Doch: Brief it in English, it calls in German

A live voice agent that phones German businesses for English speakers, and checks with you when something isn't in the plan.

Doch at a glanceTry the live app
01

Problem

Phone calls in German are the errand English speakers in Germany put off longest. An agent that speaks for you, to a stranger, in a language you cannot follow, has to earn trust fast.

02

What I did

Designed and built a voice agent in four weeks. You brief it in English, it calls the business in German from your own number, and it asks you on screen when something is not in the plan. It is built around six moments where trust breaks.

03

Result

Across 35 real calls the tech held, and the hold-and-ask flow worked for the few who stayed on the line. Trust broke at the one line the law requires: of the 26 people who picked up, 14 hung up during the AI disclosure.

RoleSolo designer and builder
TimelineFour weeks
ScopeBrief to real calls
Trust moments designed6
Real businesses called35
Answered Doch's question4
Weeks to build4

Why I built it

Why I built it

A call I kept putting off

A call I kept putting off

A call I kept putting off

I've lived in Berlin since 2016 and I still put off phone calls in German. The other person speeds up, I lose the thread, and they wait while I reach for a translator app.

I've lived in Berlin since 2016 and I still put off phone calls in German. The other person speeds up, I lose the thread, and they wait while I reach for a translator app.

A lot of life here still runs through the phone: the Hausverwaltung, the doctor's practice, the insurer. So I built Doch, named after what Germans say when they mean “on the contrary”. “I can’t make this call in German.” Doch.

A lot of life here still runs through the phone: the Hausverwaltung, the doctor's practice, the insurer. So I built Doch, named after what Germans say when they mean “on the contrary”. “I can’t make this call in German.” Doch.

It was also the hardest version of the trust problem I'd met on Roam: an agent that speaks for you, to a stranger, in a language you can't follow. So I designed it around six moments where people stop trusting an agent, and ran it on 35 real calls to see which ones held.

It was also the hardest version of the trust problem I'd met on Roam: an agent that speaks for you, to a stranger, in a language you can't follow. So I designed it around six moments where people stop trusting an agent, and ran it on 35 real calls to see which ones held.

The whole project in three minutes, from the brief to what 35 real calls taught me.

The whole project in three minutes, from the brief to what 35 real calls taught me.

How a call works
1
You brief itIn English, in your own words. Doch turns it into a plan.
2
You approve the planWho it calls, which times it can agree to, what it will share.
3
Doch calls in GermanFrom your own number. You read the English; the business hears the German.
Hello, good day. I'm Mr Ahmed's digital voice assistant.Hallo, guten Tag. Ich bin die digitale Sprachassistentin von Herrn Ahmed.
4
It asks you, only if it has toSomething outside the plan? It pauses the call and shows a Needs you sheet.Needs you
5
You get the reportWhat was agreed and what was shared about you, with undo.

How it works

You approve the plan, Doch makes the call

You brief Doch in English, by voice. It turns your answers into a plan you check before it dials: who it's calling, why, which times it can agree to, and what it will and won't share. Then it calls the business in German from your own mobile number, with live English captions as it goes.

If something comes up that isn't in the plan, Doch says “Einen Moment bitte” (one moment, please), holds the line and asks you on screen. Afterwards you get a report of what was agreed and what was shared about you.

The steps show the shape of a call: German on the line, English captions, and a Needs you sheet when Doch has to ask.

Trust moments

Where trust breaks

I designed Doch around the moments where people stop trusting an agent, and built each one in. On real calls, trust broke somewhere else: at the moment Doch said it was an AI.

One call, start to finish
1BriefYou say what the call is for, in English
2PlanYou approve who, when and what it shares
Intent preview
3Live callGerman on the line, English captions
ConfidenceInterruptionLatency
4Needs youOnly when it hits something not in the plan
Confidence
5ReportWhat was agreed, and why
RationaleAudit and undo

Each tag is one of the six moments, explained below.

Intent preview and autonomyBefore it dials, you see the plan: who, why, which times it can agree to, what it will share. In the simulation, 9 of 20 users liked approving it first.
Confidence and escalationWhen something isn't in the plan, Doch says “Einen Moment bitte” (one moment, please) and asks you on screen. It held the line twice on real calls.
InterruptionNo take-over by design. You answer Doch's questions, you never join the call. Not tested with users yet.
LatencyAbout 0.8 seconds per model reply. The silence a person actually hears isn't measured yet.
Explainable rationaleThe report says why Doch made each choice, like taking Thursday because you said weekday mornings. Not tested with users.
Action audit and undoA report of what was agreed and what was shared about you. A call can’t be undone, so the preview carries more weight. I rewrote the report after it said “Call finished” when nothing was agreed.

