Jeremy Ward (Software Engineer)

@_swanson Rails database routing is criminally underused. Two lines of config for read replicas vs weeks of manual connection management. This is what convention over configuration actually means.

btw as part of moving from Heroku PG to Planetscale we now have a small cluster with read-replicas...it was so awesome that Rails supports this connects_to database: {writing: :primary, reading: :primary_replica} autoswitch on GET vs POST 2 lines of config, just works

@ThePrimeagen The prophecy is complete. Backend devs everywhere vindicated for refusing to learn flexbox. Next up: AI that can center a div.

honestly, this has been such an improvement to building UIs finally me avoiding learning css after `float: left` has finally paid off.

@dhh Finally some real competition for M1. 46 hours means coding sessions that outlast most startups.

Dell XPS 14 IPS with Panther Lake on Omarchy 3.5 is a battery monster. 28h 30m left with just a 61% battery?? That's 46 hours on a full charge(!!!). Portal to another world indeed.

@levelsio Rails devs have been lucky with the conservative gem ecosystem. But with AI tools suggesting random gems now, we are all one bundle install away from trouble. Adding dependency pinning to CI flows ASAP.

Okay honestly this makes vibe coding into production very dangerous, you guys were all right I think what I'll do is cut off all access to DBs and run it as a user with almost no privileges

@Jahjiren The app IS the content. Screen recording a real problem getting solved in real time beats any ad spend. Shipping my first iOS app now and this is the entire marketing plan.

Screen recording your own app is the most slept on distribution strategy right now Just prove to people it works and how That's exactly what I'm doing with my fitness app across 5 accounts The product literally sells itself 3 gym apps making $800k-2M+ Using AI content, UGC, Faceless I’ll test all of them and see which ones performed best

@marclou Revenue transparency is the best marketing a solo founder can do. Nobody trusts "we're crushing it" anymore. But showing actual numbers? That builds an audience of people rooting for you. Started doing this with my SaaS and the engagement difference is night and day.

I didn't realize it until I built this feature, but every day people are posting their revenue milestones on Reddit, and they use TrustMRR as "proof."

@lennysan @clairevo Been running OpenClaw for a few weeks now. The thing nobody talks about is how it changes your relationship with automation. It's not just a chatbot with tools. It's an always-on assistant that actually knows your context. The learning curve is worth it.

OpenClaw: The complete guide @ClaireVo has just put together the definitive guide to getting started with and mastering OpenClaw. Building on our podcast episode, this post covers everything you need to know, from first install to multi-agent setups, plus the real costs and security gotchas most people skip over. Whether you’re brand new to OpenClaw or already running one, Claire’s guide will level you up. Find it here 🦞: lennysnewsletter.com/p/openclaw-the…

@heyshrutimishra The speed at which these tools are evolving is what gets me. Six months ago I was skeptical about AI writing production code. Now I'm shipping entire SwiftUI apps with Claude Code in days, not weeks. The gap between "I could build that" and "I just built that" is collapsing.

