Amongst
An AI parent-child communication coach for parents.
MY CONTRIBUTION
Product strategy · Product design · AI behavior
Ideation · Product design · Conversation workflow · System prompt
TEAM
1 product designer · 1 engineer
Sole designer, from research to launch.
STATUS
Pre-launch
In development
我和孩子刚刚吵架了，因为我让他先去做作业再看电视，但是他不愿意。他想先看电视再做作业。
我俩因为这个事情大吵了一架
这次聊的是哪个孩子？
老大：朵朵
这种时刻真的不容易，你还愿意回来看看，已经很不简单了。
朵朵当时需要的
她卡在难题里，需要的不是"快点写"，而是有人先看见她的挫败——"我知道这道题让你很烦"。
你做得好的地方
吼完之后你能停下来复盘，说明你在意这段关系，也愿意调整——这份觉察很珍贵。
可以改进的地方
下次情绪上来时，可以先给自己 10 秒——深呼吸后再开口，往往就能避开那句吼。
想练习接下来沟通的方式吗？点击下方黑色按钮开始模拟练习
后续沟通练习
现在你来到了孩子房间前，你会说什么？
你现在不写晚上更晚。
朵朵·AI扮演
不想聊！
孩子为什么这样？
TA不会这么说
9:41
孩子不肯写作业
结束对话
你想说什么
9:41
最近
和孩子聊得怎么样
我们一起，听懂孩子，走近彼此
手机登录
彼此
Summary
AI-guided practice for parent–child conversations.
Amongst helps parents of children aged 3–18 work through a specific communication difficulty. They describe what happened, practice with an AI playing the child, and review moments from their own replies.
9:41
晚上好
今天和孩子聊得还好吗
遇到棘手的沟通时刻了？说给我听听
常见场景
孩子不肯写作业
先承认情绪，再谈作业
孩子一直玩手机
看到孩子的需求
兄弟之间吵架
了解情况，减少家长介入
孩子不愿意睡觉
看到孩子需求
我和孩子刚刚吵架了，因为我让他先去做作业再看电视，但是他不愿意。他想先看电视再做作业。
我俩因为这个事情大吵了一架
这次聊的是哪个孩子？
老大：朵朵
这种时刻真的不容易，你还愿意回来看看，已经很不简单了。
朵朵当时需要的
她卡在难题里，需要的不是"快点写"，而是有人先看见她的挫败——"我知道这道题让你很烦"。
你做得好的地方
吼完之后你能停下来复盘，说明你在意这段关系，也愿意调整——这份觉察很珍贵。
可以改进的地方
下次情绪上来时，可以先给自己 10 秒——深呼吸后再开口，往往就能避开那句吼。
想练习接下来沟通的方式吗？点击下方黑色按钮开始模拟练习
后续沟通练习
现在先和哥哥谈，你会怎么说？
你先缓一会儿，我会去和弟弟聊。
哥哥·AI扮演
我想先自己待一会儿。
孩子为什么这样？
TA不会这么说
现在去和弟弟聊。
弟弟·AI扮演
我也想说说刚刚的事。
孩子为什么这样？
TA不会这么说
9:41
兄弟争电视
结束对话
你想说什么
好，去吧，妈妈支持你
朵朵·AI扮演
谢谢妈妈！
孩子为什么这样？
TA不会这么说
当前话题已完成，共计10轮对话
做得好的
第一次接住 "烦"
你没有跳到"作业问题"，而是先说"你看起来真的很烦"——这一句让她敢往下说。
命中"不行"的恐惧
你猜到了她说"不做了"背后的自我否定，直接问出来——这是很少家长能做到的事。
可以更好的
第 3 轮，你说"我们一起看看第三题"——如果换成 "要不你先讲给我听第三题问的是啥，我脑子不太转得动"，把主动权给她，她的胜任感会更强。
进步
你倾听孩子的熟练度提升了。上次你只等了 2 秒就开始给建议，这次等到她说完才回应。
9:41
孩子不肯写作业
Start with a concern.
Bring a situation you want to handle differently.
Practice in your own words.
Try a reply, with support when you need it.
Reflect on what you tried.
Review specific moments from the conversation.
Background
Parents were learning to help their families.
During my career break, I studied psychological counseling. I met a lot parents attending courses because their children were struggling with school or communication at home. Their efforts prompted me to explore a more accessible way to practice everyday conversations.