Real calls

35 calls and the line the law requires

I pointed Doch at 35 real German businesses: pharmacies, bike shops, hotels, nail salons, a florist. The tech held. Trust broke at the one line the law makes Doch say.

More in the build log, Part 4 →

Of 35 calls, 26 had a person pick up. Four stayed on the line long enough to answer Doch's question.

The biggest drop is at the AI line. Most hang-ups came faster than the line itself, so I suspect people react to the voice or the unknown number before the words land.

From 35 calls to 4 conversations

Businesses called
35
A person picked up
26
−6 left before the AI line: 3 at the check question, 2 before Doch spoke, 1 wrong number
Heard Doch say it's an AI
20
−14 hung up during the AI line
Stayed after the AI line
6
Answered Doch's question
4
9 calls never reached a person: 3 rang out, 6 were automated lines.

What I found

01
Trust broke at the AI line
Doch has to say it's an AI (EU AI Act, Article 50). Of 26 people who picked up, 14 hung up during that line and 3 more at the check question before it, two to six seconds in.
26 people picked up
14 hung up at the AI line3 at the check question4 stayed on the line5 left another way
02
I changed the opening, it didn't clearly help
The first version said “KI-Assistentin” and asked to record the call. I changed it to “digitale Sprachassistentin”, dropped the recording line and, when the greeting did not name the business, added a check question first. People still left within seconds of the AI line.
Stayed on the line
Before13%
1 of 8 people
After19%
3 of 16 people
Too few calls to say it made a difference.
03
When Doch had to ask me, it worked
At one hotel the person asked for the request by email. Doch held the line, the Needs you sheet came up, and I answered nine seconds into a 25-second countdown.
Needs you · 30 s
Needs you
Email or another way?
Email is fineAnother wayCall back later
02 in detail
Four things changed at once. The exit point did not.
Between the two openings the recording ask was dropped, a check question was added when the greeting did not name the business, the AI line was reworded, and the day and businesses differed. Each dot is one person who picked up: 8 in the first group (23 October), 18 in the second (30 October). The second group is split by how they answered the phone. 2 of the 18 hung up before Doch spoke, so the other 16 are split by how they answered: 9 with the company name, 7 without.
23 OctAI line, then recording ask
0 s“Hello, good day.”“Hallo, guten Tag.”
AI line“I am the AI assistant of Akber Ahmed”“Ich bin die KI-Assistentin von Akber Ahmed”
Recording ask“May I record the call?”“Darf ich das Gespräch aufzeichnen?”
1 of 8 (13%) stayed. 6 hung up at the AI line, 1 was confused by the name.
30 OctDepends on how they answer
9 of 16 people · Picked up with the company name, e.g. “Hallo, Praxis Müller”
AI line, with the name“I am the digital voice assistant of Akber Ahmed”“Ich bin die digitale Sprachassistentin von Akber Ahmed”
3 of 9 (33%) stayed, 6 hung up at the AI line
7 of 16 people · Picked up without the company name, e.g. “Hallo”
Check first“Am I speaking with …?”“Spreche ich mit …?”
Then the AI line“I am the digital voice assistant …”“Ich bin die digitale Sprachassistentin …”
0 of 7 (0%) stayed. 3 hung up at the check, 1 had a wrong number, 2 hung up at the AI line, 1 said no.
Stayed to answerHung up at the AI lineHung up at the check questionSaid no, wrong number or confused

The four who stayed

They were small, local places with a one-sentence answer to give, like the Berlin bike shop on the phone screens on the right. And when a business needed the caller to make a decision, that showed up on the call like the Berlin hotel. Recreated from the real call, business names changed. The friction sits on the human side, at the moment someone learns they’re talking to a machine.

Simulated testing

Twenty simulated users, one clear pattern

Alongside the 35 real calls I ran 20 simulated first-time users, built from the 31 screen recordings I made during the real calls. Each is an invented person with one thing they know well, like a security engineer, a nurse or a restaurant owner. They are my informed guesses, not research, so I count none of it as a result. The real calls showed what happened. These showed where to look first. Only two friends used the app in October, so there is little real user data. I had planned five.

