AI Newsletter - 7/14
Sharing what my curiosity mind reads/thinks about the technology and the world each week. Inspired by Josh Woodward.
Thoughts:
In my work, I’ve engaged extensively in prompt engineering and explored several vibe coding projects. While coding has become the easiest part, the real challenge lies in effectively communicating my requirements to the AI model—much like how we communicate with humans. This led me to realize that, even in the age of AI, communication remains the most critical skill. It’s about articulating our thoughts clearly and effectively.
Recently, I watched a YouTube video titled The New Code — Sean Grove, OpenAI, where Sean eloquently explained why structured communication is the new “code.” He mentioned, “Code only accounts for 20% of the impact, while the remaining 80% comes from structured communication.” By structured communication, he refers to a process involving six key steps:
Engaging with users to understand their challenges.
Distilling these discussions into concrete goals to address those challenges.
Strategizing ways to achieve those goals.
Sharing these plans with colleagues.
Translating the plans into code.
Testing and verifying that the outcomes align with the plans and original objectives.
While vibe coding allows us to generate code almost instantly, the real effort now lies in how effectively we create structured communication with the AI model. Sean also remarked, “Structured communication is the new bottleneck,” and I’ve experienced this firsthand. My engineer constantly seeks clarification on requirements, underscoring the importance of clear and structured communication.
What I read/watch:
#1/ Danger of AI in mental Care - A long way to go to use AI as the therapist, but good use case on training therapist or agent to operation for therapist.
#2/ The New Code — Sean Grove, OpenAI - The structured communication is the new code and it is the bottleneck of the vibe coding.
#3/ The Deep Story On The Waymo Vs Tesla Robotaxi Battle - Tesla is behind Google for 8 years from safety standard. Google used Ladar and map to create its unique advantage and hardware will get cheaper.
#4/ China’s Moonshot AI launched K2 open source agentic model - the cost of agent will be so cheap!
#5 Meta's AI Strategy - acquiring the talent to fill the gap
#6/ How Walmart built an AI platform that makes it beholden to no one - Impressive business case on build-versus-buy dilemma as most company will opt for buying. Massive scale with 3 million daily queries from 900,000 weekly users.
#7/ **Are we designing AI products all wrong?** - Chat interface is just a lazy design because it still require too much user effort to prompt.
#8/ China is building an entire empire on data - “China’s planners see data as a factor of production, alongside labour, capital and land.” When you have the largest set of the data in the world, what would you do? It simply enables far more intelligence for the country.
Twitter:
Things that I tried:
Voice is the area I see huge promising because it is the more natural way of human interaction and tried a few product this week:
play.ai - amazing real person voice experience!
synthflow.ai - tried the real estate agent call
Data:
251 Million GPUs shipped in 2024; Nvidia crossed $1tr in June 2023 and $4tr in July 2025 - 4 times in 2 years!