Martin Orton

Just in case his wrong, I'm going to ignore his advice 🤓

Just when I thought I've seen everything! 😱

So, while AIs are busy destroying a lot of "white-collar" jobs, what "rare skills" will save you from the carnage? Ask an AI to fix a bug in your "vibe code", and it'll do it in about four seconds. 🤖 Beautifully. Confidently. Sometimes it's even the right bug. That's the part nobody puts on the pitch deck. AI doesn't make you faster at being right. It makes you faster at whatever you asked for. Ask the wrong question and you now get the wrong answer at scale. 🚀🔥 So which human skills actually hold their value? I've read some posts and thought about it, and these six seem legit. After 25 years of building for the web, I'd struggle to argue with any of them: 🎯 Problem Framing and Critical Thinking: Working out what the problem actually is before anyone starts solving it. AI answers the prompt you gave it, not the one you meant. Garbage in, garbage out. It's only as good as what you feed it. 🔍 Verification and Judgment: Models will hand you wrong numbers and dodgy code with the swagger of someone who just discovered crypto. Someone has to check. I always find it hilarious when some sales pundit claims they've "vibe coded" something without knowing even the basic fundamentals of development. A "normie" shouldn't "vibe code"! Period! I don't care what you think! 🧩 Systems Thinking: AI is great at one piece. Security, latency, cost and the legal team are still your problem. 📚 Domain Depth: General knowledge is now free. Knowing your field properly is the bit nobody can prompt their way into. 🤝 Trust and Accountability: Explaining the risk honestly and putting your name on the decision. Nobody's firing the chatbot. ⚡ Fast Learning: Learn the tool, build something small, get feedback, adjust. Repeat every few months, because the tools change that often. Notice what's not on the list? Typing speed. Memorising syntax. Knowing which button to press. The routine stuff was always the least valuable part of the job. AI just made it obvious. (Mildly uncomfortable, but obvious.) #AI #FutureOfWork #SoftwareDevelopment

🤖 AI made knowledge cheap. It also made real experts harder to spot. Your client can now get a decent overview of almost any topic in 30 minutes with ChatGPT. They're also drowning in AI news and AI pitches. So the cold outreach, the simple demo, and the "educational" call that worked in 2023? Dead in 2026. ☠️ I read a sharp piece on this that borrows "the flip" from The Four Conversations: getting the client to see you as the prize, not the supplicant. 🏆 ✅ Flip happens: you get the interesting work and the respect. ❌ No flip: you're rushed, commoditised, and underpriced. 💸 It's harder now, but not impossible. Here's what you can do: 🤝 Build real, face-to-face relationships and give value before asking for anything. 🧪 Run divergent projects (internal tools, simple AI replacements for Canva or Premiere) that spark client imagination. 🧠 Go deeper than the shallow AI research everyone else is handing over. ✍️ Publish long-form content consistently, because expertise shows over time. 🔒 Double down on existing clients. Deliver a 10/10 and trust becomes your moat. 🚀 Partner with organisations that already have your ideal customers and align incentives so they want to send clients your way. The uncomfortable bit is that you can't hack your way into being seen as an expert. You have to actually deserve it. 🛠️ When everyone has the same AI, trust, and depth, only relationships are left. Treat your business like a long-term asset, not a lead-gen funnel.🌱 Which of these are you actually doing right now?👇 #AI #Consulting #Freelance #PersonalBrand #TrustIsTheMoat

