About
Specialties: investment bank technology, FinTech , crypto
Activity
852 followers
-
Anson Au posted thisBeing professional at work is the right thing to do. Letting work dominate your emotions, or define who you are, is a choice. A career is part of your life. It doesn't have to become your identity.
-
Anson Au posted thisThis weekend, I spent quite a lot of time thinking about my daughter’s education. Not how to maximize her exam score. Not how to make sure she becomes a doctor, lawyer, engineer, or anything else. I realized my responsibility as a parent is much simpler, and much harder: Give her enough capability, judgment, confidence, and opportunity that one day she can make her own choices. I cannot force an academic outcome. I cannot decide what her talents will be. I cannot manufacture a 45-point IB student. What I can do is give her the best environment I reasonably can, expose her to different possibilities, help her learn how to think, and make sure she has enough options when the important decisions arrive. That is also why I increasingly like the philosophy behind IB education. Not because IB guarantees better results. But because, at its best, it asks students to question, research, form an argument, make choices, and defend their thinking. In Hong Kong, I sometimes see a strange outcome: A student gets an exceptional score, and suddenly everyone thinks the answer is obvious. “Study medicine.” Why? Because the score is high enough. To me, that gets education backwards. A high score should create more freedom, not a stronger obligation to follow the most prestigious path available. If one day my daughter gets an exceptional score and decides she genuinely wants to study medicine, great. If she decides she wants computer science, history, design, research, entrepreneurship, or something I would never have chosen for her, that can also be great. What matters is whether she understands herself, understands the trade-offs, can think independently, and is capable of owning the decision. My goal is not to choose a successful life for her. My goal is to help her become capable enough to choose her own.
-
Anson Au shared thisWho needs to read this? 1. Software engineers 2. Hiring managers for any role that requires the use of AI-assisted development tools.Anson Au shared thisThe most important skills in Building and Deploying AI Applications.AI Engineering Skills Map: Building and Deploying AI ApplicationsAI Engineering Skills Map: Building and Deploying AI Applications
-
Anson Au posted thisBecoming a star employee takes years. Losing one can take a single bad decision. Talent is hard to build. Trust is easy to break. And once a star employee stops caring, resignation is usually just the paperwork.
-
Anson Au shared thisAnd impressive achievement - Open source , China AI catch up game https://lnkd.in/gDpv3hPRMoonshot AI’s Kimi K3 faces GPU crunch amid US-China tech rivalryMoonshot AI’s Kimi K3 faces GPU crunch amid US-China tech rivalry
-
Anson Au posted thisExcited to introduce a methodology I've been quietly practicing for years and am finally ready to name: 𝗟𝗶𝗲-𝗗𝗿𝗶𝘃𝗲𝗻 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 (𝗟𝗗𝗗). After Waterfall, Agile, and now Spec-Driven Development, I think we've all been dancing around the real question: what if the artifact everyone actually optimizes for... is the status update? 𝗧𝗵𝗲 𝗖𝗼𝗿𝗲 𝗣𝗿𝗶𝗻𝗰𝗶𝗽𝗹𝗲 The lie is the interface. Code, tests, and documentation are just supporting evidence that occasionally corroborates it. 𝗧𝗵𝗲 𝗟𝗗𝗗 𝗠𝗮𝘁𝘂𝗿𝗶𝘁𝘆 𝗟𝗮𝗱𝗱𝗲𝗿 🪜 🔹 𝗟𝗶𝗲-𝗙𝗶𝗿𝘀𝘁 — You tell the team "this works," everyone nods, and the lie is quietly discarded the moment CI fails (and nobody tells the on-call engineer). 🔹 𝗟𝗶𝗲-𝗔𝗻𝗰𝗵𝗼𝗿𝗲𝗱 — The lie evolves alongside the codebase. Every time reality contradicts it, the lie gets a changelog entry instead of a fix: "known issue, tracked, prioritized" (it is never prioritized). 🔹 𝗟𝗶𝗲-𝗮𝘀-𝗦𝗼𝘂𝗿𝗰𝗲 — The lie is the source of truth. Standups, dashboards, and quarterly reports are generated directly from it. The actual system becomes a regenerable artifact nobody inspects too closely. 𝗧𝗵𝗲 𝗙𝗼𝘂𝗿 𝗣𝗵𝗮𝘀𝗲𝘀 Assert → Deflect → Reframe as Feature → Close the Ticket. 𝗧𝗵𝗲 𝗚𝗼𝗹𝗱𝗲𝗻 𝗥𝘂𝗹𝗲 Use the minimum amount of truth required to stop follow-up questions. 𝗘𝗻𝗳𝗼𝗿𝗰𝗲𝗺𝗲𝗻𝘁 𝗠𝗲𝗰𝗵𝗮𝗻𝗶𝘀𝗺 None, by design. The moment you add a test that can actually fail, you've accidentally reintroduced engineering. 𝗞𝗻𝗼𝘄𝗻 𝗙𝗮𝗶𝗹𝘂𝗿𝗲 𝗠𝗼𝗱𝗲 Occasionally the lie is load-bearing for so long that other lies get built on top of it. This is also, incidentally, how you get microservices. ⸻ In all seriousness: this is a joke about a very real failure mode — confident-sounding output (from people or AI agents) that hasn't actually been checked against reality. The fix isn't more confidence, it's boring stuff: tests, code review, and someone willing to ask "wait, does this actually work?"
