🚀 𝗧𝗵𝗲 𝘀𝘂𝗯𝘁𝗹𝗲 𝘀𝗶𝗴𝗻𝘀 𝘆𝗼𝘂𝗿 𝗔𝗜-𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗰𝗼𝗱𝗲 𝗶𝘀 𝗶𝗻 𝘁𝗿𝗼𝘂𝗯𝗹𝗲 🚀 How do you know when to stop vibe coding and start verifying? Watch for the signals: 🔄 𝗥𝗲𝗵𝗮𝘀𝗵 𝗹𝗼𝗼𝗽𝘀: prompting slight variations over and over without progress 💥 𝗦𝗵𝗼𝘁𝗴𝘂𝗻 𝘀𝘂𝗿𝗴𝗲𝗿𝘆: one small change triggers cascading edits everywhere

"Watch for signs of trouble: lots of mocking, complex setup, too many dependencies—especially needing to modify other parts of the code. When you see those signs, stop vibe coding and read the code." Check out new #Radar article by @andrewstellman.bsky.social: bit.ly/4nQMeOx #AI #VibeCoding

🚀 𝗧𝗵𝗲 𝗔𝗜 𝗿𝘂𝗹𝗲 𝗲𝘃𝗲𝗿𝘆 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗻𝗲𝗲𝗱𝘀: 𝗧𝗿𝘂𝘀𝘁 𝗯𝘂𝘁 𝘃𝗲𝗿𝗶𝗳𝘆 🚀 AI-generated code looks right a lot of the time—but it’s not guaranteed. It’s stitching patterns together, not reasoning about your architecture or long-term design. That’s why I argue the mindset has to be 𝘁𝗿𝘂𝘀𝘁 𝗯𝘂𝘁 𝘃𝗲𝗿𝗶𝗳𝘆.

🚀 𝗘𝘅𝗰𝗶𝘁𝗲𝗱 𝘁𝗼 𝘀𝗵𝗮𝗿𝗲: 𝗺𝘆 𝗻𝗲𝘄 @oreilly.bsky.social 𝗥𝗮𝗱𝗮𝗿 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝗷𝘂𝘀𝘁 𝗱𝗿𝗼𝗽𝗽𝗲𝗱! 🚀 I’m really excited about this one. It’s called “Trust but Verify” and it digs into how we should be using AI coding tools. AI can speed up development in amazing ways, but that doesn’t mean you can switch off your brain.

🚀 𝐓𝐡𝐞 𝐀𝐈 𝐩𝐢𝐭𝐟𝐚𝐥𝐥 𝐧𝐨 𝐨𝐧𝐞 𝐭𝐚𝐥𝐤𝐬 𝐚𝐛𝐨𝐮𝐭… 🚀 Here’s something subtle but dangerous: 𝐜𝐨𝐧𝐭𝐞𝐱𝐭 𝐝𝐫𝐢𝐟𝐭. With humans, misunderstandings show up in meetings. With AI, the answers keep sounding confident until suddenly it suggests a fix that makes no sense. 🧵

🚀 𝐀𝐫𝐞 𝐩𝐫𝐨𝐦𝐩𝐭 𝐥𝐢𝐛𝐫𝐚𝐫𝐢𝐞𝐬 𝐭𝐡𝐞 𝐧𝐞𝐰 𝐭𝐞𝐦𝐩𝐥𝐚𝐭𝐞 𝐭𝐫𝐚𝐩? 🚀 In the 1990s, companies built massive templates and forms for requirements documents, hoping the right format would guarantee the right system. It didn’t. That didn’t work, and it often made things worse… 🧵

🚀 𝐏𝐫𝐨𝐦𝐩𝐭 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐢𝐬𝐧’𝐭 𝐧𝐞𝐰—𝐰𝐞 𝐣𝐮𝐬𝐭 𝐜𝐡𝐚𝐧𝐠𝐞𝐝 𝐭𝐡𝐞 𝐧𝐚𝐦𝐞 🚀 In the late 1960s, the NATO Software Engineering Conference kicked off what became known as the “software crisis.” The problem wasn’t a lack of technical skill… 🧵

