Because I don't feel like giving him. Hahahha. I "trained it" myself at home for a week before porting to office notebook.

When I meant "train AI", I don't mean training it technically. I mean 1) Skills (slash commands): your repeatable workflows packaged so one command replaces pages of prompting. 2) Memory and preferences: what it already knows about your role, your rules and what "good" looks like to you. 3) Project instructions and corrections: every time you fixed its output and the fix stuck.

We are prepared.

A colleague wanted to do what I do: schematic, component placement and PCB routing with Claude Code. I helped him install it and get through the IT policy. That part was easy. Then we ran the same prompt on both machines. Mine delivered. His had no idea what was being asked. The gap is the setup. I work through a handful of slash commands (skills), and each one replaces pages of prompting he'd have to write from scratch. My AI already knows my context, my rules and what a good result looks like to me. He saw it right away. Same model, same prompt, very different output. I'd always thought of the time spent building skills as overhead. Now it looks more like capital. A new AI starts at zero, while mine starts with the compounding already done.

Happened to me a few times when context is >80%. Nowadays, I keep my context window to less than 70%

AI is cheap right now because consumers are being subsidised. That won't last. Use it while it's still this affordable.

Claude code on opus 5.5, feels dumber yesterday. Hope it recovers and be well today.

I see. That's why I am an engineer. I only know 1s and 0s. Hahaha.

The current trend is to RE/replace expensive legacy software with open source. Doable for 80% of the software out there. Interesting times. Another industry is going to be disrupted.

Oh, how ? Fsd approved in msia?

Interestingly, I think the world is heading toward hybrid. Frontier models and open weights, both running side by side. Frontier is how you find out what's possible. You pay for the best model to show you how a job gets done fast, what the ceiling looks like, which workflows are worth building at all. Open weights is where you actually build. Sovereign AI. Internal super agents that sit on your own hardware, touch your own data, and never leave the building. No per-token bill scaling with every new use case either, which matters once agents run all day. So one is for discovery, the other for deployment. Rent the frontier to learn the playbook, then own the stack to run it. Will open weights catch up enough to make frontier optional? Maybe. For now I'd want both. Just my thoughts.

The barrier to coding has collapsed. Total collapse. It is no longer who can code but who can create the pipeline and orchestrate agents to do so.

Domain knowledge is what separates AI slop from finished work. You can tell when something looks right but isn't. Run five agents at once. Output multiplies but review capacity stays flat and decision fatigue sets in fast.

Harder to for IT policy to block CLI. 😬

If you have been working with/in G, you will understand how it works. Before a statement is made, rigorous discussion on what-if would have been deliberated. Most angles would have been covered, minimum surprises when something is executed. This is akin to typical behaviour of an adult vs teenager. Teenager don't think much, just do it and think how to handle when music comes. It can be a good thing too, given more mistakes to learn from. Contrary, adult tends to think more before doing, trying to cover all angles before implementation. Yes, it will take longer to push something but the risk/ worst case scenario are mitigated. Neither is bad. But it depends on the circumstances — how high are the stakes and the consequences of failure.

That's also true on propaganda. I guess it works for both sides.

There are reasons why some countries implement certain rules, even it may be unpopular or "limit" human rights. Physical rallies. Spreading false falsehoods. Mass immigration of different cultures. No correction on what's right or wrong. Not holding falsehoods spreader accountable. Over time, without correction, false news may become the truth to some. The repeated exposure (to fake news) brainwashed the brain. That is the danger of social media. That's how social fabric falls to the populism politics, prevalent in today's world of extremist propaganda. I value safety in my community/ country where my family can walk freely without fear.

There are some people who complain about this and that, blaming others for something, whereby others may not have control over, or just want to vent their frustration on others. Blame father. Blame mother. Blame government. Blame everyone except themselves. Why? It subconsciously make them feel better, relieve of any ounce of self responsibility and accountability. It is exactly this kind of negativity and blaming attitude that limit one's growth. Be grateful for what we have. Don't blame others. Work on it. Positivity breeds positivity and leads to growth and a happier life. That's why it is often said that it is important to practise gratitude. 🙏

Nice! I have mostly completed step 3 for what is possible within my company IT policy. Some are airgapped so I need to trend carefully. Now doing step 4 to remove myself from eval. But of course, step 1 to 3 are ongoing processes.