Gadi Shamia

CEO and co-founder of @Replicant_AI.

[Gadi Shamia @gadishamia]

Most contact center AI projects never leave pilot. Here's why.

Companies aren't struggling to build one bot. They're struggling to operationalize AI across 150+ intents, policies, and edge cases — without creating another silo.

After automating 1B+ minutes on the phone, we built something different: AI that builds itself from your best agents' calls.

AAA went from 23% → 75% automation success with @Replicant_AI.

Watch how we do it 👇 [video] 📢

[Gadi Shamia @gadishamia]

@Jason @uniqlo Wait until @chamath hears that you buy undergarments that are not washed with the tears of virgins and woven from lotus silk

[Gadi Shamia @gadishamia]

People on X: "Running AI locally on your phone is so cool!"

Everyone who used Claude Code on a laptop: "What I really want is a 1-hour battery life for my iPhone."

[Gadi Shamia @gadishamia]

@morad @grok which tesla models have this feature

[Gadi Shamia @gadishamia]

@gregkellyusa Security level that fits a mall entrance.

[Gadi Shamia @gadishamia]

@doritsos The same $625M exit for a company that raised half as much and has no special preferences on its cap table could have been a whole different story. It's a complex world of special snowflakes vs. one standard story.

[Gadi Shamia @gadishamia]

Claude is down. That's the breaking news.
But if you check its status page, you'd see it is down, A LOT.

AI is not a toy anymore. You should treat it as such.

https://t.co/QfdRRR0pWO

[Gadi Shamia @gadishamia]

United Airlines employees- when you get on board, do you connect automatically to the wifi?

This is the only explanation I have to why no one from United's IT fixed the FIVE minutes process of getting online.

[Gadi Shamia @gadishamia]

Tesla does it, and I can see why it is important in this moment when lane assist and self-driving are almost good enough to be tempting not to focus, but not good enough to not be supervised.

Road safety is one of the things that can't be controlled only by the free market IMO. If you decide to take a risk, you are risking everyone around you.

[Gadi Shamia @gadishamia]

@Jason offset the decline of the second home tax?

[Gadi Shamia @gadishamia]

I gave it a call (I am a customer)

What works well:
- You can interrupt the agent
- it stays on task
- Good knowladge base integration

What needs improvement:
- Data collection is weak. It had a very hard time with email and its verification code, even when I spelled it phonetically
- It can't perform tasks; it deflected me to the website to change my plan, for example.
- Conversation UX was suboptimal.

Still, I am happy to see more companies going AI-first. It is so obviously the future.

[Gadi Shamia @gadishamia]

נכון עקרונית אבל הרשה לי לתקן כמה דברים:
הטסלה היא לא אוטונומית לחלוטין. מצלמה מפקחת על העיניים ונסיון להשתמש בטלפון או לקרוא יגרור התראה וביטול המצב ה״אוטונומי״ כך שלא ניתן לעבוד או לקרוא.
נהיגה בכביש מהיר מצריכה מעט התערבות לעומת העיר כך שנהיגה באמת די עצמאית. אני נוהג קבוע חמש שש שעות על כבישים מהירים עם התערבות כל שעה בערך.
נוסעים כאן הרבה אבל אף אחד לא חושב שעתיים מהבית לעבודה זה קרוב.

הבעיה העיקרית בחישוב שלך הוא זמן. 15 שעות לעומת 5. 10 שעות עבודה שוות לרוב האנשים יותר מהפער בעלות.

[Gadi Shamia @gadishamia]

@RMB I meant that we will see RFPs for years, long after their ineffectiveness is clear.
BTW, do you think RFPs are useful?

[Gadi Shamia @gadishamia]

RFPs are dead. They just don't know it yet.

In a time when you can vibecode a new feature in hours, does it matter to spend months asking companies what they have right now?

You are measuring the past when you need to measure the future and the vendor's innovation velocity.

[Gadi Shamia @gadishamia]

@RMB Like anything in the enterprise, they will die hard (2,3…)

[Gadi Shamia @gadishamia]

Someone asked me today if UX is important in the era of LLMs. It is. A conversation is an interface, and how you run it matters. This is something @AnthropicAI is usually better at than @OpenAI.

As another example, it took ChatGPT about a year to catch up with this feature, which makes editing a prompt possible again.

[Gadi Shamia @gadishamia]

Sometimes a product earns your loyalty all over again.

My @googlenest alerted me that my AC had failed — hours before I would have noticed on my own. By the time I would have walked into a hot house, the tech was already done fixing it.

I liked Nest before; now I love it

(It was the capacitor, if anyone's curious.)

[Gadi Shamia @gadishamia]

Welcome back to "What's wrong with your customer service?" — Episode: @EufyOfficial

Someone waiting at my door. I'm miles away. No remote access to my smart lock. Eufy support? Email only. Marked URGENT.

In the next day, 1,258 words of copy-paste troubleshooting that ignored everything I said.

🔑 "I'm out of the house" — said twice. Ignored.
🔑 "Both locks affected" — said three times. Kept asking "which one?"
🔑 "Not an internet issue" — said clearly. Still got router/WiFi reset suggestions.

This wasn't AI. Humans on templates... somehow worse than a 2020 chatbot.

It resolved itself after 24 hours, exactly the system outage I flagged in my FIRST email.

In 2026, companies still skipping proper AI front-ends + empowered humans? It's a solved problem. Fix it.

20+ smart lock brands exist. Why choose the one that leaves guests (or family) stranded outside?

Share your stories — let's learn the patterns.

[Gadi Shamia @gadishamia]

My father's people philosophy at work:

↑ Kick upward
→ Respect your peers
↓ Be kind to everyone below you

I've turned this into my favorite reference check question for executive candidates:

"In one sentence, describe how [candidate] treats their peers, their reports, and their leaders."

I learn more from this question than almost any other.

[Gadi Shamia @gadishamia]

@Eric_M_Stevens Exactly. Same problem lazy product managers create :)

[Gadi Shamia @gadishamia]

The real problem with AI “vibe coding” isn’t the coding. It’s the spec.

In customer service, we see agents built from outdated knowledge bases, partial system data, and agent interviews.

That’s a bad spec. Garbage in, garbage out.

The real spec already exists: millions of successful agent calls.

📽️ how @Replicant_AI does it https://t.co/XSJIA169u6