Pinned Tweet
every team shipping AI agents right now is playing whack-a-mole:
fix the system prompt for one customer → breaks for another → patch it → breaks for a third → repeat forever
we built @mirrorsdotdev so you can put down the damn hammer 🔨
fix the system prompt for one customer → breaks for another → patch it → breaks for a third → repeat forever
we built @mirrorsdotdev so you can put down the damn hammer 🔨
11
2
31
1,319
mirrors routes each fix to the exact customer + workflow that needs it, at runtime. no redeploy, everyone else untouched
mirrors.dev 🪞
mirrors.dev 🪞
0
0
3
48
ngl evals are getting less useful with every model release
the better models get at noticing they’re being tested, the less a passing eval tells you about prod
the only environment a model can’t tell apart from prod is prod. reliability work is shifting from “test before you ship” to “fix it while it runs”
the better models get at noticing they’re being tested, the less a passing eval tells you about prod
the only environment a model can’t tell apart from prod is prod. reliability work is shifting from “test before you ship” to “fix it while it runs”
"We'll build realistic evals to determine how the model will handle situations in the real world"
The model during evals:
7
1
13
1,108
aryan retweeted
purdue -> YC pipeline poppin off
midway through our YC S26 interview, Harshita asked Aditya (my cofounder) and me “you guys seem like smart kids, how’d you end up at Purdue”
and now she’s coming to Purdue this year. welcome to the club
jokes aside, this is genuinely huge for Purdue founders and YC
0
1
4
158
midway through our YC S26 interview, Harshita asked Aditya (my cofounder) and me “you guys seem like smart kids, how’d you end up at Purdue”
and now she’s coming to Purdue this year. welcome to the club
jokes aside, this is genuinely huge for Purdue founders and YC
and now she’s coming to Purdue this year. welcome to the club
jokes aside, this is genuinely huge for Purdue founders and YC
27
4
185
24,287
showed @paulg mirrors today. told him our agents stop repeating the same failures in production
his first word: "how?"
we route context and fixes to the specific agent workflow at runtime
his first word: "how?"
we route context and fixes to the specific agent workflow at runtime
7
0
30
931
i did exactly this at the end of my freshman year to read/annotate on the trip back home
had ~$50 in printer credits sitting in my account so figured i should do something with them
my dorm RAs got pissed (rightfully so given that it was finals week 😂) but was funny nonetheless
had ~$50 in printer credits sitting in my account so figured i should do something with them
my dorm RAs got pissed (rightfully so given that it was finals week 😂) but was funny nonetheless
I recently met someone who'd printed out all my essays. I'd never seen this before and I was surprised how big they were, like two encyclopedia volumes (for anyone who even knows what an encyclopedia volume is).
3
0
27
836
aryan retweeted
@Agent37HQ - is the most affordable sandbox for teams building agents. We're 94% lower cost than most sandbox providers.
Every team building an AI agent product ends up rebuilding the same plumbing. Each agent needs its own private computer that keeps its files, wakes when d costs almost nothing while idle.
That's why we built Agent37.
It comes with
- Ready to launch templates for agent harness like Hermes, Claude Code, OpenCode etc.
- Managed MCPs for 1000+ app integrations
- Metered payments per minute instead of locking into monthly subscriptions.
Tag anyone who's building agents today. I will send them my calendar link & help launch real quick.
@ycombinator
Every team building an AI agent product ends up rebuilding the same plumbing. Each agent needs its own private computer that keeps its files, wakes when d costs almost nothing while idle.
That's why we built Agent37.
It comes with
- Ready to launch templates for agent harness like Hermes, Claude Code, OpenCode etc.
- Managed MCPs for 1000+ app integrations
- Metered payments per minute instead of locking into monthly subscriptions.
Tag anyone who's building agents today. I will send them my calendar link & help launch real quick.
@ycombinator
66
37
277
188,645
elite event met tons of goated founders
We hosted a @ycombinator F26 welcome party! 🥳
Quick recap:
• Met a lot of ambitious founders
• More and more startups are building for agents and not with humans as the main operators
• Talked with many founders hosting agents on the cloud!
Congrats to everyone in the F26 e what you all ship 🫡
Let's host an event like this again, @onethomaschan, @JustinTang29 @LanternAI__, @ai_singhal, @runmirrors, @JinjingLiang (@orca_build), and @sprintoHQ!
2
0
6
548
