Fadi Zuabi

ML Engineer @Solidigm 🤖 Bay Area 📍Dad 👶🏼👧🏻

[Fadi Zuabi @fadi_zuabi]

next time you're struggling to setup an AI meta layer tool, just copy paste this to any AI chat you use:

"Setup this [Skill/ MCP/ Agent/ etc] for me <link to the tool or its docs>. If you need any auth or help from me, give me steps to do it so you can continue "

[khariya @khariya65912325]

@fadi_zuabi Another cool trick is to use + to create unique email like [email protected]

[Fadi Zuabi @fadi_zuabi]

@sattyyouneed make money more than it cost. not productivity, real ROI

[Fadi Zuabi @fadi_zuabi]

well put, @OrenMe and @vasuman

Fun fact: most of us were afraid AI will take our jobs, and now we're disappointed that it didn't.

AI will generate more work and will need more workforce. the outcome is human advancement, not replacement. History proved this too many times.

Transition is hard and we're in that mud now.

[Oren Melamed @OrenMe]

We all thought AI will take our jobs, but it was eventually @vasuman with this killer post

Adoption was a lie, you don’t need to code more/faster!

When code is cheap and you can ship 100 features, knowing which 10 to ship matters more

Judgment is the key…

The funny thing I come to realize now is that the laws of physics still remain - your end users don’t want you to ship 1 trillion features a month just because u now can, they need the one they need - and we now need to go upstream and start figuring out how to create value with AI on business and product level and not just last mile coding and architecture

The industry is around it - we call it second brain, company brain, LLM wiki etc but the pattern is emerging - code is not enough to run the business and we need to see how we bring more people to the table

BTW - if we talk physics - macro vs micro - small company vs big company is not the same - if u run small shop you can probably feel the momentum faster and maybe this won’t echo as strong as it does if you have history, legacy and hundreds to thousands of developers already iterating on your business code

But the bottom line is again - judgment - on what to do when you can, hypothetically, do anything

[Fadi Zuabi @fadi_zuabi]

@burkeholland one supreme crunchy is not enough! they even made it smaller. agree though two was a bad idea.

[Fadi Zuabi @fadi_zuabi]

double down. Been observing very similar results

[Mario Zechner @badlogicgames]

recommended reading. similar observations on my end, albeit at much smaller scale.

[Fadi Zuabi @fadi_zuabi]

you can also add "+" and a note to any email not just gmail.

[email protected] is same as [email protected]

[Fadi Zuabi @fadi_zuabi]

why don't many people know that you can add dots "." anywhere in your username in gmail?

[email protected] is same email as https://t.co/[email protected]

I find it useful to give different versions of my email for different services.

[gabriel @g_br_l]

when 1 (one) real user actually signs into your fully vibecoded saas

[Fadi Zuabi @fadi_zuabi]

@ForwardEditor this is the best way to share AI products imo. share the prompt not the response!

[Fadi Zuabi @fadi_zuabi]

credit @veritasium

[Fadi Zuabi @fadi_zuabi]

you would think this is our "nowadays"

[Fadi Zuabi @fadi_zuabi]

@OpenAIDevs @DanWahlin @awsdevelopers @cursor_ai @github @code @vercel dragging our ass into the childish forward slash backwards slash mess again. bravo

[Fadi Zuabi @fadi_zuabi]

Most use cases will not need advanced AI patterns. most work you can do or learn with any tool that makes AI calls. Almost all harnesses come batteries included nowadays.

1/ ✅✅ AI knows how to do most stuff nowadays. and if you know what to expect and how to validate the work, just do it!

2/ ✅❌ in some cases, e.g. internal tools and processes, AI might not be trained on it and will try to relate to what it's trained on. This is where you would watch it make the mistakes, collect the knowledge and gotchas, and bundle to a skill or some file system of your own.

3/ ❌✅ If you're exploring new areas, AI probably knows it, but you don't know how to validate the work, that's a risk of AI deception. Learn the topic by asking more and explicitly tell the model that the topic is new to you and it needs to guide you.

4/ ❌❌ This is where most AI builders are 😅. The topic is new to you and models haven't been trained on it. This is where advanced patterns like parallel and multi layered patterns become useful.

[Fadi Zuabi @fadi_zuabi]

@burkeholland I talk to it like I would in a team's group chat.
we should be nice because eventually will work with it in group chats, not because it'll become a robot and get back at us.

[Burke Holland @burkeholland]

You never really think about it, but chat is kind of a bad interface.

We only use it because nobody knows what you are going to want to do with the agent.

But you do. So build the agent interaction you want.

[Fadi Zuabi @fadi_zuabi]

this is exactly what I teach in workshops now. it takes 4 hours to get this idea through to 15 engineers, but totally worth it. I ask folks to start an automation and let them ask it to "do the x task and create a dashboard for me to leave feedback and communicate with you" - it evolves with each run.

session bridge is another killer feature everyone is loving in copilot, when they see two chats talking to each other is like a dream coming true.

oh and an aha moment is when I show the chat debug view and how/why sending "hi" is also sending 20k in tokens.

thank you! I learn a lot from your videos.

[Fadi Zuabi @fadi_zuabi]

@XMoney spot the bug ;)