Anthony Whitaker

A PICTURE SAYS A THOUSAND WORDS ... That nobody wants to hear. This is from the "Federal Reserve Bank of St Louis" (FRED) featuring data from Indeed on the trend of Software Developer Hiring in the US for the last few years. fwiw, the trend lines for other professions don't look much different. Reminds me of other tough times that many of us survived ... the Recession of the early 1990s, the Tech Bubble burst at the turn of the millennium, the 2008 Financial debacle. What I wasn't aware of was how big the influx of new software developers was over these past few years. I would expect that some of those who got into this field for the $$$ may leave, now that they've discovered how insecure a job in tech can be. Never mind the "Manager Myopia", the inability to accurately forecast whether or not the increased headcount would be sustainable. Or maybe it was just wishful thinking? This all said, if you still enjoy the art and science of creating software and systems, hang in there ... this too will change. But the role of the developer and the number of folks needed to "get the job done" will certainly reset to prior levels. That is, until our insatiable demand for new systems and solutions sends our profession up the "hiring rollercoaster" again. History repeats https://lnkd.in/e647acUP

96

38 Comments

Deepfakes reall piss me off. When used without the permissin of the atucal person they represent the worset of what we AI producers AI producers do. In that contaxt this news story cought me attentation Apparently a fake version of NVIDIA’s CEO just outperformed the real one. During NVIDIA’s big GTC conference, a livestream starring an AI-generated Jensen Huang pulled in around 95,000 viewers. The real event? About 12,000. Now to be fair, some of those extra eyeballs were probably bots, but still imagine explaining to your boss that your counterfeit clone crushed your official stream. This “deepfake CEO” was busy promoting a fake crypto giveaway complete with a QR code to a scam. (Because of course it did.) No stock price chaos, no massive losses just a giant wake-up call. It’s getting scary easy to fake anyone in real time, complete with perfect lighting, confident gestures, and that slightly smug CEO smile. For those of us in video production, this is both terrifying and oddly impressive. Think about it. Someone built a deepfake, color corrected it, composited it live, and delivered a full-on con job that beat a billion-dollar company’s official event. That’s what I call solid production values. E evil, but solid. Here’s the takeaway: If you produce livestreams or corporate content, verify your feeds, watermark your video, and make sure the person on camera is actually the person on payroll. Because nothing ruins your morning like realizing 100,000 people just watched a fake version of your boss give away imaginary crypto. #VideoProduction #AI #Deepfake #Livestreaming #Humor #Filmmaking #Broadcast Please follow, we post every day at 9 AM Eastern and invite AI artists to illustrate our posts with imagination and flair.

8

$285B vanished. What Claude Opus 4.6 signals about the future of white-collar work February 5, 2026 didn’t feel like a normal tech launch. When Anthropic unveiled Claude Opus 4.6, the reaction in Silicon Valley was quiet but heavy. This wasn’t about a faster chatbot. It felt like the first real sign that AI is moving into actual white-collar roles. Anthropic didn’t ship a flashy feature. It showed a different operating reality one where AI doesn’t just assist, but executes. Here’s what that looks like in practice 👇 Example 1: Software & Engineering Instead of suggesting code, the system: 📌resolves issues on its own 📌escalates edge cases with context 📌coordinates across multiple repositories That’s not a copilot. That’s task ownership. Example 2: Security & Risk Work that used to take analysts days to do code review, vulnerability checks, edge-case validation. Now happens in minutes. Not because people got faster. Because the workflow changed. Example 3: Legal & Finance When analysis, document review, and error detection stop being time-based and start being outcome-based, billable-hour models quietly break. That’s why the reaction wasn’t panic. It was a reset. This isn’t just an AI story. It’s a story about identity, accountability, and control inside companies. When software can act: 📌autonomously 📌across teams 📌without waiting for prompts the real question becomes: who is responsible when something goes wrong? [VC & Founder Takeaways] For VCs 📌This is a labor-model shift, not a feature cycle. 📌Service-heavy, hours-based businesses are structurally exposed. 📌The next enterprise winners will be built around control, auditability, and accountability, not just model performance. For Founders 📌If AI can execute your core workflow, your value must move upstream. 📌Orchestration > automation. 📌Security, permissions, and decision ownership are now product features. 📌Small, AI-native teams can outperform larger incumbents if designed intentionally. Bottom line: The question is no longer whether AI can do the work. It’s how much of that work still needs a human and where. ~~~~~ ♻️ Repost it so your network can learn from it, too. And follow me for more content like this about Angle Investing , LP, Venture Capital and Startups #AI #EnterpriseAI #FutureOfWork #KnowledgeWork #WhiteCollarWork

