Chicken Little

Take a moment to understand this. Once you internalise this, your investment returns will follow suit. $tsla $pltr

Markets are always volatile. That's where the alpha comes from. You buy when the price falls so far that the valuation stops making sense, then you load up the truck. But prices only get that cheap during a barrage of bad news. They never do when the headlines are good. The news looks legit, the real hit to the business is often small, and the market overreacts anyway. That's why investing feels so hard right now. You need real conviction, your own independent view, and the nerve to act when the moment comes. And when there's nothing to do, you do nothing.

Not mine but found it useful. Thanks and credit to @beamnxw

Spent 80% of today onboarding colleagues onto our Claude Teams license. Setup, saving tokens, connecting their agents to tools like electronic drawing software. Question after question, desk after desk. Some of it is IT and infrastructure, which isn't my background. I learned it on my own so I could teach it. On top of that I'm running five agents myself, so my head was switching between my agents, their agents, and their setup problems for hours. By evening I'd done almost none of my own work. The rollout helps the division, but the unit economics for me look bad. One person absorbing the support cost for everyone, with no line item showing it, doesn't hold up for long. I also miss the recognition. I'm not sure my colleagues will remember who set this up once it becomes part of their own workflow. I hope my boss sees the hours. Maybe it's just a Friday. Either way, I'm writing a setup guide and a token-saving cheat sheets.

Loading up the truck again. 😊

Learning used to be starting from basic, foundation, brick by brick. With smart model like opus and fable, it is the opposite. I give the input and output, let it build and figure. It is often smarter and find better ways to solve the issues. Then I RE and learnt from it 😬

With AI at work, I do things faster. 1. I cannot give the impression that I am THAT fast otherwise boss/colleagues expectations will be unreasonably high. So I intentionally submit with a delay. 2. I have much more free time at work because my agents are doing for me. So I need to act a bit busy, otherwise I get more work to do. Hahaha. 3. I used to be limited by tokens (standard plan) but no longer (premium plan). Now my limit, beside my time is computer hardware. I need to remind Claude not to spawn more than 5 agents per project (running 5-8 project at any one time) because my computer runs low on memory. All in all, I think I am at least 5x more productive even with all the "self induced delay". 😂

With proliferation of AI, it is all about ideas. Almost everything that wasn't possible a decade ago, can be made possible now. How do you get ideas? Read. Talk to people. Watch podcasts. Think. Resilient. Be curious. It is no longer optional to learn AI, but imperative in today's world, similar to what Microsoft office was two decades ago.

From Boris Cherny, the creator of Claude Code. He says, “In a year or two, whether you learn to code won’t matter.” Instead, there are three things you should do. 1. Play with AI tools. Try them without fear and learn. 2. Stay on the frontier. Change how you use them every time a new model comes out. 3. Build a harness. Bundle tools, memory, rules, and feedback loops together to suit your work, and keep it running.

AI enables a lot of unimaginable stuff possible in a good way, yet it creates a lot of AI slop or software that looks impressive. AI gen software resembling CRM or ERP looks impressive but no company will pay for it, unless it is an established brand. Either give it foc to gain exposure or reframe the problem statement. It is about credibility or complimentary.

Thanks for hosting. @AWS

I have been building with opus 5.5. Today, I got fable 5.2 to critique codebase by opus. Generally ok with some enhancement. Not much different but the tokens burnt rate is fast.

Don't believe everything that you read in the media, especially social media. Their intention is to capture eye ball. So practise Independent thinking. Especially important, as an investor because you don't want to be misled by the media and make stupid, potentially life changing decisions. Go back to the fundamental question: What drives stock price in the long run?

AI has dramatically changed the way we work. We need to let go of old practices, e.g. writing code (which I enjoy), and embrace developing/building code. Or stop using expensive legacy software and develop our own. Or stop mundane and repetitive workflows and invent new ones with AI tools. We are nearing AGI in some areas where we humans cannot compete with AI. There is no choice. It is an unstoppable force that will push non-practitioners into oblivion. On the flip side, the superpower that AI provides is orders of magnitude greater than any technology we have seen before. It multiplies existing domain knowledge and enables crossover into adjacent fields with much more ease. "AI will not replace you at work, but someone who learns AI will."

I usually plan my schedule for the following week on prior Friday. I forgot that it was a holiday yesterday and plan a number of activities. Now, I got to catch up the activities today. 😂

Remember that stock price is driven by business fundamental over long run. Typically not by -a single person. -a single event. Remove emotion out of the equation. Time to load the truck when transitional events drive stock prices down temporarily.

Most people are still debating whether AI will take their job. The ones pulling ahead stopped debating a while ago and just started using it. I see the gap widen at work every month. Think of it as unit economics. One person with AI produces what a small team used to, and the cost barely moves. Same salary, a much bigger top line. Employers will do that math faster than employees will. I see three groups forming. Superusers use it to work faster, learn faster and compound. Casual users treat it like a friendly chatbot and a better search bar. The third group never learns to use it properly. Group two feels safe today. But the margin between the groups compounds, and compounding is brutal to catch up with. This holds even if AGI never shows up.

Funny 😂😂😂

I feel for the current batch of fresh grads. They graduated just before AI skills became a baseline. They also haven't had time to build domain knowledge. Compare that with a senior who already has the domain and only needs to pick up AI. That is a much shorter climb, and the senior wins that matchup most days. Now look at the teens in school today. They are picking up AI as part of their education. By the time they enter the workforce, they will be fluent. What they lack then is domain knowledge, and that gets built on the job like it always has. So this batch sits in the middle with neither. Seniors have the experience, the next wave will have the AI skills, and the people graduating now have to compete against both. Is that fair? Probably not. It is the market they walked into. It is hard on the employer side too. A fresh grad today costs the same salary but takes longer to produce anything useful. You need to teach domain and AI at the same time. On a quarterly view, the unit economics look bad. Top line contribution is low, and the margin is negative for a while. But I think the math changes if you stretch the horizon. Someone hired now builds both skill sets inside your own systems, on your own problems. By year three or four, they know the domain and they know AI, and they know how your company actually works. That is hard to hire in from outside. The companies willing to look past the short term cost will end up with a bench nobody else has. Most will pass, because the payoff shows up late and the cost shows up now. Just my thoughts on it. Hope the ones who do get a chance make the most of it.

I get asked about AI workflows at work a lot, and it feels good. I'm trying to keep that feeling away from my self-worth. Anyone can be replaced at work, and with AI, almost anyone can build software by asking the right questions. What I want from all this sharing is a bigger stack of knowledge I can carry anywhere, whoever I work for. My output is AI multiplied by what I know. Everyone gets the multiplier, so the domain knowledge is the part I own. Teaching colleagues showed me something. I assumed they could do what I do, and they couldn't. The gap was the skill files I've built over months and the persistent memory that tells AI what I want. I say "build software that does ABCD" and it already knows the language, the UI and my usual constraints. A pipeline then pushes the result to whichever network needs it. Cost per output keeps falling while my context keeps compounding. If the job disappeared tomorrow, that's the asset I'd keep. That is my secret edge.