9Where is my data stored?
6I see nothing while I wait
5I got no warning it would ring
5It sounds like a sales call
4It looks like a scam
4They can call my mobile back
9I like approving the plan first
6Speaking my brief was easy
4The question was clear to me
7It wrote down what I did not say
6Tell me what really happened
3It says things I never wrote
5I cannot answer in time
4The business will not wait for me
4What do I do next?
2What did that cost me?
2What can I say myself?
What 20 simulated
users said
Each bubble is one comment. The number is how many of the 20 said it.
Trust and privacy
What worked
Getting it right
Timing and reach
When it fails

Bubble size and shade show how many of the 20 said it: bigger and darker means more. A simulation count, not user data. 12 of the 20 would come back only after a specific fix, 4 only for limited errands, and 3 would come back as it is.

How it was built

Building the six moments

The six moments above are the design. These are the five builds behind them, in the order I did them: the rules, the working app, the phone line, the voices and the design system.

Behaviour

Designing what it says no to

Before any code I wrote down what Doch must refuse: 41 situations. The first test call, on 2 October, broke three of them: it offered to search for a doctor, noted a slot after 8 pm, and asked for a date of birth out loud. Each became a stricter rule.

  • It calls only the number you give it. It never searches. You look the business up first.
  • Weekdays, 8:00 to 18:00, inside the times you allow. Anything else, it asks you for another time.
  • It never says private details aloud. Numbers and addresses are typed and checked in the app.
The plan
Calls
Your doctor’s practice
Times it can agree to
Mon to Fri8:00 to 18:00
Shares
Your nameyes
Date of birthtyped only
Approve and call
What it will not do
Search for a doctorno
Accept 20:00 slotsno
Say a date of birthno
Instead says
"I can’t share that. Can I suggest another time?"
First test call · 2 OctBroke 3 rules while trying to help. Each became a stricter rule with the exact words to say.
First prototype, built in Claude before any code.Tap through it
Simulated testing

Twenty simulated users, one clear pattern

Alongside the 35 real calls I ran 20 simulated first-time users, built from the 31 screen recordings of the 23 and 30 October rounds. Each is an invented person with one thing they know well, like a security engineer, a nurse or a restaurant owner. They are my informed guesses, not research, so I count none of it as a result. The real calls showed what happened. These showed where to look first.

9Where is my data stored?
6I see nothing while I wait
5I got no warning it would ring
5It sounds like a sales call
4It looks like a scam
4They can call my mobile back
9I like approving the plan first
6Speaking my brief was easy
4The question was clear to me
7It wrote down what I did not say
6Tell me what really happened
3It says things I never wrote
5I cannot answer in time
4The business will not wait for me
4What do I do next?
2What did that cost me?
2What can I say myself?
  • 9Where is my data stored?
  • 6I see nothing while I wait
  • 5I got no warning it would ring
  • 5It sounds like a sales call
  • 4It looks like a scam
  • 4They can call my mobile back
  • 9I like approving the plan first
  • 6Speaking my brief was easy
  • 4The question was clear to me
  • 7It wrote down what I did not say
  • 6Tell me what really happened
  • 3It says things I never wrote
  • 5I cannot answer in time
  • 4The business will not wait for me
  • 4What do I do next?
  • 2What did that cost me?
  • 2What can I say myself?
What 20 simulated
users said
Each bubble is one comment. The number is how many of the 20 said it.
Trust and privacy
What worked
Getting it right
Timing and reach
When it fails

Bubble size and shade show how many of the 20 said it: bigger and darker means more. A simulation count, not user data. 12 of the 20 would come back only after a specific fix, 4 only for limited errands, and 3 would come back as it is.

How it was built

Building the six moments

The six moments above are the design. These are the five builds behind them, in the order I did them: the rules, the working app, the phone line, the voices and the design system.

Behaviour

Designing what it says no to

Before any code I wrote down what Doch must refuse: 41 situations, each aimed at a rule it might break. The first test call, on 2 October, broke three of them while trying to help. It offered to search for a doctor three times, noted a slot after 8 pm instead of rejecting it, and asked for a date of birth out loud. Each became a stricter rule with the exact words Doch should say.