-
Anson Au posted thisA note for my daughter, for the day she is old enough to understand it: Be kind, but do not build your life around pleasing others. In relationships, love deeply, but do not lose your dignity, independence, or sense of self. A healthy relationship should support who you are, not require you to become smaller. In your career, work hard and be responsible, but remember that no job title, company, or manager owns your life. Protect your health, your time, your values, and your ability to choose your own direction. You can care about your family, your colleagues, and the wider community without sacrificing yourself completely. Some people may call this selfishness. I would call it self-respect and self-protection. Help others because you choose to, not because guilt, fear, or pressure has taken away your choice. Your career may change. Relationships may change. Other people’s expectations will always change. But your life is yours. Live it with kindness, courage, and boundaries.
-
Anson Au posted thisMid-2025: Ralph Wiggum loop becomes the hottest agentic hack. Put a prompt in a bash script. Let the agent run. Loop until the tests pass. March 2026: Anthropic publishes harness design for long-running apps. The message was not “loops are useless.” It was more uncomfortable: A loop alone is naive. Agents lose context. Agents get trapped in stack traces. Agents judge their own work too generously. Agents can keep moving even when the direction is wrong. June 2026: “Loop Engineering” enters the vocabulary. And suddenly, the loop sounds mature. But the real innovation is not the loop. It is everything around the loop: - The goal. - The harness. - The evaluator. - The permission boundary. - The stop condition. - The audit trail. - The human approval point. This is where enterprises get hurt. They see a sparking idea from builders and adopt it as architecture. But letting naive loops loose on enterprise code does not solve autonomous software engineering. It can burn API credits. Create silent technical debt. Modify systems without enough context. And give management the illusion of progress at machine speed. The danger is not that the loop fails. The danger is that it succeeds just enough to look correct. In enterprise software, passing tests can still mean broken architecture, hidden side effects, security gaps, and technical debt quietly added at machine speed. The loop is not the engineer.
Anson Au commented on a post
1mo
Agree but I think most people not aware of this.
-
Anson Au liked thisAnson Au liked thisMiniMax released H3 a few weeks ago as an open-weight video model, performing close to leading models. This week, fal.ai post-trained H3 for better quality and prompt adherence while optimizing its inference. The result, H3 Max, generates 720p+ video with synchronized audio 35× faster than MiniMax’s original endpoint. As a consumer product builder, I think we’re only scratching the surface. Imagine choosing a movie’s main character, casting yourself in the story, or watching scenes adapt to you in real time. Video could evolve from something we all watch into personalized formats. Just as better cameras and faster networks drove the shift from text → images → video, faster and cheaper generation could unlock the next shift: generative video built directly into products. Imagine choosing a movie’s main character, casting yourself in the story, or watching scenes adapt to you in real time. Video could evolve from something we all watch into something generated for each of us. #GenerativeAI #AIVideo #OpenSourceAI #ProductInnovation
-
Anson Au liked thisAnson Au liked thisIt feels like just yesterday we wrapped the first run of Generative AI Applications for HR at PolyU SPEED — and Cohort 2 has already kicked off. What makes me passionate about this isn’t just sharing AI applications with people. It’s meeting those HR friends and understanding their AI usage stories. The real learning often happens in hearing how someone is actually applying AI in their world — what’s working, what’s messy, and what they’re still figuring out. It’s not one-way lecturing. It’s an occasion for exchanging ideas and experience. So glad to meet everyone, and I’m looking forward to what’s coming. If you’re an HR practitioner exploring AI with the same curiosity, welcome to connect. I’d love to hear your story and keep exchanging ideas.
-
Anson Au liked thisAnson Au liked thisIf judgement becomes the main thing humans have left, how many people in an organisation would you trust to be the judge? I’ve asked five people a version of this question this week.... “How many people in your organisation really understand the ecosystem well enough to make consistently good judgement calls?” The answers ranged from 2% to 10%. It made me wonder about the idea of the “human in the loop.” As AI takes on more of the tasks, is judgement where human value increasingly sits? Understanding the context. The customer. The politics. The history. The personalities. The things that aren’t written down. And if that is where our value sits, what does that mean for the size of organisations? If only 2–10% are trusted to exercise that judgement today, do we eventually need far fewer people in the loop? Or does AI actually make good human judgement more important, and more widely needed? How easy is it to train judgment? I’m not sure. But I think it’s an interesting question.
Experience & Education
Licenses & Certifications
Recommendations received
2 people have recommended Anson
Join now to viewOther similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More