🚀 𝐀𝐈 𝐦𝐚𝐝𝐞 𝐦𝐞 𝐟𝐞𝐞𝐥 𝐥𝐢𝐤𝐞 𝐈 𝐤𝐧𝐨𝐰 𝐦𝐲 𝐜𝐨𝐝𝐞 𝐛𝐞𝐭𝐭𝐞𝐫… 𝐛𝐮𝐭 𝐢𝐬 𝐢𝐭 𝐫𝐞𝐚𝐥? 🚀 This surprised me. I’ve always cared a lot about architecture, separation of concerns, and keeping classes decoupled. Normally that takes a ton of focus, because it’s so easy to get pulled down into the weeds of the implementation . 🧵

"Talking about prompt engineering today is really just continuing a much older conversation about how developers spell out what they need built, under what conditions, with what assumptions, and how to communicate that to the team." New #Radar article by @andrewstellman.bsky.social: bit.ly/4nzKfxQ

With Copilot and ChatGPT, I can make sweeping changes and lean on AI tests, which frees me to think more about architecture and design instead of getting lost in the weeds. It feels like I understand my code better.

🚀 𝐓𝐡𝐢𝐬 𝐢𝐬 𝐰𝐡𝐚𝐭 𝐀𝐈-𝐚𝐜𝐜𝐞𝐥𝐞𝐫𝐚𝐭𝐞𝐝 𝐭𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐝𝐞𝐛𝐭 𝐥𝐨𝐨𝐤𝐬 𝐥𝐢𝐤𝐞 🚀 This Reddit post hit me hard. A new developer tried to build a project using Cursor and ChatGPT. After three months, every small change broke everything else. Debugging took days. Energy ran out. They gave up.

Issues that seem small at first can compound quickly, making code harder to understand, maintain, and evolve. To really see that danger, you have to look at how AI is used in practice—which for many developers starts with #VibeCoding. @andrewstellman.bsky.social bit.ly/3K2DXbz #GenAI

🚀 𝐌𝐲 𝐧𝐞𝐰 @OReilly.bsky.social 𝐑𝐚𝐝𝐚𝐫 𝐩𝐢𝐞𝐜𝐞 𝐢𝐬 𝐮𝐩: “𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐀𝐈-𝐑𝐞𝐬𝐢𝐬𝐭𝐚𝐧𝐭 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐃𝐞𝐛𝐭” 🚀 “Vibe coding” is real and it works, but there are hidden dangers that can creep in if you never stop to review what the AI just gave you. 🧵

"A rehash loop is a signal that the AI ran out of context. When you’re stuck in a rehash loop, treat it as a signal instead of a problem. Figure out what context is missing and provide it." Learn what to do when #AI gets stuck in this new #Radar article by @andrewstellman.bsky.social: bit.ly/4p5wBEn

🚀 𝐌𝐲 𝐧𝐞𝐰 @oreilly.bsky.social 𝐑𝐚𝐝𝐚𝐫 𝐩𝐢𝐞𝐜𝐞 𝐢𝐬 𝐮𝐩: “𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐭𝐡𝐞 𝐑𝐞𝐡𝐚𝐬𝐡 𝐋𝐨𝐨𝐩” 🚀 If you’ve used AI for code, you’ve probably run into this: the model circles the same few solutions that don’t work, but you can’t to nudge it past to get what you need. That’s the 𝐫𝐞𝐡𝐚𝐬𝐡 𝐥𝐨𝐨𝐩—and it’s maddening.

My new @oreilly.bsky.social Radar piece is up, and I'm so excited about it! Ever been stuck in a rehash loop when you're coding with AI? The AI keeps circling the same solutions. This is an important signal that 𝐭𝐡𝐞 𝐀𝐈 𝐫𝐚𝐧 𝐨𝐮𝐭 𝐨𝐟 𝐜𝐨𝐧𝐭𝐞𝐱𝐭. Check out my article and learn what you can do about it! 👀

🚨 𝐍𝐞𝐰 𝐎'𝐑𝐞𝐢𝐥𝐥𝐲 𝐑𝐚𝐝𝐚𝐫 𝐩𝐢𝐞𝐜𝐞: 𝐌𝐂𝐏, 𝐃𝐞𝐞𝐩 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧, 𝐚𝐧𝐝 𝐁𝐢𝐠 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬 🚨 I'd love for you to check out my latest piece published in @oreilly.bsky.social Radar. “𝑴𝑪𝑷 𝑰𝒏𝒕𝒓𝒐𝒅𝒖𝒄𝒆𝒔 𝑫𝒆𝒆𝒑 𝑰𝒏𝒕𝒆𝒈𝒓𝒂𝒕𝒊𝒐𝒏—𝒂𝒏𝒅 𝑺𝒆𝒓𝒊𝒐𝒖𝒔 𝑺𝒆𝒄𝒖𝒓𝒊𝒕𝒚 𝑪𝒐𝒏𝒄𝒆𝒓𝒏𝒔.”