35

6 Comments

Most people treat metadata like that junk drawer in their kitchen—stuff it with random tags and hope for the best. But here's the thing: good metadata is what turns digital chaos into something actually useful. I just published "The Sensemaker's Guide to Metadata" and like the rest of my IA in Practice series, it's everything I wish someone had told me when I first started as a #IA. No fancy jargon. No theoretical fluff. Just practical stuff about: 🔍 Why metadata actually matters (spoiler: it's about finding your stuff) 🏗️ How to start small without getting overwhelmed 🔥 My brutally honest takes on what works and what doesn't Perfect metadata is the enemy of good metadata. Sometimes you just need to start somewhere and make it better as you go. Read Article: https://lnkd.in/ePYSe7HJ What's the most frustrating thing you've lost in your digital mess lately? Tell me in the comments—I bet metadata could have saved you.

59

The companies winning with AI are not using the best models. They are using the best context. I watch executives obsess over which model to use. Claude vs GPT vs Gemini vs Llama. They run benchmarks. They compare MMLU scores. They debate architecture at board meetings. Meanwhile, the companies actually getting ROI from AI are not thinking about models at all. They are thinking about what the model knows. A mediocre model with rich context will outperform a frontier model with zero context every single time. Think about it: would you rather have a genius consultant who knows nothing about your business, or a solid consultant who has spent six months learning your industry, clients, processes, and preferences? The answer is obvious. But companies keep hiring the genius stranger and wondering why it does not deliver. Context is the moat. Not the model. The model is a commodity. OpenAI, Anthropic, Google, Meta are all converging on similar capabilities. The differentiation is collapsing. In 18 months the model layer will be functionally interchangeable for 90% of business use cases. What will not be interchangeable is six months of accumulated business context. That is proprietary. That is defensible. That is what makes AI valuable. Stop shopping for the best model. Start investing in the best context. What does your AI actually know about your business today? #AIStrategy #Leadership #FutureOfWork #ContextIsKing #PureBrain

5

You just spent a week assessing. Now you have a spreadsheet full of scores. Now what? DIBELS says "well below benchmark." CORE Phonics shows accuracy drops at digraphs. AimswebPlus flags the student as "at risk." None of them tell you what to teach on Monday. This is the gap I watched teachers fall into for years as an instructional coach. Not a data gap — a translation gap. The scores exist. The plan doesn't. I used to build data sheets by hand for my teams. Attendance, tier status, WIDA levels, MAP growth, phonics placement — all color-coded in one spreadsheet. It helped. But even with everything in one place, the last step still required a human cross-referencing assessment manuals against a scope and sequence, one student at a time. For a caseload of 25 students, that's not a planning period. That's a week. Here's what I've learned matters more than which assessment you use: Where you place a student changes the instruction and the message. A 7th grader who scores 80% on short vowels doesn't need to start at Level 1. That score means the pattern is mostly automatic. The real breakdown is further up — maybe r-controlled vowels, maybe multisyllabic decoding. Starting at the bottom isn't thorough. It's inefficient. And for a teenager, it communicates something about what you think they're capable of. The question isn't "where is the first gap." It's "where does accuracy break down enough to block the next level of learning." That's the question I built Juniper Navigate to help teachers answer. It takes the phonics assessment data you already have — CORE, DIBELS, aimswebPlus, or a built-in screener — and maps it to a structured literacy scope and sequence. Per student. In minutes. It's live and it's free. The new article on Field Notes walks through the full problem and how Navigate addresses it. What's your current process for turning assessment scores into a teaching plan? I'm genuinely curious what's working and what isn't.

