That's how you turn AI from autocomplete into a real engineering teammate. #AIEngineering #ClaudeCode #Cursor #SoftwareEngineering

Microsoft and @satyanadella just laid out their vision for LLMs, but it doesn’t make sense. LLMs are a very useful way to query data. For searching documents it obviously works great for non-developers. But part of the reason it works so well, even though it’s a very sloppy way to query data, is that the response can be very sloppy. That’s because users immediately curate the results and simply perform a new query if they don’t like the results. This is a similar situation to having code autocomplete when the developer is watching and immediately changes the results when wrong. But where I think non-developers disconnect from reality is when they think they are creating automation that is reliable when they run a query that appears to work. They are actually performing a less reliable query than the worst trained database guy, but they don’t know it. Imagine a database guy that has no training and writes database queries and immediately ships whatever appears to work. It’s going to be insanely unreliable for two reasons. First it probably didn’t really work the first time, but it seemed like it because he just glanced at the output. Second, when the data in the database changes the output will miss things or include things it shouldn’t. These type of errors are fine when search documents, but if they are included in automation that needs to be reliable it doesn’t work at all. This is why if you think your “prompts” or “prompt history” is going to become your companies IP (the automation that makes your company valuable) you are going to be disappointed.

All 3 are live on the Apify store: • Google Trends Keyword Monitor • Google Trends Content Calendar • Google Autocomplete Keyword Scraper Reply with a seed keyword and I'll run a free sample so you can see real output. #SEO #KeywordResearch #buildinpublic Links below 👇

The autocomplete one is my favorite. It expands each seed keyword 3 ways: • Questions (how/what/why) → intent • Prepositions (for/with/vs) → commercial • Alphabet a–z → coverage 2 seeds → 486 unique keywords in 17 seconds. No signup to try.

The fix wasn't fighting the blocking — it was changing the data source. → Trends tools: SerpApi (free 100/mo key, no proxy) → Keyword discovery: Google's PUBLIC autocomplete endpoint — no key, no proxy, just works Pick sources that don't require an anti-bot war.

Stripe built Minions. Ramp built Inspect. Coinbase built Cloudbot. Async coding agents are becoming standard infrastructure at serious eng orgs. Not autocomplete. Agents that take a ticket and come back with a PR. LangChain open-sourced the pattern. It's called open-swe. A framework for building your company's internal coding agent. Comment @openswe on a Linear issue, mention the bot in a Slack thread, or tag it in a GitHub PR review. It spins up an isolated cloud sandbox with full shell access and starts working. It reads AGENTS.md from your repo root, runs linters and tests before committing, then opens a draft PR linked back to the originating ticket. Change your mind mid-run? Send a message. Middleware injects it before the agent's next model call. What stands out: • Runs parallel tasks, no queuing • Curated ~15 tools, not Stripe's ~500 • Sandboxes: Modal, Daytona, Runloop, or LangSmith • Built on LangGraph Deep Agents, you compose instead of forking The honest read: in-house coding agents took Stripe-sized teams to build. This is the same pattern with an upgrade path. Free and open source, MIT licensed, in Python. ⭐ 9,963 stars on GitHub. 🔗 GitHub link in the comments 👇

Dear Microsoft, when I hit the Windows Start menu key and start typing a word to autocomplete a search, I never, ever, EVER want it to return results of something not on my computer. Ever. Like, ever, ever, never.