MCP basically wires your AI assistant straight into your dev environment. It sends files, code snippets, even whatever you just highlighted—automatically. It’s really useful. But if you’re in a regulated industry, it can be a compliance nightmare.

MCP is the new shiny thing in dev tools. And sure, it’s useful. AI assistants have always peeked at your files, your cursor, your selections—but MCP formalizes it. Now there’s a standardized pipeline sending that context around to... wherever?

🚨 𝐌𝐂𝐏: 𝐓𝐡𝐞 𝐧𝐞𝐰 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐯𝐢𝐭𝐲 𝐛𝐨𝐨𝐬𝐭 𝐭𝐡𝐚𝐭 𝐜𝐨𝐮𝐥𝐝 𝐛𝐫𝐞𝐚𝐤 𝐲𝐨𝐮𝐫 𝐜𝐨𝐦𝐩𝐥𝐢𝐚𝐧𝐜𝐞 🚨 MCP—the Model Context Protocol—plugs AI assistants directly into your environment. Your open files, your editor selections, even the last thing you typed. More in my @oreilly.bsky.social Radar piece... 🧵

🚨 𝐌𝐲 𝐎'𝐑𝐞𝐢𝐥𝐥𝐲 𝐋𝐢𝐯𝐞 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐜𝐨𝐮𝐫𝐬𝐞 𝐨𝐧 𝐮𝐬𝐢𝐧𝐠 𝐀𝐈 𝐟𝐨𝐫 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭 𝐬𝐭𝐚𝐫𝐭𝐬 𝐢𝐧 𝐨𝐧𝐞 𝐡𝐨𝐮𝐫! 🚨 ✨ You can still attend for free with an @oreilly.bsky.social trial! Details below. Copilot and ChatGPT for Java and C# Developers 📅 Live April 16 🔗 learning.oreilly.com/live-events/...

🚨 𝐋𝐚𝐬𝐭 𝐜𝐡𝐚𝐧𝐜𝐞! 𝐌𝐲 𝐎'𝐑𝐞𝐢𝐥𝐥𝐲 𝐋𝐢𝐯𝐞 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐜𝐨𝐮𝐫𝐬𝐞 𝐢𝐬 𝐭𝐨𝐦𝐨𝐫𝐫𝐨𝐰 🚨 ✨ You can still attend my @oreilly.bsky.social course for free! Details below. Copilot and ChatGPT for Java and C# Developers 📅 Live April 16 🔗 learning.oreilly.com/live-events/... This isn’t just about cool demos or AI tricks.

🚨 𝐋𝐚𝐬𝐭 𝐜𝐡𝐚𝐧𝐜𝐞! 𝐌𝐲 𝐎'𝐑𝐞𝐢𝐥𝐥𝐲 𝐋𝐢𝐯𝐞 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐜𝐨𝐮𝐫𝐬𝐞 𝐢𝐬 𝐭𝐨𝐦𝐨𝐫𝐫𝐨𝐰 🚨 ✨ You can still attend for free! Details below. Copilot and ChatGPT for Java and C# Developers 📅 Live April 16 🔗 learning.oreilly.com/live-events/... This isn’t just about cool demos or showing off AI tricks.

OpenAI just released ChatGPT 5… and wow. I was fixing a bug in a shell script I use to edit video files. It thought for 29 seconds and this is what it came back with. WTF 🤦🏻 I’ve been getting responses like this all day. Is anyone else seeing GPT-5 completely lose its 💩, or is it just me?

OpenAI just released ChatGPT 5… and wow. I was fixing a bug in a shell script I use to edit video files. It thought for 29 seconds and this is what it came back with. WTF 🤦🏻 I’ve been getting responses like this all day. Is anyone else seeing GPT-5 completely lose its 💩, or is it just me?