This one is for the Flat Earthers!😉

🔥 AI coding is making some developers worse at software development! And I say that as someone who's fully embracing AI. The problem isn't AI writing code. The problem is developers who stop writing code altogether. 🤖 If an AI agent can build a feature in 10 minutes that would have taken me two hours, I’m going to use it. But I still want to understand what it built. Why? Because when the generated code fails, produces unexpected behaviour, or makes a fundamentally bad architectural decision, prompting the AI again won't necessarily fix the problem. In fact, it could make the problem worse. 🧠 Understanding the code is the answer. There’s a huge difference between: "The AI wrote it, so it must be fine." and: "The AI wrote it, but I understand exactly what it's doing and why it did it." That distinction will become increasingly important. 💻 I still believe developers should write code manually. Especially when learning something new. Writing it yourself forces you to confront: 🧩 How the underlying technology actually works 🔍 Where the constraints and edge cases are 🚩 What bad code looks like 🏗️ Whether the architecture makes sense 🐛 Why something is breaking ⚙️ What trade-offs are you actually making AI is exceptionally good at generating solutions. But it doesn't magically give you the experience required to judge those solutions. And that's what should worry all developers. If someone has never built something from scratch, how do they know when an AI-generated solution is brilliant, mediocre, or fundamentally flawed? They don't. They have to trust the machine. And trusting generated code you don't understand is a very dangerous professional habit. The irony is that AI could actually make good developers dramatically more valuable. People who understand architecture, systems, databases, APIs, security, performance, and the underlying code can use AI to move ridiculously fast. 🚀 AI doesn't have to make technical expertise obsolete. It potentially makes technical expertise more valuable. My approach is simple: ✅ Let AI do the boring work. ✅ Let AI do the repetitive work. ✅ Let AI accelerate the work. But don't outsource your understanding. You still have to understand the big picture and the fundamentals. Because if the AI disappears tomorrow, you should still know how the system works. And if it doesn't disappear? You'll be the person who knows whether what it produced is actually any good, and you WILL be held accountable for what you or the AI generated. That's a much more valuable skill than simply knowing how to prompt it. I cringe when I see these hacks on YouTube selling the fantasy of "I've created this money-making app using AI in one day, delivering $ xxx MRR, and no developers are needed." These people don't know a single thing about coding, and what they do is extremely misleading!😠

🌍 Your passport is now worth less than your employer! ➡️🏢 Sounds crazy, right? But hear me out. There is a method in my madness. 😊 For millennia, where you were from shaped almost everything about your life. That's changing. Fast! Countries mostly provide you with tangible benefits: roads, police, firefighters, and perhaps, if you're lucky, healthcare. And that's assuming you're not living in a corrupt 2nd/3rd world country run by a dictator or kleptocracy (but that's a different subject altogether). In return for these oh-so-wonderful "benefits", they take 20–55% of your income in taxes. But the world is becoming digital-first. And by that I mean an AI‑driven world. The real leverage isn't physical infrastructure anymore. No, no, it’s now: 🔐 How strong is your cybersecurity 🤖 Which AI models can you access ⚙️ How much compute/tokens can you burn 🧠 What internal tools, automations, and infrastructure can you access 🚀 And lastly, whether you can stay on the cutting edge of AI No country is going to give you unlimited tokens, internal LLMs, or world‑class development tools. Great companies will. A great employer can offer you: ✅ Access to the latest (sometimes internal) AI models ✅ Self‑hosted/local models and powerful internal software ✅ Serious investment in your skills and education ✅ Proper security and infrastructure so you can build fast and safely In other words, a company is the new country. Sounds bizarre if you think about it, but that's the world we're living in currently. Over the next decade, who you work for will matter far more than which passport you hold. We're already seeing AI companies with more practical power than many nation-states. Can you imagine 10,000 coordinated AI agents targeting a country's digital infrastructure? That's not sci‑fi anymore. It's now sci-fa (is that even a word?) So if you're planning your career in 2026 and beyond, don't just ask: "Which country should I move to?" 🌎 Start asking: "Which company's infrastructure, tools, and AI ecosystem do I want to plug my life into?" Because your employer might end up being the most important "country" you ever join.