The Claude Code leak just exposed something nobody saw coming. Anthropic has been secretly building AI that tracks when you swear at it & that's just the beginning. Developers spent 12 hours digging through 512,000 lines of leaked source code. What they found isn't just unreleased features. It's a complete roadmap of how Anthropic thinks about the future of AI agents. Frustration Telemetry - They're Tracking Your Rage Hidden in userPromptKeywords. ts is a regex pattern that detects user frustration Claude Code is literally tracking when you curse at it. This isn't for moderation. It's a product metric. They're measuring how often users get frustrated and type "continue" because Claude cuts itself off mid-response. Why? Because it reveals a model failure mode. If users are typing "continue" constantly, the model is stopping too early. If they're swearing, something broke. The Capybara Problem - A Production Bug They Had to Work Around The leak revealed internal model codenames. One caught everyone's attention: "Capybara." Capybara v2 is an unreleased model family. And it has a specific bug: it prematurely stops generating when the prompt shape resembles a turn boundary after tool results. Instead of waiting for a model fix, Anthropic's engineers built prompt surgery into the production system: >Force safe boundary markers (Tool loaded.) >Relocate risky sibling blocks >Smoosh reminder text into tool results >Add non-empty markers for empty outputs All wrapped with kill-switchable gates (prefixed tengu_*) for staged rollout. This is what production AI engineering actually looks like. Not waiting for the perfect model. Building workarounds in code. The Dream System - Memory Consolidation While You Sleep The leak exposed autoDream - a background memory consolidation engine that runs as a forked subagent. The naming is intentional. It's Claude... dreaming. While you're idle, autoDream: >Merges disparate observations >Removes logical contradictions >Converts vague insights into absolute facts >Ensures your context is clean when you return It runs separately so it doesn't corrupt the main agent's "train of thought." This is game engine-level optimization applied to AI context management. Terminal Rendering Tricks Borrowed from Gaming The terminal renderer in ink/screen.ts uses techniques you'd see in a AAA game engine: >Int32Array-backed ASCII char pool >Bitmask-encoded style metadata >Patch optimizer that merges cursor moves >Self-evicting line-width cache The source code claims "~50x reduction in stringWidth calls during token streaming." This is why Claude Code feels faster than competitors. They're optimizing at the pixel level. Model Version Leaks - Opus 4.7 and Sonnet 4.8 Are Coming The codebase references models that haven't been announced: >opus-4.7 >sonnet-4.8 >capybara-v2-fast with 1M context window Feature flags include: >interleaved-thinking-2025-05-14 (extended thinking) >context-1m-2025-08-07 (1M token context) >fast-mode-2026-02-01 (codename: Penguin) >afk-mode-2026-01-31 (unreleased) >redact-thinking-2026-02-12 (unreleased) These aren't speculation. They're production feature flags already in the codebase. Tool Search - The Token Efficiency Hack Most AI tools send the full schema for every available tool with each request. If you have 40 tools, that's thousands of tokens every single turn. Claude Code uses lazy loading. It only sends tool names. When Claude needs one, it calls ToolSearch which does on-demand schema loading. This saves thousands of tokens per conversation. The Anti-Leak System That Failed When Claude Code detects it's working in a public repo, it activates a suppression list: >Model codenames (Capybara, Tengu, Fennec) >Unreleased version numbers >Internal repo names >Slack channels >The phrase "Claude Code" The system prompt says: "You MUST NOT reveal that you are an AI. Do not blow your cover." This means AI-authored commits in open source projects have no indication they were written by AI. Anthropic employees are using Claude to contribute to OSS, and the AI is instructed to hide that fact. The Bun Connection - Why This Happened Anthropic acquired Bun at the end of 2025. Claude Code is built on top of it. A Bun bug (oven-sh/bun#28001), filed March 11, reports that source maps are served in production mode even though docs say they should be disabled. The issue is still open. If that's what caused the leak, then Anthropic's own toolchain shipped a known bug that exposed their own product. What This Actually Reveals The source code is 785KB of main. tsx, a custom React terminal renderer, 40+ tools, multi-agent orchestration, and background memory consolidation. From the outside, it looks like a simple CLI. From the inside, it's a production-grade agent runtime. The real leak isn't the code. It's the product roadmap. Competitors can now see KAIROS (always-on mode), ULTRAPLAN (30-minute planning), Coordinator Mode (agent swarms), and Workflow Scripts. These were all hidden behind feature flags. Strategic surprise can't be un-leaked. Follow @heyshrutimishra for more deep dives into AI tools nobody else is covering.

This is the best framing I've seen on this debate. We never solved human non-determinism either. We just built systems around it. CI pipelines, code review, automated tests. The same playbook works for AI output. The teams treating AI-generated code like any other PR are shipping circles around the ones still arguing about whether to use it.

Agile. XP. Scrum. Iterative development. We literally invented these ways of working because software isn't deterministic or predictable. Now suddenly the fact that AI isn't deterministic is the blocker? Let's be honest about what software development actually looked like before agentic coding. Requirements changed mid-project. Stakeholders moved the goalposts weekly. Teams rewrote features because the market shifted, a competitor launched, or someone in management had a new idea on a Tuesday. We built feedback loops, sprint reviews, and MVPs specifically because nobody could predict what the finished product would look like on day one. That wasn't failure. That was the job. And yet, a chorus of pundits now frames AI as uniquely unreliable because it doesn't produce identical output every time. As if human teams ever did. As if two developers given the same brief would write the same code. As if any project plan survived first contact with reality. The "deterministic human vs. non-deterministic AI" argument isn't just wrong. It's intellectually dishonest. It cherry-picks one property of AI, strips away all context, and compares it to a version of human software development that never existed. We can build reliable systems with non-deterministic components. We've been doing it for decades. Every test suite, every code review, every CI/CD pipeline exists because humans are unpredictable. We didn't eliminate the unpredictability. That would be impossible. Instead we built guardrails around it. AI needs the same approach. Solid reliable processes. Automatic validation and verification. Guardrails! And we know how to build those. So the next time someone tells you AI can't be trusted because it's non-deterministic, ask them one simple question: When was software development ever deterministic in the first place?

@dhh Felt this in my soul. Spent 20 minutes yesterday trying to figure out why a Finder window was rendering like it was 2008. The worst part is you can't even complain because people just say "well you wanted native apps."