I wonder if this would get Copilot to ignore anything you added to the project instructions in .github/copilot-instructions

🚀 𝐂𝐨𝐩𝐢𝐥𝐨𝐭 𝐫𝐞𝐚𝐝𝐬 𝐲𝐨𝐮𝐫 𝐜𝐨𝐦𝐦𝐞𝐧𝐭𝐬. 𝐇𝐞𝐫𝐞’𝐬 𝐰𝐡𝐚𝐭 𝐭𝐡𝐚𝐭 𝐦𝐞𝐚𝐧𝐬 𝐟𝐨𝐫 𝐲𝐨𝐮𝐫 𝐜𝐨𝐝𝐞 🚀 I’ll be teaching this live next week, and this part always gets devs thinking differently about how they code with AI. ✨ Sign up today and take advantage of a free 10-day O’Reilly trial—just in time for the course:

The more you give, the more you receive.

WEIRD AI hallucination of the day: ChatGPT accidentally turned into a Taiwanese YouTuber. Apparently “please like and subscribe” is the new lorem ipsum? So I hit the mic button to dictate a comment in ChatGPT, then immediately tapped the checkmark before saying anything.

🚀 How Copilot “reads” your code—and how to help it do better 🚀 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 𝗱𝗼𝗲𝘀𝗻’𝘁 𝗷𝘂𝘀𝘁 𝗴𝘂𝗲𝘀𝘀. It pulls in code from all around your cursor, checks variable names, comments, and patterns across your whole project, and tries to figure out what you’re building.

Exactly 7 years ago I got to watch Allie Quigley absolutely crush the

Here’s one of my favorite AI tips for getting better SQL queries from Copilot (or ChatGPT, Claude, etc.). Before you ask it to write the query, ask it to give you a query that lists the column names and types for the table you’re working with. Then run that and paste the results back into the chat.

A little career advice for junior folks. People don't always appreciate "constructive" feedback.

Here’s one AI technique I use all the time. When I want to make sure the AI really understands a piece of code, I ask it to refactor it. I’ll give it something specific to do. Want to try this yourself? Here are a few ideas: • Split this method into smaller ones

Why I'm not (too) worried about AI taking all our dev jobs, part 47: designing software isn't about making statistically average decisions, it's about understanding your users and their problems and coming up with something new to solve them.

Whenever I do a training session with this slide, the word “partner” always seems to trigger someone. I actually like that—it means some people are skeptical, which is probably a good thing now that AIs are getting increasingly autonomous, at least when it comes to letting them run wild in our code.

🚀 𝐂𝐨𝐩𝐢𝐥𝐨𝐭 𝐜𝐚𝐧'𝐭 𝐫𝐞𝐦𝐞𝐦𝐛𝐞𝐫 𝐚𝐧𝐲𝐭𝐡𝐢𝐧𝐠—𝐮𝐧𝐥𝐞𝐬𝐬 𝐲𝐨𝐮 𝐠𝐢𝐯𝐞 𝐢𝐭 𝐚 𝐦𝐞𝐦𝐨𝐫𝐲. 🚀 I got this great Copilot tip straight from one of my contacts at GitHub, and I wanted to share it with you. It started out as something one of their engineers did on their own project—and it turned out to be so useful…

i mean this is kind of the key problem. there's already a way to 'prompt' a computer to perform a task. its called programming. anything else implies the output is inevitable, or is trying to upscale from a lossy compressed source

🚀 𝐓𝐡𝐞 𝐃𝐨 𝐖𝐡𝐚𝐭 𝐈 𝐌𝐞𝐚𝐧𝐭, 𝐍𝐨𝐭 𝐖𝐡𝐚𝐭 𝐈 𝐒𝐚𝐢𝐝 𝐏𝐫𝐨𝐛𝐥𝐞𝐦. 🚀 This is something dev teams have been dealing with forever. You're a product owner. You write some great requirements or user stories and give them to the team. The team builds exactly what was asked for…

🚀 𝐂𝐨𝐩𝐢𝐥𝐨𝐭 𝐰𝐨𝐧’𝐭 𝐫𝐞𝐚𝐝 𝐲𝐨𝐮𝐫 𝐦𝐢𝐧𝐝. 🚀 𝐼𝑓 𝑦𝑜𝑢 𝑤𝑎𝑛𝑡 𝑔𝑜𝑜𝑑 𝑠𝑢𝑔𝑔𝑒𝑠𝑡𝑖𝑜𝑛𝑠, 𝑦𝑜𝑢 ℎ𝑎𝑣𝑒 𝑡𝑜 𝑔𝑖𝑣𝑒 𝑖𝑡 𝑠𝑜𝑚𝑒𝑡ℎ𝑖𝑛𝑔 𝑡𝑜 𝑤𝑜𝑟𝑘 𝑤𝑖𝑡ℎ. Copilot, Claude, ChatGPT, and other AIs don't know what you’re actually trying to build. They can’t see your design decisions or your goals. They only have the code, the comments around…