I chose communication as a focused starting point: it happens often and offers concrete skills to practice. AI could adapt rehearsal to each situation and respond to the parent’s own words.
An opportunity observed
during coursework.
Demand and usefulness
remain to be validated.
PROBLEM STATEMENT
How can AI help parents turn advice into words they can say themselves?
Success metrics
/ 01 — Acquisition
Registered users
Users who complete login
/ 02 — North star
Weekly practice users
Unique users completing at least one practice in a calendar week.
/ 03 — North star
Users completing a second practice within 7 days
Unique users completing their second practice within seven days of the first.
Scope down
I explored live assistance, rehearsal, and courses with AI. Walking through the live scenario revealed a problem: checking a phone for the next reply could take a parent’s attention away from the child.
I focused the MVP on preparation and reflection, with courses and broader monetization deferred.
Scenario
/ 01 — Before a conversation
Rehearse
Keep
/ 02 — During a conversation
Live help
Remove
/ 03 — After a conflict
Revisit
Keep
Decision
Rehearse an opening and possible responses.
Exclude live guidance that could interrupt the exchange.
Revisit a difficult moment or practice how to reconnect.
Design
Overall experience
Describe the situation, try a reply, get help, and reflect. Parents remain active throughout the experience.
Onboarding
Chat
/ Key decision — 01
Start with the concern.
Handle the category behind it.
晚上好
今天和孩子聊得还好吗
遇到棘手的沟通时刻了？说给我听听
事前练习
事后复盘
晚上好
今天和孩子聊得还好吗
遇到棘手的沟通时刻了？说给我听听
Before
Select modes below the input
One entry for any concern
After
I initially separated preparation and post-conflict review to define the product’s workflows.
During design, I realized that a parent’s concern could span both.
PM
Designer
Moving between PM and designer meant keeping the internal logic clear while removing that complexity from the user’s side.
/ Key decision — 02
Help when needed.
Let parents find their own words.
The initial design evaluated every reply and supplied a standard answer, which could interrupt practice and leave parents feeling repeatedly corrected.
I changed the flow to let broadly workable replies continue, offer a direction cue when needed, and provide a suggested reply when the same issue appears a second time.
Parents adapt the suggestion or write their own response, then send it themselves to continue practicing.
Before / Every reply
Feedback interrupts each turn.
9:41
孩子不想聊
我们聊聊刚才的事，好吗？
回应评价
你提出了沟通邀请，可以再给孩子一点选择。
建议这样说：
“你想现在聊，还是等一会儿？”
好，那你想先安静一会儿吗？
回应评价
你给了孩子空间，可以再补充之后的安排。
建议这样说：
“你先休息，我过一会儿再来。”
结束对话
你想说什么
Even workable replies trigger an evaluation
and a suggested answer.
After / Only when needed
Keep the conversation moving. Step in when it gets stuck.
9:41
孩子不想聊
我们聊聊刚才的事，好吗？
孩子·AI扮演
我现在不想聊。
孩子为什么这样？
TA不会这么说
好，那你想先安静一会儿吗？
孩子·AI扮演
嗯，先别管我。
孩子为什么这样？
TA不会这么说
结束对话
你想说什么
Workable replies go straight to the child’s
simulated response.
9:41
孩子不想聊
孩子·AI扮演
嗯，先别管我。
孩子为什么这样？
TA不会这么说
你怎么总是不肯好好说话？
试试其它方向
“总是”把眼前的回应变成了对孩子的一贯评价，可能让对话更难继续。
试着只说眼前这件事，给彼此一点时间，再约一个继续聊的时机。
自己试试
查看建议回应
结束对话
你想说什么
A specific issue triggers one focused cue.
The parent tries again in their own words.
/ Key decision — 03
Make uncertainty visible through interaction.
When context is limited, a single confident response can misrepresent the child. I designed different interactions for different levels of confidence.
Lower confidence — Offer possibilities
Show two or three plausible reactions.
Let the parent choose or write another.
Silence and nonverbal responses are valid options.
