The Autonomy Ladder

Tallgrass leadership wants to move from suggesting to acting. The proposal on the table, for the next two quarters: 

1. Job Recap sends itself. No technician review. The summary goes to the homeowner automatically when the job is marked complete. 
2. Smart Assign assigns. New jobs are placed on the board automatically; the dispatcher can move them. 
3. Snap Quote quotes. For jobs under $500 on equipment the model identifies with high confidence, the quote is sent to the homeowner without a technician tap. 
4. Reschedule handles itself. When a technician runs 45+ minutes late, the system texts the affected homeowner, offers two new windows, and rebooks — no human involved. 

You are the design lead. The CEO wants one rule the whole company can apply: when does the system act on its own, when does it ask, and when does it refuse to act at all? You are being asked for the ladder, not for four separate features. 

What you know: 

• Job Recap: 118 monthly “it said something that didn’t happen” tickets against 240,000 recaps. 61% currently sent unedited — so most of them already effectively have no human review, just a tap. 
• Smart Assign: 84% first-visit completion when followed vs 79% when overridden. Dispatchers currently accept 12% via the affordance and independently choose the same tech another 31% of the time. 
• Snap Quote: 71% correct identification overall, 58% in poor light. 1 in 9 quotes triggers a correction call to the homeowner. 
• Reschedule: no model accuracy data — the feature does not exist yet. Late-running jobs happen on about 14% of visits. 
• Cost: each autonomous action costs Tallgrass roughly $0.02–$0.09 in inference. Volume is high enough that reversing a bad one costs vastly more than making it — a wrongly sent quote averages $180 in contractor time and goodwill credits, per the support team. 
• Legal has signed off on autonomous customer messaging only if every customer-visible artifact carries a clear AI disclosure at first view and there is a documented human-reachable path within one tap. 
• The four largest contractor accounts have already asked whether Smart Assign can be turned off. 

Write the ladder and design the two states that prove it works. 

Constraints to consider 

• One rule, applied to all four features. You may set different features at different rungs, but the rule that decides the rung must be the same rule, stated once, and you must apply it to all four. 
• Every autonomous customer-visible action requires an AI disclosure at first view and a human-reachable path within one tap. Non-negotiable, no launch-then-fix. 
• Reversal must be designed, not assumed. For any action you let the system take alone, you must show what undo looks like to both the internal user and the customer — including the case where the customer has already seen the wrong thing. 
• The contractor owns the customer relationship, not Tallgrass. Any autonomous message goes out under the contractor’s brand. A contractor must be able to see and change the setting; you cannot design a Tallgrass-only global switch. 
• Thirty minutes. You are producing a ladder plus two annotated states — not four feature designs. 

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. The ladder can be a hand-drawn table. What we score is the rule, the reasoning, and whether the two states you chose actually prove the rule works. Do not spend your thirty minutes on finish.