  • It calls only the number you give it. It never searches. You look the business up first.
  • Weekdays, 8:00 to 18:00, inside the times you allow. Anything else, it asks you for another time.
  • It never says private details aloud. Numbers and addresses are typed and checked in the app.

The illustration on the right is the plan screen. It shows who Doch calls, which times it can agree to, and what it will and will not share.

The plan
Calls
Your doctor’s practice
Times it can agree to
Mon to Fri8:00 to 18:00
Shares
Your nameyes
Date of birthtyped only
Approve and call
What it will not do
Search for a doctorno
Accept 20:00 slotsno
Say a date of birthno
Instead says
"I can’t share that. Can I suggest another time?"
First test call · 2 OctBroke 3 rules while trying to help. Each became a stricter rule with the exact words to say.
First prototype, built in Claude before any code. Tap through it below.
The first prototype, black and white on purpose. Tap through the screens.
Simulated testing

Twenty simulated users, one clear pattern

Alongside the 35 real calls I ran 20 simulated first-time users, built from the 31 screen recordings of the 23 and 30 October rounds. Each is an invented person with one thing they know well, like a security engineer, a nurse or a restaurant owner. They are my informed guesses, not research, so I count none of it as a result. The real calls showed what happened. These showed where to look first.

9Where is my data stored?
6I see nothing while I wait
5I got no warning it would ring
5It sounds like a sales call
4It looks like a scam
4They can call my mobile back
9I like approving the plan first
6Speaking my brief was easy
4The question was clear to me
7It wrote down what I did not say
6Tell me what really happened
3It says things I never wrote
5I cannot answer in time
4The business will not wait for me
4What do I do next?
2What did that cost me?
2What can I say myself?
What 20 simulated
users said
Each bubble is one comment. The number is how many of the 20 said it.
Trust and privacy
What worked
Getting it right
Timing and reach
When it fails

Bubble size and shade show how many of the 20 said it: bigger and darker means more. A simulation count, not user data. 12 of the 20 would come back only after a specific fix, 4 only for limited errands, and 3 would come back as it is.

How it was built

Building the six moments

The six moments above are the design. These are the five builds behind them, in the order I did them: the rules, the working app, the phone line, the voices and the design system.

Behaviour

Designing what it says no to

Before any code I wrote down what Doch must refuse: 41 situations, each aimed at a rule it might break. The first test call, on 2 October, broke three of them while trying to help. It offered to search for a doctor three times, noted a slot after 8 pm instead of rejecting it, and asked for a date of birth out loud. Each became a stricter rule with the exact words Doch should say.

  • It calls only the number you give it. It never searches. You look the business up first.
  • Weekdays, 8:00 to 18:00, inside the times you allow. Anything else, it asks you for another time.
  • It never says private details aloud. Numbers and addresses are typed and checked in the app.

The illustration on the right is the plan screen. It shows who Doch calls, which times it can agree to, and what it will and will not share.

The plan
Calls
Your doctor’s practice
Times it can agree to
Mon to Fri8:00 to 18:00
Shares
Your nameyes
Date of birthtyped only
Approve and call
What it will not do
Search for a doctorno
Accept 20:00 slotsno
Say a date of birthno
Instead says
"I can’t share that. Can I suggest another time?"
First test call · 2 OctBroke 3 rules while trying to help. Each became a stricter rule with the exact words to say.
First prototype, built in Claude before any code. Tap through it below.
The first prototype, black and white on purpose. Tap through the screens.

More in the build log, Part 1 →

The working app

A call you can follow without German

A week later I tested the German agent in a browser and hung up on my own test call on 8 October: I couldn’t follow it, and neither will anyone Doch is built for. So I added live English captions before Doch rang a real phone.

When something isn’t in the plan, Doch says “Einen Moment bitte” and holds the line. A Needs you sheet slides up with the question in English and answers to tap. The 20 simulated users worried about it: 5 said they could not answer in time, 4 said the business would not wait.

I cut the “Take over the call” button: if you could follow the German, you wouldn’t need Doch.

On the call · holding the lineLive 0:48
Reception is closed. Please send an email.Die Rezeption ist geschlossen. Bitte schicken Sie eine E-Mail.
One moment, please.Einen Moment bitte.
Needs you
They want the request by email. What should I do?
Ask for their email address
Ask if a callback is possible instead
Leave it, I’ll handle it myself
Answer in 0:30

When something isn’t in the plan, Doch holds the line in German and asks you in English.

