday until everything's in" [21]. It runs on a calendar, so it actually happens, and he reports that nobody has been annoyed by it because they agreed to it [22]. The reminder lists only what is still outstanding, since a client who already sent the logo and gets asked for it again reasonably concludes you did not read their email [23].

Credentials get separate handling. A WordPress admin password pasted into an email thread sits in plaintext in two inboxes, and in every backup of both, indefinitely, long after the project is over [24]. "I've been handed live production credentials over Gmail more times than I can count," he wrote [25]. A password manager's one-time share link costs the client one extra click and takes the credential out of permanent storage in both places [26].

This is one practitioner's account of his own pipeline. The post says the difference in response time was bigger than he expected and does not give figures [16], so what is on offer is a causal story about five-page marketing sites, not a measurement. For the format to transfer, the assets have to exist somewhere already, and the first ask has to be short enough to finish in one sitting: he writes that five items get finished and twenty get postponed [18].

He ran all of this by hand for a long time, with a checklist template, a calendar reminder to follow up and a separate secure link for logins [27]. It works, he wrote, and stops scaling at about three simultaneous projects [27].

What to watch

  • Whether the developer publishes before-and-after response times for the rewritten request. The post does not include them.
  • Whether the manual process gets tooled, since the available text breaks off at the point where it stops scaling past about three simultaneous projects.
  • Whether clients accept one-time share links from a password manager they do not already use. That is the only step that adds work on their side.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories