Your professor moves the problem-set deadline.
CS 170 · problem set 4
due Thu → Fri
↳ for the Brief · problem set 4 now due Friday
noticed · matched · updated
Your AI
Doppelgänger.
Knows what you’re trying to make happen. Stays caught up with your life. Designed to keep working toward your goals when you’re not there.
11:48 PM · You went offline
The research
We wanted to know if this problem was actually ours.
So I started interviewing ambitious students myself — people balancing research, recruiting, internships, startups and far more than classes — about how they actually keep it all together.
- Walk me through everything you’re trying to make happen right now outside of just keeping up with classes.
- When you sit down in the morning and need to figure out what matters today, what do you actually check or look through?
- Tell me about the last time something important almost slipped through the cracks—or actually did. What happened?
- Think about the last time you had to piece together what was going on before you could actually do the work. What were you looking for, and where did you have to look?
- What have you tried to make all of this easier—Calendar, Notion, task managers, ChatGPT, anything else—and how well has that worked for you?
- During your busiest weeks, what part of managing everything takes the most mental energy that isn’t actually the work itself?
- If you could permanently eliminate one piece of that management overhead without changing the actual work you do, what would you choose—and why?
An ambitious student’s Tuesday
A day, kept by hand
- 9:00CS lecture
- 11:00Research meeting
- 1:00Internship
- 3:30Recruiting interview
- 5:00Startup meeting
- 7:00Club responsibilities
- 11:59Project deadline
- A professor moves a deadline
- An internship contact replies
- A recruiter asks for your availability
- A collaborator sends feedback
- Tomorrow’s meeting moves
Who is keeping track of all of this?
Right now, you are.
One life. Two instances.
Kairo gives you a doppelgänger.
You
Kairo
A recruiter emails you
Person recognized
recruiting situation · openedYou reply with your availability
Waiting on the recruiter
open loop · createdYou close your laptop
The state stays where it is
context · keptasleep
The recruiter replies
open loop · waiting → resolved
Your application moves to interviewing
pursuit · applied → interviewingYou come back
The Brief is already updated
brief · caught upThe BriefYour interview is confirmed for Tuesday.
Illustrated · pursuits and open loops are integrated and tested, not yet live; a Brief fed by them and always-on cloud execution are next.
Night shift
You stop.
Kairo is designed to keep the state.
or step to a moment
You went offline
CS 170
Research lab
Recruiting
Your research mentor replies.
Research lab · onboarding
lab access · waiting → resolved
↳ for the Brief · nothing left to chase
noticed · matched · resolved
A recruiter asks for interview times.
Recruiting · product internship
applied → interviewing
↳ reply with your open times · drafted in Gmail
noticed · matched · updated · prepared
You come back.
The Brief
Problem set 4 is now due Friday.
Your mentor sent the lab access.
A recruiter wants interview times — a reply is drafted in your Gmail.
Waiting for your approval · you send it
Illustrated · one night, as Kairo is designed to keep it. Today Kairo runs while its local server is on; situations, open loops and pursuits are integrated, not yet live; always-on cloud execution is next.
You were
asleep.
Your doppelgänger wasn’t.
One night, as Kairo is designed to keep it · always-on cloud execution is next
How Kairo works
A doppelgänger needs more than memory.
It needs a model of your world.
One example, all the way down. A student applied for an internship and is waiting for a recruiter to confirm an interview. Overnight, the recruiter replies. Here is what sits under the one line Kairo shows.
Surface
Your interview is confirmed for Tuesday.
nextNothing needed from you before Tuesday — interview prep could be ready for you to review.Next
State
situationInternship recruiting · interviewing
open loopinterview confirmation · waiting → resolved
pursuitthe internship · applied → interviewing
commitmentyou · interview Tue 3:30 PM · active
Model
personthe recruiter · one person across work email and calendar invite
belief v1interview · “Tue or Wed” inference ← your first exchange
belief v2interview · “Tue 3:30 PM” fact ← the recruiter’s reply
Evidence
provenancev2 ← Gmail · recruiter → you · “Re: interview times”
historyv1 kept beside v2 · never overwritten
Source metadata + preview snippet · no message bodies stored
gmail snippetrecruiter → you · “Thanks for applying — could you share a few times…”
gmail snippetyou → recruiter · “Tuesday or Wednesday afternoon works…”
gmail snippetrecruiter → you · “Confirmed — Tuesday at 3:30 PM…”
calendarInterview · Tue 3:30 PM · invite accepted
Store richly. Reason deeply. Display minimally.
Assistants wait for instructions.
Kairo keeps the thread.
A typical on-demand assistant
Kairo
Surface
State
Model
Evidence
Source
An answer, for this request.
assembled when you ask
assembled when you ask
fetched for this request
searched just now
Your interview is confirmed for Tuesday.
open loop · waiting → resolved
the recruiter · interview time v2
← the recruiter’s reply
Gmail · recruiter → you
you ask → it searches → it reasons → it answers
life changes → Kairo notices → the model updates → you return to it
model and sources built·state layer in progress
Most AI rebuilds your context when you ask. Kairo is being built to keep it standing before you ask.
This isn’t a concept.
What exists on the main line today, what is integrated but not yet live, and what comes next — checked against the repository, not the roadmap.
AI has learned how to do work.
It still doesn’t know what work matters to you.
Models can write, research, browse, code and use tools. To work on your behalf, they also need an accurate model of you:
- your goals
- your people
- your history
- your commitments
- your unresolved situations
- the state of your world
Kairo is building that persistent context layer — for ambitious students first.
There should be
another you.
We’re building it.
Nick Zhao. 18. UC Berkeley.
Last summer I built two iOS apps while traveling through high-altitude mountains — staying online by routing a Wi-Fi router onto a tour bus. Now I’m building Kairo, and I’m a business development and partnerships intern at Berkeley SkyDeck.
Your life doesn’t stop when you close your laptop.
Neither should
your context.