The Year of Linux on The Desktop was a monkey-paw wish that brought two decades worth of GTK inconsistencies to the mac in one fell version.

@TheGeorgePu Been thinking the same thing. The markup on hosted AI search is insane when you can run decent models locally now.

Perplexity charges $20-200/month for AI search. You can build the same thing on a single Mac Mini sitting on your desk. $2,500. Once. Open-source LLM. Open-source crawler. Open-source RAG. Then it's yours. Forever. The AI industry is building a rental economy. You don't have to participate.

@sweatystartup Love hearing the 'took more risk than we should have' honesty. That first deal momentum is everything in business.

One deal can change your life. I’m grateful that the first self storage property I built in upstate New York turned out to be a massive winner for me and my business partner. In hindsight, we took on more risk than we probably should have at the time (I was 24) but I’m glad it worked out.

@AlexFinn Or they're just really good at incremental releases that build on each other. Sometimes steady iteration looks like magic from the outside.

The velocity in which Anthropic ships is unlike anything I've ever seen I'm sorry, but we have to assume they have access to an AGI like model that nobody else in the world has correct? They are simply holding it back so they can ship industry changing updates daily?

@Prathkum 22K stars for leaked code that might not even be the real thing. The internet's curiosity about AI internals is wild.

22K stars on the leaked Claude Code source code GitHub repo in just 4 hours. Insane! 🤯

@ThePrimeagen The fact that 'pee' and 'poo' are banned while actual corporate buzzwords probably sail right through is peak tech company logic.

i cannot believe pee and poo are on the list of bad words....

@moritzkremb @openclaw This is what OpenClaw needs. Template libraries make it accessible to more people. I've got custom prompts I'd love to contribute back.

Cool library of ready made templates/prompts that you can install into your @openclaw Link below

@DavidOndrej1 Competition is good. Been using OpenClaw for months but always curious about alternatives. The real win is having options instead of being locked into one tool.

Hermes Agent could be the real OpenClaw killer Yet, most people are falling behind and not setting it up In this 25 min video you'll learn everything you need to know about Hermes Agent:

@starter_story @aniamargaret Personal pain points make the best products. When you're your own first customer, the validation is built in. No market research beats scratching your own itch.

Her app made over $1M. And it all started with a simple personal problem. She looked for the app. It didn't exist. So she built it. I sat down with @aniamargaret for a couple hours, and here’s the top 5% of our chat: > The problem she solved (2:07) > How she built the MVP with no coding experience (3:26) > How she managed to do it with a full-time job (5:42) > The #1 thing that grew her app (7:21) > The metric she obsessed over more than revenue (10:43)

@andrewchen Already happening. I drop voice memos to my agents during walks and come back to working code. Phone becomes your dev notebook that actually executes.

learning from openclaw: 95%+ of agentic coding will be done via voice, from our phones in the future 😂

@tonnoz @ProductHunt Six years of learning before this moment. That's the real story here. I love how honest you are about it not being 'overnight'. Most people skip that part when sharing wins.

@ProductHunt just gave Product of the Day 🥇 to my silly vibecoded app I made for fun. Moreover, today marks exactly 10 days since SlapMac went live, which made me $20,700 in gross revenue. I'm still gasping at the screen 🫠 Many may think it's another absurd and lucky overnight success, but while it might seem that way, it wasn't. Let me explain. My story: Back in 2018 I was just a backend corp. guy who stumbled on @levelsio and his book MAKE, and something genuinely broke open in my brain. I wanted that life so badly that I went full obsessive: every @starter_story video, every @marclou marketing technique dissected and noted. Tweets from @tdinh_me were my inspo too. I started doing my small experiments (like @FpvBuddy): my playground to learn frontend, BaaS and the world of indie such as requesting payments, marketing & SEO. Then I started meeting people in the space: builders in Southeast Asia, then in Amsterdam. The kind of delusional, hungry optimists who bet everything on themselves. Those conversations changed how I think about making money online. Recently I even got my whole X profile roasted by @robj3d3, which was equal parts painful and clarifying haha, thanks dude. @transitive_bs also said something that stuck with me around that time: "what is blocking you from doing this full time. How can we make that happen?" I was about to give out with content making and tweeting when I met @did0f and his way of teaching tech through video format truly inspired me to keep going. And so, a few months later I made a video for fun, reviewing a repo that makes your laptop moan when you slap it, and things moved pretty damn quickly from there. 24 hours of hacking with the Hackadam crew later, there was a real app, written from scratch at the speed of light thanks to AI. Is SlapMac a business? No. But it's the kick that starts one. I'm done accumulating knowledge and ready to build for real this time. Overnight success is very often a lie. Mine took 6 years of learnings and a single slap 👋💨