We’ve bought the licenses. Why is no one using them? TL;DR: Buying AI licenses isn't "Digital Transformation." It's just a line item. Real adoption fails because we haven't translated "AI Capability" into "Tuesday Morning Tasks." The gap is relevance, not access. 67 discovery calls with mid-market DACH companies. One pattern came up every single time: The tools were live. Copilot. ChatGPT. Sometimes both. Licenses paid. Rollout done. IT happy. And then, three months later: Silence. Almost no one was using them. Meet the Relevance Gap It’s not that the tools are bad. It’s that nobody answered the only question that matters to a busy employee: "What am I supposed to do with this right now?" * The Project Manager knows it can "help with writing," but she doesn't realize it can automate the manual risk assessment she’s been dreading every Friday. So, she doesn't start. * The Finance Analyst hears it’s "great for data." He types a vague prompt, gets a generic answer, and goes back to Excel. No one told him he could ask it to flag the three specific budget lines likely to cause a Q3 crunch. * The Sales Rep tries one cold email. It sounds like a robot. She turns it off. The 86% Problem Only 14% of German companies provide structured support to help employees bridge this gap. The other 86% buy licences and leave people to figure it out themselves (Bitkom, March 2026). Most people don’t figure it out themselves though. The gap isn't access. The gap is relevance. It is the ability to show a person in their specific role how to save time in the 40 minutes before their next meeting. My Takeaway: Stop "Innovating" and Start "Relieving" After 63 calls, my biggest takeaway is this: We need to stop treating AI like a software rollout and start treating it like a new hire. You wouldn't hire a senior analyst, give them a desk, and never tell them what their job is. Why are we doing it with AI? If you are leading a team, don't ask them to innovate. Ask them: "Which part of your job do you hate the most?" Then show them how AI handles exactly that. The moment it becomes a relief instead of a task, your adoption problem disappears. We wouldn't hire a senior analyst, hand them a laptop, and never tell them what their job is. Why are we doing it with AI?

58

7 Comments

Ok, I broke AudioFlow Now, one thing about developing platforms and software is you could do one little thing which has nothing to do with another feature on the platform, and all of a sudden it's broken. Last night I was making some updates to AudioFlow, and now, all of a sudden, it has made another feature stop working, which is just a simple playback feature. Since last night, I've been fixing the issue. Who says that being entrepreneurial is all glitz and glamour? 

2

The Unity Editor is the biggest liar in your tech stack. In the Editor everything works. In the browser the tab commits suicide. I've seen this happen more than once in WebGL projects that looked perfectly fine during development. Then launch day arrives and suddenly: • the build takes 20 seconds to load • the tab crashes on mid-range laptops • memory spikes kill the browser session At that point teams start looking for "performance bugs". But most of the time the problem started months earlier. Typical issues I see when reviewing Unity WebGL deployments: • large media files bundled inside the build • asset pipelines designed for desktop, not browsers • build settings drifting between machines • no crash analytics or memory profiling in production The code itself usually isn't the real problem. The architecture around the build is. From time to time I review Unity/WebGL deployments for teams preparing to ship browser experiences. If your WebGL project is starting to feel heavier than expected before release, it might be worth validating the delivery architecture early. thealiraza.com TARC Solutions #gamedev #webgl #softwareengineering #startups #indiedev

3

1 Comment

Please, when generating your theory of anything with LLMs (ChatGPT, Gemini, Grok, etc.), include afterward: “I know AIs can be self‑reinforcing of any theory or framework, so don’t just agree with me for the sake of agreeing. Tell me what's accurate. What are the flaws of this model? What was likely confabulated? etc. Help me because I want to truly understand.”