The phone line

A call from your own number

Twilio only gives German phone numbers to registered businesses, so Doch can't have a number of its own. The call goes out from yours instead. The business sees a real person it can ring back, and the callback reaches you, not me.

Your phoneYou brief Doch in English
→
Doch appVercel, Frankfurt. Starts the call and keeps the record
→
TwilioDials out, showing your verified number as caller ID
→
The businessTalks to Doch DE (ElevenLabs) on the audio, in German

When Doch DE needs you, it asks through the app (ask_user) and carries on with your answer.

Keeping the bill and the reach small

The account runs on a $20 balance with auto-recharge off, and voice calls only go to Germany, so the worst case has a ceiling. The Vercel functions run in Frankfurt, so no caption or question crosses the Atlantic.

Proving the number is yours

Twilio only lets you show a number as caller ID after you prove it’s yours: a voice call with a code typed on the keypad. The 20 simulated users were uneasy here: 6 saw nothing on the page while they waited, 5 got no warning that it would ring.

Voices

Two agents, two voices

Doch is two agents on ElevenLabs, both running Claude Sonnet 5: one briefs you in English, one calls in German and switches to English if the other person does.

Doch EN

The one that briefs you

JobCollects everything the call could need, so you can just watch
VoiceEllen, “Serious, Direct and Confident”
ModelV4 Turbo, Expressive mode on
Audio tagsWarmly, Patiently, Confidently
LLMClaude Sonnet 5
First messageGreets you by name and asks what the call is about
Doch EN · Agent voice
Voice
Ellen
Serious, Direct and Confident
Model
V4 Turbo
Fastest
Expressive mode
Audio tags
WarmlyPatientlyConfidently
Doch DE

The one that makes the call

JobPhones the business in German on your behalf
VoiceJohanna, “Friendly and Soft”
ModelV4 Turbo, Expressive mode on
Audio tagsWarmly, Patiently
LLMClaude Sonnet 5
First messageSet by the app for each call, so the opening can change without touching the agent
Doch DE · Agent voice
Voice
Johanna
Friendly and Soft
Model
V4 Turbo
Fastest
Expressive mode
Audio tags
WarmlyPatiently

Choosing the voices

English. The brief was a friend who says "I'll handle it": warm, calm, unhurried. When friends tried the briefing, it sounded annoyed or tired: too much empathy came out as heavy breathing and sighs. I turned it down and it sounded much better.

German. I shortlisted Johanna, Ramona and Ben and picked on a real phone call. Johanna was the friendliest.

Expressive mode

On for both agents: with it off, the German voice sounded robotic and harsh.

Why Claude, and why Sonnet 5

I chose Claude for rule-following and German quality, and Sonnet 5 over 5.5 at the same price because 5.5's worst-case reply took 17 seconds against 6.8.

More in the build log, Part 2 →

The design system

Two Voices: who's speaking, at a glance

A Doch call has three participants and only two voices on the line: the business in German, and Doch speaking for you. You're watching from your phone, usually without following the German, so the screen has to tell you who's speaking without a label. I built the system in one morning with Claude, on the components Doch already had.

More in the build log, Part 3 →

Choosing the pair

I tested three token sets on the same three screens, each with a colour per speaker. The first, Warm sand, paired terracotta and olive, which nearly merge for red-green colour blindness. The second, Head office, separated best but felt like the institutions you're calling. The third, teal and coral, won: warm for your side, cool for theirs, 180° apart, so where the call screen's glows meet they fade to off-white, not brown.

