Six Seconds and a Disclosure

You are the designer on Job Recap at Tallgrass. After a technician finishes a visit, the model drafts a plain-language summary of what was done, what was found, and what the homeowner should do next. The technician reviews it and taps Send. The homeowner receives it as a text message with a link to a web page. 

Roughly 240,000 recaps go out a month. 

How the technician’s screen works today: 

Technician taps Generate Recap at the end of a job. A full-screen overlay appears with a centered spinner and the word “Generating…”. The overlay blocks the whole app. Nothing else is tappable. When the model returns, the overlay is replaced by the draft in an editable text field, with a Send to Customer button below it. If the request fails, the overlay is replaced by a small grey toast at the bottom of the screen: “Something went wrong. Try again.” The draft field is empty. 

How the homeowner’s page looks today: 

Tallgrass logo, the contractor’s logo, the technician’s name and photo. A heading: “Your visit summary.” Four to six paragraphs of the drafted summary. A line at the very bottom, in 11px grey: “Summaries may be generated with assistance from automated tools.” A button: “Book a follow-up.” 

Ten weeks of data: 

Signal	Value
Recaps generated	240,000 / mo
Median generation latency	6.2 s
p95 generation latency	19 s
Generation requests abandoned before return (tech backgrounds the app or taps away)	22%
Generation failures (timeout or error)	4.1%
Of failures, technician retried	38%
Recaps sent without any technician edit	61%
Recaps sent with an edit of 10+ characters	39%
Homeowner opens the recap link	74%
Homeowner taps “Book a follow-up”	9%
Support tickets: “the summary said something that didn’t happen”	118 / mo
Support tickets: “recap took too long / never generated”	402 / mo

Two more facts. First, Tallgrass’s ~90 contractor customers in Ireland and the Netherlands are now covered by AI transparency rules that require the customer to be clearly informed that content was AI-generated, in a way that is clear and distinguishable — legal has told the team that an 11px grey line at the bottom of the page does not meet the bar, and has given you one release to fix it. Second, the technician is being paid per completed job and closing out a job is the last thing standing between them and the next one. 

Redesign the technician’s generation experience and the homeowner’s disclosure. You have one release. 

Constraints to consider 

• You cannot make the model faster. 6.2 s median and 19 s p95 are what they are. You may change what the interface does during, before, and after that time. 
• You cannot require the technician to edit the draft. 61% send it untouched and a mandatory edit step will be defeated by typing a space. Whatever you do about the 118 monthly “it said something that didn’t happen” tickets has to work on a technician who is not going to read carefully. 
• The disclosure must be clear and distinguishable on the homeowner page, and it must be there at first view — not behind a tap, not in a footer. This is a legal requirement, not a preference. Contractors will object that it undermines their brand; that objection does not change the requirement. 
• One release, one designer, and the mobile app ships every two weeks — the homeowner web page can ship any day. If your plan depends on both landing simultaneously, say so and say what happens if only one does. 
• No new data source and no new model call. You cannot add a second “fact check” model pass. 

AI Usage Guidance 

We expect you to use AI tools. We evaluate how you use them — not whether you use them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation. 

The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not. 

Mandatory AI question for your video: Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn’t meet the bar, what you did differently, and why. 

Communication clarity matters for this role. We assess structure, stakeholder awareness, and ability to simplify complex ideas — not presentation style or accent. 

A note on tools and fidelity. We are not scoring visual polish and we are not scoring your tool. Photographed paper sketches score exactly the same as a Figma file. What we score is whether the annotations and the critique tell us what you decided and why. Do not spend your thirty minutes on finish.