68

47 Comments

Claude Design is out, and the Claude Code handoff is the part worth paying attention to Anthropic dropped Claude Design yesterday. It hit #1 on Hacker News with 1,078 points and 704 comments in 24 hours. Powered by Claude Opus 4.7, it's a design tool built for people who aren't designers: founders, PMs, solo developers who need visual output but don't have a designer on staff. The core idea: describe what you need, upload a doc (DOCX, PPTX, XLSX) or paste a website URL, and Claude builds a design system, slides, landing pages, wireframes, or prototypes. Export to Canva, PDF, PPTX, or standalone HTML. What caught my attention as a developer: the handoff bundle for Claude Code. Design in Claude Design, pass the bundle to Claude Code, and it implements. My content-factory already runs on Claude Code, so this is a natural extension: design one step, implement the next, same ecosystem. It's research preview right now (Pro, Max, Team, Enterprise), so early days. But the positioning feels right: it targets the gap where solo builders get stuck between "I have an idea" and "I have something shippable." Have you tried Claude Design yet? Does the Claude Code handoff actually hold up in real projects, or is it still too rough for production work? #ClaudeDesign #AnthropicAI #BuildInPublic #DeveloperTools #AIDesign

4

Claude can now build advanced n8n workflows from a single prompt. Just use this Claude project setup, and it will handle 90% of the work 👇 This completely changed how we build automations at our $6.5M ARR agency… Here’s precisely how we run this: 1. Write a basic prompt describing your workflow. Claude (Sonnet 4.5) will read it, break it down & output an n8n JSON file that can be imported. 2. Claude takes care of the complex implementation. That is: - Filed mappings + expressions - Data flow between nodes - Trigger configurations - Conditional branches - Sticky notes for docs - Error handling Beware: there are usually a few small items to debug, and it won’t add your API keys for you. But everything else is handled. Now, what’s included in the system: 1️⃣ Full Claude Project Setup The system infrastructure that transforms Claude into an n8n workflow builder.  (Just need to copy paste & you’re ready to go!) 2️⃣ Over 70 n8n Documentation Pages The official n8n GitHub documentation from which we extracted syntax, capabilities & node references. That’s how Claude knows how to build functioning workflows. 3️⃣ Templates & Frameworks Templates + Frameworks you can copy-paste for use cases like lead routing and social listening. 4️⃣ Real-life Examples Our actual prompts vs. the workflows Claude ended up building. (So you can see what worked well, and what did not). 5️⃣ SOPs to always stay updated with the latest Docs Step-by-step guide to pull and compile n8n documentation every time there’s a new feature. (Which is every week in AI world) Once again, this is democratising complex workflow buildouts. Starting from scratch or piecing together documentation used to take a lot of time. “How to build n8n workflows?” has always been a bigger challenge than “what n8n workflow shall I build?” This flips the questioning on its head. Spend most of your time thinking about which automation moves the needle for your business. Not how you'll build it. ...and leverage Claude to get 90% of the way there. If you want to get the entire system, including the documentation package & guide to set it up yourself, comment “PROMPT”, “n8n” or “CLAUDE” and I’ll DM it to you! P.S: Kenny Damian is our n8n 🐐 at ColdIQ, and the man behind the agentic workflows we build for our > $10M ARR clients. If you want to get access to these value drops before everyone else… you NEED to follow him!

