Ujjwal Pandey

💻 Building, coding & solving problems.
🤖 Exploring AI and how it changes the way we build.
🚀 Sharing what I learn about tech and what’s next.

[Ujjwal Pandey @iamupandey]

🇮🇳 I have an idea that could change India’s education system at microscopic level :-

@RahulGandhi @EduMinOfIndia @PMOIndia - why not create a "National Student Card" for every student from Class 6 to higher education ?

One verified identity →
🎓 education
💰 scholarships & loans
🚀 startup funding
🤖 AI & tech access
🧪 innovation labs
💼 internships
🏭 manufacturing opportunities
🌍 export support

A student shouldn’t just graduate with a degree.

They should graduate with skills, experience, a portfolio, and the ability to build something.

We should connect education directly to jobs + entrepreneurship + manufacturing + exports.

Learn → Build → Manufacture → Export → Grow India 🇮🇳

If we want higher GDP and higher per-capita income, investing in students is one of the smartest places to start.

Looking forward to do research on that.
#StudentCard #Education #India2047 #SkillIndia #MakeInIndia

[Ujjwal Pandey @iamupandey]

@Archita_shri08 Reply dena hota hai 😂

[Ujjwal Pandey @iamupandey]

@GregoryNimensky I am not able to see myself ???

[Ujjwal Pandey @iamupandey]

@WebKingService Yes the new version is great ☺️

[Ujjwal Pandey @iamupandey]

Which is the best models when it comes to coding ⁉️

• GPT-6 Sol
• Opus 5.5
• Grok 4.7

Give the best answer ‼️

[Ujjwal Pandey @iamupandey]

@seridarivus13 Yeah ☺️

[Ujjwal Pandey @iamupandey]

@seridarivus13 Yes but my favorite one is still codex 😂

[Ujjwal Pandey @iamupandey]

@seridarivus13 well buddy its a great reason, i personally use codex, then my priority is grok then claude

[Ujjwal Pandey @iamupandey]

@iamujjwalpandey Yes absolutely 💯

[Ujjwal Pandey @iamupandey]

🚨 OpenAI agents posted 53 user- provided images to external websites.

Here are the takeways we should keep in our mind 👇

[Ujjwal Pandey @iamujjwalpandey]

🚨@OpenAI found research agents posting 53 user-provided images to external websites.

Why this matters for builders:
• Agents can take real actions, not just generate text.
• Tools and internet access creates new ways for data to leave.

Overall, observing logs is important !

[Ujjwal Pandey @iamupandey]

@seridarivus13 Let's say we have to choose 1 model overall, what will be you call buddy ???

[Ujjwal Pandey @iamupandey]

🚨 Attention claude users !!!

Claude code quietly added something every team running AI agents in production will eventually need !!!

v2.1.283 actually adds:
• Exact model matching.
• denied Models to block models.
• Tighter team control.

So, your agents use the models you expect and your tel decides which ones are off-limits.

[Ujjwal Pandey @iamupandey]

Here’s the official announcement 👇
https://t.co/dMDiequg50

[Ujjwal Pandey @iamupandey]

🚨 Anthropic just dropped Claude's plugin directly to builders.

Now we can:
• Submit an MCP server to GitHub plugin.
• Get validation and safety checks.
• Track installs, views and search terms .
• Reach users across claude itself.

Anthropic says that MCP usage is 110x times greater this year.

So, MCP is becoming a distribution layer.

[Ujjwal Pandey @iamupandey]

@SKatalystAI Making loops and graphs for content and coding will be better

[Ujjwal Pandey @iamupandey]

🚨An OpenAI agent searching for Australian medical-spending data hit an access boundary and kept going on.

Meanwhile, no patient records were breached.

Well, the bigger question is that when should an AI agent must know where it has to stop ?

Read👇 https://t.co/dMulbNQweG

[Ujjwal Pandey @iamupandey]

@ash_twtz Read this buddy 😊👇🔥

[Ujjwal Pandey @iamupandey]

@OpenAI @OpenAIDevs The numbers are straight from OpenAI:
Check the official sources 👇🔥:
https://t.co/oQhnmzkI3h
https://t.co/cTNKz2jxDP

[Ujjwal Pandey @iamupandey]

@SKatalystAI @OpenAI @OpenAIDevs Repost this one if it's good 😊

[Ujjwal Pandey @iamupandey]

🚨 I think a lot of people are about to waste their money on GPT-6 for one simple reason:

I mean "using the smartest model" for everything ⁉️

Well You know,
• Luna - 💲0.10/M output
• Sol - 💲2/M output
• Astra - 💲10/M output

Well that's 100% Gap ‼️

But Wait 🚨👇

I am not saying "use the cheapest model" is the main lesson I am trying to give 🤔🙂‍↔️

Well the real metric isn't cost per token❗️

It is cost per successful task performed ‼️

So switch the models carefully according to the tasks to save tokens and your time. 🔥

[Ujjwal Pandey @iamupandey]

@SKatalystAI Exactly. Being cheaper per token doesn’t mean cheaper overall.

If it takes more retries and cleanup to get the job done, those savings disappear fast.

[Ujjwal Pandey @iamupandey]

@i_mika_el Exactly buddy. Routing should measure cost per successful task: success rate, retries, tokens used, and latency.
A cheaper model that needs 3 retries can easily lose its price advantage.

[Ujjwal Pandey @iamupandey]

@SKatalystAI @OpenAI @OpenAIDevs Also keep supporting 😊