Skip to content
refs
REF-1908Source onlineBlueprint rev 01

Computer Agent Completes a Vendor Form

A presenter sets up a cross-app task, then a long desktop recording proves the agent can find missing data and submit the form.

Maker @anthropic-aiFor AnthropicOriginal post ↗
Cut timeline · 14 shots02:03
00:0000:3001:0001:3002:00

Field notes / why it works

The task is specific by 0:29, and the most-replayed section at 0:38 shows its exact instruction over a spreadsheet and blank form. A 45.25-second desktop shot sustains attention through about 14 UI changes, including a failed lookup and fallback, before a visible successful submission around 1:47.

01

Plate 01 / The format in one breath

A researcher-like presenter briefly explains an automation task, then yields the frame to a real desktop recording. The screen proof shows an instruction, a failed first lookup, a second source, field transfer, and a successful submission. A short presenter return interprets the result and closes on availability.

Why it works

  • The promise is concrete by 0:29: complete a vendor form using information scattered across apps, so viewers know what success will look like.
  • The most-replayed span, 0:38.1–0:39.4, lands exactly as the instruction appears over the spreadsheet and blank form. Viewers can inspect the task themselves.
  • From 0:55 to 1:40, one desktop shot stays up for 45.25 seconds while about 14 internal visual steps show progress. The eventual submission at 1:47 pays off the wait.
02

Plate 02 / Format card

Reference
Platform · aspect · lengthYouTube · 16:9 · 2:02.9 · source 640×360 at 23.98 fps
Pace14 shots, 13 confirmed cuts, 6.3–7.3 cuts/min including two uncertain changes; 23.4 total picture changes/min; mean shot 8.78 s, median 3.94 s
Script327 spoken words, 164 wpm across full runtime, about 230 wpm within speech; 14 long sentences; no spoken questions
VoiceCalm, low-register, conversational technical presenter; analytical rather than salesy
On screenRoughly 35% presenter, 57% desktop workflow, 8% disclaimer/end card (inferred from shot durations)
CaptionsNo ongoing subtitles; small presenter lower-third, a dark disclaimer card, and native text inside the desktop UI
SoundSpeech-led with a subdued tonal bed (music classification inferred); no clear accent on most cuts; measured −14.7 LUFS integrated
StructureIntroduction → task setup → desktop proof → completed form → availability/forward-looking close; result at about 1:47

Pro / 13 plates + recipe

The first plates are open. The full dissection is Pro.

Get the timed shots, type and sound specs, and the runnable recipe.

Open the full blueprint
  1. Plate 03 / The hook, frame by frame (first 3 seconds)Pro
  2. Plate 04 / Structure (beat sheet)Pro
  3. Plate 05 / Shot grammarPro
  4. Plate 06 / Visual style guidePro
  5. Plate 07 / Voice and deliveryPro
  6. Plate 08 / Sound designPro
  7. Plate 09 / PackagingPro
  8. Plate 10 / Script templatePro
  9. Plate 11 / Production planPro
  10. Plate 12 / PromptsPro
  11. Plate 13 / Edit recipePro
  12. Plate 14 / QA checklistPro
  13. Plate 15 / Make it yoursPro
  14. Recipe / runnable promptPro