شركة Anthropic أطلقت موديل جديد من عائلة Mythos اسمه Claude Fable 5 .. وقفلوه بعدها 😂 وده مش مجرد موديل “أذكى شوية” في كتابة الكود .. ده موديل معمول للمهام الطويلة والمعقدة: migrations كبيرة، تنفيذ features كاملة، مراجعة outputs بالـ vision، وكتابة tests يتأكد بيها من شغله. وده معناه إننا داخلين مرحلة جديدة في الـ Software Engineering. كل ما الـ AI بقى أقوى، قيمة المهندس اللي فاهم architecture، trade-offs، testing، security، product context، وmaintenance بتزيد أكتر. ويبقى السؤال الحقيقي هنا: هل إحنا بنتعلم نشتغل مع النوع ده من الموديلات كمهندسين… ولا لسه بنتعامل معاها كأنها autocomplete متطور؟ في العدد 101 من نشرة اقرأ-تِك الأسبوعية اتكلمنا عن Claude Fable 5، وليه إطلاقه مهم للمطورين، وإزاي ممكن يغير شكل شغلنا في الفترة الجاية. newsletter.eqraatech.com/p/v… هل شايفين إن الموديلات دي هتخلي شغل المطورين أسهل … ولا أصعب؟ شاركونا رأيكم 🚀

I rewrote my most-read Claude Code piece for 2026. When I first wrote it, Skills, native subagents, and hooks didn't exist. Now they're most of the point. The mistake people still make: treating Claude Code like a faster autocomplete and fussing over the prompt. The real payoff is the harness you build once and reuse on every task: - a tight CLAUDE.md (bigger is NOT better, it spends attention every turn) - a real permissions policy, not --dangerously-skip-permissions everywhere - skills for every workflow you'd otherwise re-explain - hooks for the non-negotiables the model can't be trusted to remember - subagents to fan out, inline reasoning to stay deep The prompt is the cheap part. The configuration is the moat, and it compounds. Full rewrite, plus the repos I keep open source (slopless, claudelicious): glyf.cc/claude-code-2026

I used the Chinese one. The context window is small. Just use for autocomplete

Composer 2.5 is the best coding model. Amazing alpha in keeping it in the sidebar and use it as a super fast autocomplete sidekick.

tired: cursor tab wired: NVIDIA CUDA Autocomplete

👉🏻 Amazon is regretting AI 🔥 youtube.com/watch?v=0vvVo0Um… AWS engineer asks Amazon’s Q AI to fix a bug… it deletes the entire production environment. 13hr outage. Then more AI meltdowns: 120k orders vanish, 6.3M orders lost in one day, site basically offline. They laid off 16k engineers, pushed AI hard, now need seniors to babysit the AI that replaced them. Mo Bitar (@atmoio) nails it: gave root access to autocomplete. Classic hype → chaos loop. Bring back the humans! youtube.com/watch?v=0vvVo0Um… #AI #Tech #Amazon

Moved from Cursor to @zeddotdev @claudeai a few months back and I'm not going back. After years on VS Code, Cursor felt like the right successor, until it got so laggy on macOS I was restarting it a few times a day. Waiting for your IDE to be responsive should be illegal. Just let me edit my code. Zed is fast enough that you forget waiting was ever a thing, and the Claude agent sits right next to the editor, so I kept the workflow I liked. I still miss Cursor's autocomplete. Nothing matches it yet. Small price for an editor that doesn't fight me. Highly recommend the combo.

Boris Cherny on betting that a coding agent could "do all of it" When Boris Cherny looked at the AI coding tools that already existed, he saw a category that was thinking too small. "If you look at the coding products, they were all pretty simple... complete the word, complete the sentence. That was the extent of AI in coding." So his team decided to go the other way: "We just wanted to make a way bigger bet. And our bet was we think actually a coding agent can do all of it." That bet has reshaped what his day actually looks like. He contrasts how engineering worked not long ago with how it works for him now: "A year and a half ago, you wrote the code by hand. And sometimes you press tab and it would autocomplete a line. Now I talk to my pod and it writes the code. And then while it does that, I talk to the next pod and it writes some code." Instead of working line by line, @bcherny describes orchestrating multiple agents at once. Anywhere from a handful to a few thousand running in parallel at any given moment. Asked how much code Claude is writing internally for engineers, his answer was blunt: "It's writing almost all of it on my team. So for me personally, it's been writing 100% of my code for at least 6 months." And he's clear about how that shift feels: "The work of engineering has just completely changed. I feel like I suddenly have superpowers. I have like a jetpack." Media: Bloomberg Originals