Avi CT

I like bringing good products into the world. Building the next one. Previously @qbiq_ai.

[Avi CT @avichaict]

@NYCMayor Helping this post go viral so the world can see you as you are

[Avi CT @avichaict]

@keshet_ohad @BarkolAmir לא מאמין לסקרים אחרי טראומות לאומיות בסדר גודל כזה. אולי אנ�� טועה. מקווה שלא

[Avi CT @avichaict]

@ybhrdwj Bro, Meta invented network effects

[Avi CT @avichaict]

When does the bug bounty cancellation wave start? Let’s see who calls it.

[Avi CT @avichaict]

@jonatanu @gadi_eisenkot אתה בן אדם ��כ אפס.

[SHY @ShyLivnii]

https://t.co/yBsxIu3lz3 brought us one step closer to agi.

i’m serious.

i think the road to agi has two phases:

adoption → rebuilding

adoption - building the bridge between the old world and ai.

rebuilding - throwing away what we built in that phase and rebuilding the foundation around ai.

right now, most of the effort goes into accelerating existing workflows. claude code, dots, harvey, cyera, devin you name it. they are all for one mission - automating our current workflows.

the real question is: how will the next phase look like?

in most industries i think we are too far away to tell.

but the only real exception right now is the software development industry.

i believe software is in the last third of the adoption phase.

let's go through a simple software-dev workflow most engineers will recognize, and divide each step in the loop to internal (inside the model) & external.

“resolve this bug, test and deploy the fix”

intent ("solve the bug, make no mistakes") - external
↓
gather context (knowledge, logs, code) - external
↓
implement loop - both
↓
test - external
↓
push & deploy - external
↓
update context - external

now lets imagine we have an architecture (like chatjimmy) with 100x lower latency and a 1000x larger context window & ability to update its own weights (go wild)

if you rebuilt that same flow around such inference, would you still write it the way it is now?

would python still act like python?

would pytest work the same way?

would bugs live in jira, docs in confluence, logic in code, and code in github?

i think it would've ended up around something like:

intent (external)
↓
model (internal)

which would then raise the question -
if all of my state is inside the model, would this loop even require my intent?:

model (internal)
↓
model (internal)

🤔

[Avi CT @avichaict]

@pmitu Keep polishing your product, and keep worrying about upcoming marketing 😅

[Avi CT @avichaict]

@Shvartsbroit b2b.

[Avi CT @avichaict]

@solopribuilds Someone needs to make the ai autonomous, dah

[Avi CT @avichaict]

@KampnerSam @Muse המציאות היא תפיסת המציאות. באזז

[Avi CT @avichaict]

@AnthropicAI is spending $100M to train 10,000 FDEs.

Meanwhile, every AI homepage:

“Autonomous _____”

[Business Insider @BusinessInsider]

Anthropic funds $100M academy to train deployed AI engineers https://t.co/9kmxxXt9qf

[Avi CT @avichaict]

I think this may understate how many households are already paying for AI indirectly. If my bank pays for AI, that won’t show up as me paying for AI, even if some of that cost ultimately flows through to me.

Would be interesting to know how much B2B AI spend is ultimately funded by end users.

[Avi CT @avichaict]

I think this may understate how many households are already paying for AI indirectly. If my bank pays for AI, that won’t show up as me paying for AI, even if some of that cost ultimately flows through to me.

Would be interesting to know how much B2B AI spend is ultimately funded by end users.

[a16z @a16z]

98% of US households aren't paying for AI yet

More charts in State of Markets II: https://t.co/MTaxKUxa2w

[Avi CT @avichaict]

@zivaolam מדוייק. הפרואקטיביות גם ככ רדודה, ומציקה

[Avi CT @avichaict]

@klashsarma https://t.co/jeNcat4aWl

[Avi CT @avichaict]

@assaf_elovic He didn’t just trash the company - he trashed everyone who owns it with him. where is the responsibility to other shareholders? I would never invest alongside a person like that

[Avi CT @avichaict]

5/
אז בנינו את ה-Incident Challenge.
מעל 400 משתתפים מרחבי העולם בינתיים, מחברות רציניות.
כל פעם מערכת אחרת, incident אמיתי, והרבה מידע שצריך לחבר.

אפשר להשתמש ב-AI כמה שרוצים.

הקטע הוא שהוא לא אמור להיות מסוגל לפתור את זה לבד.

https://t.co/KU0N0UDkqb

[Avi CT @avichaict]

1/
שאלה שמעסיקה אותי לאחרונה:
אם AI כותב את הקוד,
במה מפתח תוכנה אמור להשתפר כדי להפוך ל��פתח טוב יותר?

[Avi CT @avichaict]

/4
להיכנס למערכת שאתה לא מכיר.
להבין מה קורה בין כמה סרוויסים.
לקרוא לוגים.
לזהות שהסימפטום הוא לא הבעיה.
ולהחליט מה בטוח לתקן.

[Avi CT @avichaict]

3/
היום אפשר לתת ל-AI הרבה מהעבודה הזאת.
וזה מעולה.

אבל יש פה משהו קצת מוזר:
ai לא מבין את המערכת שהוא פועל בתוכה. זה בדיוק מה שהמפתח מגשר עליו כיום. אז איך מחזקים את השריר הזה?