Software Developer. I build with AI daily and share how to earn more and stay employable. Posts on AI, careers & money π°. Cricket when India plays π
OpenAI estimates agents finish 46% of delegated 2β4 hour research tasks without intervention.
Impressiveβbut just 25% of inference/serving-code tasks. βAI can do half the workβ means little until we ask: which half, and who owns it in production? π€
We keep arguing about which AI model is βbestβ. Gemini 4 Neonβs launch has a useful reminder: Google reports 12.8% on one coding-agent setup and 39% in Antigravity. Same model, different harness. The tools around the model matter too. π
Characte
What's stopping people from making animated movies from AI and releasing in theaters ? If u u have good story then I think it's a great opportunity π€
Building with AI has made one thing obvious: generating code is getting easier. Knowing what to ask, what to reject, and what might break in production is still hard.
So where do juniors learn that judgment if nobody gives them a chance? π€
Me: I'll just ask AI to make one small change.
AI: Rewrites 4 files, changes the architecture, introduces 3 new bugs and explains why it's actually better. π
Me: I only asked you to change a button.
Every developer has been here π
At CIQ, AI reportedly writes over 90% of the code...yet the company says it canβt hire enough engineers, especially experienced ones.
If juniors canβt get the first opportunity, where do tomorrowβs senior engineers come from? π€
Hot take: Knowing 5 programming languages may matter less than knowing how to solve 1 real business problem.
AI can help write Java, Python or Go.
But understanding WHY something needs to be built? That's still on us. π€
Are developers preparing for this shift?
DeepMind's Signal work caught my eye. AI agents helped produce machine-checkable proofs for key security properties; humans still defined and reviewed what needed proving.
That's a more interesting use of AI than generating another pile of code. π
The AI agent hype is loud, but the less glamorous part is getting interesting.
LangChain just raised $300M to scale tools for monitoring and evaluating agents.
Shipping an agent is one thing. Trusting it in production is another. π
I tried to login to my personal Microsoft account, this is what I got after login page π€
UTC Date: 2026-11-13T13:12:08
Client Id: F69E708D
Session Id: 039b04-949
Client Version: 20261030.02
BootResult: network
Back Filled Error: None
err: TpError: Failed to fetch
esrc..see more