Saturday, December 27, 2025

Investing in AI: good info & book

very astute opinion, show experience!

Bill Gurley (@bgurley) is a general partner at Benchmark, a leading venture capital firm in Silicon Valley. His new book

Life is a use-it-or-lose-it proposition.
Shouldn’t you spend it doing something you love?
This book will teach you how to find your dream job and avoid a career you’ll regret—from a leading venture capitalist, based on his viral college talk.
This title will be released on February 24, 2026.




AI summary:
  • AI as a "Real Technology Wave" with Speculation (0:23-1:39): Bill Gurley discusses how every significant technology wave, like AI, inherently attracts speculation and "bubble-like behavior" due to rapid wealth creation. He emphasizes that a real technological shift and speculation often go hand-in-hand.

  • "Circular Deals" in AI Investments (3:06-3:52): Gurley explains "circular deals" where large companies like Microsoft invest in AI startups (e.g., OpenAI), and those startups then agree to purchase services from the investor. He views this as a questionable and non-transparent accounting practice.

  • Concerns for Retail Investors in AI SPVs (4:20-5:28): Gurley expresses concern for retail investors getting involved in AI-related Special Purpose Vehicles (SPVs). He warns that many promoters of these SPVs may not even have the underlying stock, calling it the "wild wild west" and advising extreme caution.

  • Challenges of Private Company Investing (7:10-8:06): The video highlights two main problems with the public investing in private companies: the majority of VC-backed private companies go to zero, and there's a lack of information transparency compared to public markets.

  • Angel Investing Strategy in AI (9:05-9:30): For angel investing in the current AI landscape, Gurley suggests looking for individuals who are highly curious, actively using AI tools, and bring a unique perspective from a specific industry that gives them an advantage.

  • Institutional Investor Focus on AI-Only Deals (9:50-10:41): Currently, institutional investors have "zero interest" in non-AI deals. This means that if an angel-funded deal isn't AI-related, it risks "dying of neglect" as it won't attract future institutional funding.

  • Protecting Your Career with AI (10:43-11:08): Gurley advises everyone, regardless of their field, to start "playing with this stuff" (AI). He believes that being the "most AI-enabled version of yourself" is the best way to protect your career from potential displacement by AI.

  • Investing in "Off the Beaten Path" AI Verticals (12:28-13:05): When investing as an angel, Gurley recommends focusing on "deeper verticals" or niche industries that are not high-priority targets for large AI companies like OpenAI (e.g., waste management). These verticals often involve specific workflows and proprietary datasets that big models won't easily "crush."

  • Importance of Proprietary Data Sets and Workflows (13:16-14:06): Successful AI investments in specific verticals will likely involve proprietary data sets and the ability to build software around existing "workflows"—tasks that need to be automated and integrated with AI, like booking tours in real estate.



wisdom for career: "don't half-ass it" => rocket fuel for life!














AI dev tool: Google Antigravity

Is this a web "dev help", or "dev replacement" tool?

While based on VS Code, it also includes
full control of Chrome web browser, and "agent manager",
to initiate and manage and test multiple projects and playgrounds in parallel.

The code editor is available, but more like a "fallback" access to check things.
In fact, Antigravity does not "like" sharing control, it does its things its own way.
Very opinionated.

So is it "good", "amazing", or "not for me"?

It is different enough, and provides free access to Google Gemini 3 Pro (unlimited?)

Worth a try for sure.

My take is that this is an evolutionary step in abstraction levels
similar to upgrade from assembly/macro (ASM) programming languages
to now common "higher level" languages like C, C++, Python, JavaScript, C# etc.

Since Google is a bit "late to the game", and has a "culture" of attempts for 10x improvements,
it is not a surprise they are "pushing the limits." As they did with Angular 2. And React "won". 

I also think this is a move to a wrong direction.
Non-developers will be lost in advanced tool,
and good developers are not good "managers,"
and good managers are not good developers, no matter abstraction level.

Time will tell... and VS Code will for sure capture some of those ideas,
same as they did with from Cursor. 

Learn the basics of Google Antigravity - YouTube

Google Antigravity: From Beginner to Expert in 14 Minutes - YouTube

Google Antigravity - YouTube

Links

Agentic AI: Language Model Usage

 Stanford Webinar - Agentic AI: A Progression of Language Model Usage - YouTube.

concept of agentic language models (LMs) and their usage. common limitations of LMs and agentic LM usage patterns, such as reflection, planning, tool usage, and iterative LM usage. Overview of LMs LM Usage and limitations Retrieval Augmented Generation (RAG) Tool usage Agentic LMs Agentic design patterns Insop is a Principal Machine Learning Researcher at GitHub Next. Previously he worked at Microsoft, where he focused on leveraging machine learning and large language models to boost engineering productivity. His projects included fine-tuning open-source large language models with internal code and text, developing document assistance tools, and applying AI to various engineering tasks. He is currently a course developer as well as a course facilitator for Stanford Online’s professional AI program. 00:00 - Introduction 00:10 - Overview of the Talk 01:50 - Training Language Models 02:30 - Modeling Objectives 04:00 - Examples of Training Data Formatting 05:40 - Applications of Language Models 06:50 - Using API for Language Models 09:00 - Best Practices for Prompt Preparation 11:10 - Importance of Clear Instructions 13:40 - Reflection and Improvement Techniques 16:30 - Tool Usage and Function Calling 20:30 - Definition of Agentic Language Models 21:50 - Reasoning and Action in Agentic Models 24:00 - Example of a Customer Support AI Agent 29:20 - Summary of Applications 36:00 - Key Design Patterns in Agentic Models 44:00 - Summary of Agentic Language Model Usage 47:40 - Audience Q&A 50:00 - Addressing Ethical Considerations 54:50 - Getting Started with Language Models 57:00 - Resources for Staying Updated 58:20 - Closing Remarks