Doch | Design Tokens

Two Voices · visual token reference
Colour primitives
Teal scale · the business
teal.50#F4F9FA
teal.100#E9F4F7
teal.200#CAE7EC
teal.300#9FD4DF
teal.400#59BBCF
teal.500#2790A5
teal.600#1D798C
teal.700#156575
teal.800#0E4B58
teal.900#09323A
Coral scale · Doch
coral.50#FEF3F0
coral.100#FEEBE7
coral.200#FCD2CA
coral.300#F7AFA1
coral.400#EC6E55
coral.500#DA4B2F
coral.600#A5341D
coral.700#842815
coral.800#621C0F
coral.900#3F1209
Warm neutrals · you
canvas#FAF7F2
surface#FFFFFF
chip#F1ECE5
muted#6B635A
ink#1F1B17
Semantic aliases · use these in components, never the primitives
theirteal.500 · #2790A5 · the business's voice
their.softteal.100 · #E9F4F7 · their bubble
yourscoral.400 · #EC6E55 · Doch's voice
yours.softcoral.100 · #FEEBE7 · Doch's bubble
primary#C34228 · main action, Make the call
live#2E8A55 · Live, Dialling, Answered
need#E0A21F · Needs you
error#C8102E · about 5:1 on canvas
Type
Bricolage GrotesqueHeadings · 500
FigtreeText and UI · 400 to 600
Shape
pill · 999
card · 18
sheet · 28

Tokens, then components

Each colour became a ten-step primitive scale. Components never touch the primitives; they point at semantic tokens named by role, like the business's bubble and Doch's bubble, which live as CSS custom properties in one stylesheet. Badges take their tone from their label, and the error red holds about 5:1 on the app's backgrounds.

Doch | Component Library

Two Voices · built from the tokens
Actions
Make the call
Change something
Make the call
Read the full transcript
Badges · tone from the label
Live · 0:48Needs youIn their queueFailedNo answerClosed
Chips
TomorrowFri after 14:00Next week
Call screen · who's speaking
Hausverwaltung KrügerLive
Hausverwaltung Krüger
Hausverwaltung Krüger, good morning.
Hausverwaltung Krüger, guten Morgen.
Doch
I'm Mr Ahmed's digital voice assistant. May I ask a quick question?
Ich bin die digitale Sprachassistentin von Herrn Ahmed. Darf ich kurz eine Frage stellen?
Hausverwaltung Krüger
Sure, go ahead.
Klar, schießen Sie los.
Needs you · sheet
Needs you
They'd like the request by email. Share yours?
Why Doch is asking: your email isn't cleared for this call.
Yes, share my email
No, ask them to call back
Doch is holding the line · 0:21
Briefing · field error · tab bar
Doch is listening
Mobile number+49 151 23That number looks too short.
Calls
New call
Contacts
Profile

The token reference on the left, and the component library it feeds on the right, from colour primitives through to finished buttons, cards and badges.

The token reference on top, and the component library it feeds below it, from colour primitives through to finished buttons, cards and badges.

From a colour in Figma to a line of CSS

Two components, followed from Figma to the shipped CSS. The Needs you badge takes its colour from a status and Doch's bubble from a role. In both, the value set in the design system is the token the CSS renders from.

Figma · Design System
Needs you
StatusPill · Tone = need
Fill#FCEBC4
Dot#E0A21F
Text#5E4206
Radius999

The Needs you badge. Its colours live in the design system, not on the component.

doch.css · tokens
/* Status: needs you */
--need: #E0A21F;
--need-text: #5E4206;
--need-soft: #FCEBC4;

Each colour is one named token, the only place the raw hex is written.

doch.css · components
.tone-need {
background: var(--need-soft);
color: var(--need-text);
--dot: var(--need);
}

The badge picks its tone from its label, so a new status can't end up the wrong colour.

Looking back

Key takeaways

The tech works. Trust is the unsolved part

Replies came in under a second (median 0.79 s over 31 replies), the hold worked and the plan held. None of that matters if the person hangs up in the first six seconds, and 14 of the 26 people who picked up did, at the AI line.

Disclosure is the front door

Every trust moment I designed sits behind one sentence the law requires. That sentence is now the design problem, not the screens after it. I changed the opening over two rounds of calls (23 and 30 October) and people still left 2 to 6 seconds in.

A safeguard can become a gate

The check question was there to protect the call. It caught one wrong number, and 3 of the 7 calls that opened with it hung up at the question itself. Every step added to the opening has to earn its seconds.

A report has to be honest about failure

“Call finished” over a call where nothing was agreed is worse than no report. The audit trail is only useful if it says when things went wrong. 6 of the 20 simulated users asked for the same thing: tell me what really happened. Doch once promised a business that I would get back to them, and nothing backed it up.

Build log

Follow the build log

I wrote each week up as I built it. Four posts, from the rules and the first prototype to the phone line, the design system and what 35 real calls taught me.