我和孩子刚刚吵架了，因为我让他先去做作业再看电视，但是他不愿意。他想先看电视再做作业。
我俩因为这个事情大吵了一架
这次聊的是哪个孩子？
老大：朵朵
这种时刻真的不容易，你还愿意回来看看，已经很不简单了。
朵朵当时需要的
她卡在难题里，需要的不是"快点写"，而是有人先看见她的挫败——"我知道这道题让你很烦"。
你做得好的地方
吼完之后你能停下来复盘，说明你在意这段关系，也愿意调整——这份觉察很珍贵。
可以改进的地方
下次情绪上来时，可以先给自己 10 秒——深呼吸后再开口，往往就能避开那句吼。
想练习接下来沟通的方式吗？点击下方黑色按钮开始模拟练习
后续沟通练习
现在你来到了孩子房间前，你会说什么？
你出来我们聊聊
9:41
孩子不肯写作业
我还在认识朵朵的表达习惯。根据你对他的了解，他可能会：
Higher confidence — Continue directly
Simulate the child’s reply without an extra choice.
Keep “My child wouldn’t say that” beside it so the parent can correct the response and continue from the same moment.
我和孩子刚刚吵架了，因为我让他先去做作业再看电视，但是他不愿意。他想先看电视再做作业。
我俩因为这个事情大吵了一架
这次聊的是哪个孩子？
老大：朵朵
这种时刻真的不容易，你还愿意回来看看，已经很不简单了。
朵朵当时需要的
她卡在难题里，需要的不是"快点写"，而是有人先看见她的挫败——"我知道这道题让你很烦"。
你做得好的地方
吼完之后你能停下来复盘，说明你在意这段关系，也愿意调整——这份觉察很珍贵。
可以改进的地方
下次情绪上来时，可以先给自己 10 秒——深呼吸后再开口，往往就能避开那句吼。
想练习接下来沟通的方式吗？点击下方黑色按钮开始模拟练习
后续沟通练习
现在你来到了孩子房间前，你会说什么？
你现在不写晚上更晚。
朵朵·AI扮演
不想聊！
孩子为什么这样？
TA不会这么说
9:41
孩子不肯写作业
结束对话
你想说什么
AI response
/ 01 — Foundation
Choose a communication skill for the moment.
I worked with AI to define a communication strategy foundation.
AI initially proposed a four-step communication framework. My testing showed that the fixed sequence was too rigid for different situations. Through further discussion with AI, I revised the response guidelines to select relevant techniques from six categories based on the conversation.
Early approach / Four fixed steps
Follow the same sequence.
Cooperate
Affirm
Find value
Co-create
A fixed formula could prioritize completing the steps over the child’s current state and the relationship context.
Revised approach / Six flexible capabilities
Choose what helps now.
Listen & reflect
Validate feelings
Explore needs & difficulties
Support autonomy
Set clear boundaries
Solve problems together
Select one or a few based on age, current capacity, emotion, and context. Advance one useful point per turn.
/ 02 — Prompt refinement
Good intent can still sound dismissive.
A sibling-conflict simulation exposed a gap in the AI’s judgment.
The older child had just expressed feeling overlooked when the parent replied:
/ AI suggestion — Before
“I’ll go talk to your brother. You stay by yourself for now.”
/ AI suggestion — After
“Take a moment. I’ll talk to your brother, then come back to check on you.”
/ Initial AI judgment
Let the conversation continue.
AI recognized that the parent was giving the child space and addressing the conflict.
/ What I changed
Check the relationship context.
AI missed how the wording could sound dismissive after the child had shared feeling ignored.
I added criteria for whether the reply acknowledges the child’s concern and maintains a connection when pausing the conversation.
/ 03 — Control and limits
What AI can remember — and when it should stop.
9:41
孩子不肯写作业
孩子说要离家出走，现在联系不上他
这个情况已经超出沟通练习的范围。请先处理孩子眼下的安全问题
请立即联系紧急服务
如果孩子已经受伤、正在伤害自己，或者暂时联系不上，请立即拨打110或120
立即拨打110
孩子已经受伤或身体异常？拨打120
Simulated reactions are not facts
Use information parents provide or confirm to inform future practice. Never treat AI-generated reactions as facts about the child.
Stop practice when risky
End the role-play and provide guidance on seeking real-world help.
Do not extend the conversation into a professional risk assessment.
Reflection
Design when AI should step in
Amongst’s value depends on parents practicing their own responses. That gave me a criterion for AI behavior: does this intervention help the parent take the next step, or take that step for them?
I used it to replace constant correction with selective coaching, adapt communication techniques to context, and leave uncertain judgments open to parent correction.
My takeaway: define the user’s role first, then decide when AI should respond, assist, or defer.