🚀 𝐖𝐡𝐚𝐭 𝐢𝐟 𝐰𝐞 𝐜𝐨𝐮𝐥𝐝 𝐡𝐚𝐯𝐞 𝐚 𝐟𝐫𝐞𝐞, 𝐨𝐩𝐞𝐧 𝐚𝐥𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐯𝐞 𝐭𝐨 𝐂𝐡𝐚𝐭𝐆𝐏𝐓 𝐚𝐧𝐝 𝐂𝐨𝐩𝐢𝐥𝐨𝐭? 🚀 Honestly, I 𝙡𝙤𝙫𝙚 seeing this. A university-led, open project aiming to build a real alternative to the closed, corporate LLMs we all use now.

🚀 𝐏𝐫𝐨𝐦𝐩𝐭𝐢𝐧𝐠 𝐢𝐬𝐧’𝐭 𝐦𝐚𝐠𝐢𝐜. 𝐈𝐭’𝐬 𝐚 𝐬𝐤𝐢𝐥𝐥. 🚀 I really hate when people act like prompt engineering is some kind of magic trick you can't understand. It just isn't. Sure, Copilot and ChatGPT can give you wildly different answers depending on how you ask.

🚀 𝐇𝐨𝐰 𝐝𝐨 𝐲𝐨𝐮 𝐤𝐧𝐨𝐰 𝐰𝐡𝐞𝐧 𝐭𝐨 𝐬𝐭𝐨𝐩 𝐚𝐧𝐝 𝐜𝐡𝐞𝐜𝐤 𝐰𝐡𝐚𝐭 𝐭𝐡𝐞 𝐀𝐈 𝐠𝐚𝐯𝐞 𝐲𝐨𝐮? 🚀 I've been thinking about that moment when you’re working with Copilot or ChatGPT and it’s all flowing. You're in that "vibe coding" zone and you want to keep going, but that flow can be risky.

🚀 𝐀𝐈 𝐝𝐨𝐞𝐬𝐧’𝐭 𝐤𝐧𝐨𝐰 𝐲𝐨𝐮𝐫 𝐩𝐫𝐨𝐣𝐞𝐜𝐭 𝐥𝐢𝐤𝐞 𝐲𝐨𝐮 𝐝𝐨. 🚀 One thing I try to get across when I teach is that AI will happily give you code that seems fine on the surface but fails in ways that are easy to miss. I've seen it spit out methods using variables that don’t even exist...

🚀 I’m excited to announce my next O'Reilly Live Training session! 🚀 𝐂𝐨𝐩𝐢𝐥𝐨𝐭 𝐚𝐧𝐝 𝐂𝐡𝐚𝐭𝐆𝐏𝐓 𝐟𝐨𝐫 𝐉𝐚𝐯𝐚 𝐚𝐧𝐝 𝐂# 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫𝐬 📅 August 12, 2025 🔗 learning.oreilly.com/live-events/... I built this session for developers who want a real, practical way to use AI tools effectively in their work.

Rainbow in Brooklyn

🚀 𝐌𝐲 𝐧𝐞𝐰 𝐎'𝐑𝐞𝐢𝐥𝐥𝐲 𝐑𝐚𝐝𝐚𝐫 𝐩𝐨𝐬𝐭 𝐢𝐬 𝐥𝐢𝐯𝐞! 🚀 The Sens-AI Framework: www.oreilly.com/radar/the-se... I’ve spent a lot of time these past couple of years thinking about how developers actually use AI tools in real projects—and where they get stuck.

I love vibe coding with Copilot. I’ll let it run wild for a bit just to see what it spits out. But sometimes I catch it doing something stupid, so I have to stop it and ask why it did that.

It's Friday! Got this fortune on Monday, waiting all week—can't wait to find out what it is!

🚨 𝗦𝗖𝗔𝗠 𝗔𝗟𝗘𝗥𝗧 (New York) 🚨 I just got a call from "Anthony at ConEd" saying my power was getting disconnected because my account wasn't paid. I logged in and checked my account: $0.00 balance, it was 100% paid up.