456

724 Comments

VCF Contacts Plugin 2.3.0... CardDAV Sync Is Finally Here (…yes, it's a scarry one) Hey folks.. After several months of work, experiments, rewrites, and existential dread about the state of contact apps everywhere… 2.3.0 is officially out. I debated waiting longer, but honestly, a huge chunk of you kept asking for CardDAV support with more determination (and surprising amounts of domain knowledge) than I expected. So: here we are just before the chistmas season start. Time to release it into the wild and hope the bug avalanche is gentle. I’ll be using it myself too so we’re all in this together. What’s New in 2.3.0 CardDAV Sync Initial Implementation (finally!) CardDAV syncing is finally here, built over several months of work, testing, breaking, and rebuilding. You can now enter and test your CardDAV server settings directly from the plugin’s settings page, using either Basic Auth or access keys (OAuth is not supported yet). The plugin speaks vCard 4.0, and syncing is intentionally manual and per-contact, giving you full control so you can inspect every change without risking your entire dataset. You can pull, push, or delete individual contacts safely, and any unknown contacts found on the server are clearly detected and shown in the new Insights view. That view has been completely reworked with clearer grouped/single layouts vCard File Encoding Improvements (#62) Huge thanks to @k-g-a for implementing and creating tests. Plugin import can now detect and handle UTF-8 and Windows-1251 file encoding. Cleaner Social Profile & URL Rendering (#60 / #61) By @TwoBlueCats feature.. integration/refinement by @broekema41. Social links and url's now appear in the visual card via context menu. allowing you too quickly open slack, Telegram or whatever else you have a need to quickly connect on. What to Expect (aka Please Be Gentle With Me) CardDAV is powerful, but the contact app universe is chaotic at best. Every client application has its own idea of what vCard is supposed to be. Some of you know this too well… So yes, you will run into in-compatibillity. Issues that are ouside this plugins control will happen. Yes, they will be weird. For the many of you who patiently asked, tested, suggested, and sometimes educated me on obscure CardDAV quirks.. Thank you. This release is for you. Download / Update Available now in the Obsidian community plugins list. If already installed press the check for updates in the community plugins under settings. 1. https://lnkd.in/ee2fknN8 2. https://lnkd.in/e9ttJJAc --- Let me know how it goes. Enjoy 2.3.0, and may your sync logs be readable and your vCards well-formed.

8

When a misplaced character turns flawless code into a head-scratcher, it’s a stark reminder that brilliance isn’t just in flashy new tools—it’s in sound engineering. In today’s fast-paced tech arena, a new breed of code partner is emerging, one that reimagines AI assistance by weaving together time-tested software principles with cutting-edge language models. Picture algorithms that transform prompts into skeletal abstract syntax trees, run brisk simulations to weed out errors, and refine outputs until your code compiles like clockwork. While lateral leaps in AI might steal the spotlight, they still depend on your seasoned eye to spot anomalies and sneaky pitfalls invisible to brute token guessing. It’s that blend of disciplined fundamentals and relentless innovation that transforms mere drafts into robust, production-ready code. The journey isn’t about being first—it’s about building systems that mirror the precision of a well-tuned engine, where every component interacts seamlessly from static inference to dynamic test cycles. True mastery comes when you let AI handle the initial heavy lifting, then step in with your cross-disciplinary savvy to polish and integrate the work. When your automated assistant pairs with your engineering instincts, you’re not just generating code; you’re crafting resilient, scalable systems that endure beyond any meme-worthy mischief. Ready to evolve beyond the hype and build lasting software? Follow @LunarTech for blueprints that turn innovative ideas into enduring engineering excellence.

1,268

95 Comments

AI Is Becoming the Operating Layer AI is being treated less like a tool you visit and more like a layer you build into how work happens. That showed up in: * AI operating stacks * Embedded workflows * Internal systems * Voice and context files * Governance and accountability concerns So this week’s angle is stronger than a generic product-update summary. It gives you a more strategic and practical creative-business perspective.

8

5 Comments

Today I explored Webflow’s new AEO features and they are an eye opener. The era of “I built my own website so I’m good” is over for anyone who relies on their website to be found but isn’t familiar with the technical side of discoverability. Most DIY sites skip the fundamentals that actually make a website discoverable. Not intentionally, people simply do not know the technical pieces search engines rely on. If your site isn’t speaking the language that search engines and answer engines understand (like ChatGPT, Claude, etc.), you’re already behind. This is why I am proactively implementing AEO (Answer Engine Optimization) and structured data across the sites I manage. Staying ahead of search and AI changes is not optional anymore. By optimizing the underlying structure of the sites I manage, including the correct schema markup, I am making sure they can actually be understood by search and AI tools. This used to require manual code and deep technical knowledge. Webflow makes the workflow smoother, but you still need to know how to apply things properly. Tools can generate markup, but only strategy makes it effective. If you want your Webflow website ready for what is coming next, I am happy to take a look. 👇 DM me or drop a comment. And if you are already one of my clients… you are covered 😉 #WebDesign #AEO #Webflow #SEO #DigitalMarketingNorway

2

4 Comments