🚀 According to Search Engine Journal, Google DeepMind just unveiled a new AI search ranking approach that could seriously shake up how search works and what SEO looks like. Instead of today's common two‑stage setup (Dual Encoders for fast retrieval + Cross Encoders for precise reranking), they propose Autoregressive Ranking (ARR): a single LLM that directly generates an ordered list of document IDs. To train it, they use SToICaL (Simple Token‑Item Calibrated Loss), which teaches the model to: 🔼 Boost the probability of tokens that lead to highly relevant documents 🔽 Push irrelevant documents further down the results 🧪 In experiments on WordNet and an e‑commerce dataset (ESCI Shopping Queries), ARR with SToICaL: 📊 Often matched the performance of Cross Encoders (the more accurate but expensive models) 🥇 Clearly outperformed Dual Encoders, especially at keeping irrelevant results below relevant ones Trade-offs remain: in some shopping‑search scenarios, one variant got slightly worse at picking the single best result, even while improving overall ranking quality. 📐 Theoretical work in the paper argues that: 📈 Dual Encoders need larger and larger embedding dimensions as the corpus grows ♾️ ARR can, in theory, rank an arbitrary number of documents with a constant‑size model This isn’t a production system yet, but it points toward a future where a single LLM could handle ranking end‑to‑end. A profound shift for search, SEO, and AI search optimisation. For now, Dual and Cross Encoders still underpin ranking, but the direction of travel is clear. 🔍🤖

🚀 The Comeback of OpenAI This might go down as one of the great stories in business history. Just months ago, Anthropic had a clear lead in AI coding, dominating with the release of Fable 5. Fast forward to today, and OpenAI has taken the crown back with GPT-6 Astra, and it's not even close. 😲 So what changed? Three things stand out: 1️⃣ Compute 💻 OpenAI took a major risk, raised more capital, signed bigger deals, and stacked up significantly more GPUs. A bold bet that's now paying off big time. 2️⃣ Focus 🎯 After a distracted 2024–2025 filled with scattered product launches (and a bittersweet GPT-5 release), leadership chose to double down on what matters most: building great models. Sora and Atlas were shut down so the team could concentrate fully on Codex and next-gen AI. 3️⃣ Product Taste ✨ From the early days of ChatGPT (which Sam Altman famously pushed for when others doubted it) to today's polished Codex app and remarkably realistic voice mode, OpenAI has consistently shown a knack for building products that feel intentional, clean, and ready for everyone. Not just experts. The result? With GPT-6 Astra, OpenAI isn't just back in the race; they're leading it again. 🏆 💡 Sometimes the biggest comebacks come down to conviction, focus, and relentless execution.

Happy Spring Day, South Africa! All the flowers in my garden!

📈 AI inference spend is growing exponentially, and it's coming out of budgets that used to go somewhere else entirely. The argument in short: 🤖 Models and agentic systems now handle work that used to need a person or a purpose-built app. Legal questions, translation, research, coding, even clicking through interfaces. 🔌 Agents keep multiplying and plugging into more APIs and tools, and people are increasingly willing to let an agent operate software on their behalf. 💸 So the money follows. Spend that went to human labour or conventional apps is being redirected to tokens. 🎯 The conclusion: if a task has a clear, verifiable outcome, an agent will eventually do it. That would make inference one of the largest markets in computing. Worth sitting with if you sell software, services, or hours.

This is shocking! 😱 And it costs a small fortune! 🤬

OMG, this is so creepy! 😱

He/she/them/they was a good little robot! Served them well. I'm dead serious! RIP 🤣 🤖 🪦

We are sooooo fucked! 😂

I love this use of mnemonics! I created a mnemonic to remember what to bring when I leave the house, for whatever reason. Things I find essential, in case I might need them for some reason. At that moment, I considered it associative memory, but in reality it was a mnemonic system.

In 1995, I was in my senior year in high school. I would have never dreamt I would see some of the tech used in this video clip. I knew some of it would eventually materialise. However, not in my lifetime. Sadly, nor the degradation of society's general ethics/morals. Perfect example of a double-edged sword, right?

Brilliant! 😂

Watch the acronym. BRICS is shedding its R in real time. 🤔