Chris Rasco

Funny watching big companies dunk on “vibe coded startups” like they just exposed something. “You can’t build stable systems like that.” “Won’t scale.” “Not production-ready.” No shit. That’s not what it’s for. You’re evaluating it like it’s supposed to replace your 200-engineer stack. Meanwhile, a 3-person team is using it to: test ideas in days get real user feedback iterate 10x faster than your roadmap allows Of course it looks messy to you. Your entire worldview is: stability process risk mitigation quarterly planning Their world is: speed learning what outcome takes less resistance to success Vibe coding isn’t trying to compete with enterprise systems. It’s a velocity engine. A test bench. A way to punch way above your weight *before* you earn the right to build it “properly.” So when a big company trashes a vibe-coded product for not being stable… it just shows they don’t understand the game being played. You’re optimizing for perfection. They’re optimizing for truth. And truth shows up faster than your next sprint cycle.

7

User training is where most technology implementations fail. You buy great software. Your team gets 2 hours of training. Then they're expected to be productive immediately. Reality: they use 20% of the features and work around the rest. This isn't a training problem. It's a design problem. Software should be intuitive for people who actually do the work. Environmental health inspectors shouldn't need to become database experts. PH360's interface follows public health workflows, not software logic: → Inspection processes match real-world procedures → Forms are organized like paper versions teams know → Guidance is contextual and job-specific → Advanced features are available but not required Make technology adoption natural, not forced. Because the best software is the software your team actually wants to use. How long does it typically take your staff to become proficient with new systems? -- Hi! I'm Jefferson the CEO of F&T Labs. I'm a former CIO of a local health department and I've taken that experience and funneled into the tools that we're building at F&T Labs. I envision a future where America's public health systems have the tools that they need to do their work. (I'm also a proud husband, dad, and bow-tie aficionado)

Hard Truth: Most "real-time" systems aren't actually real-time Here’s a hard truth: 90% of systems claiming to be "real-time" are lying to themselves and their users. If your worst-case execution time (WCET) isn’t measured, documented, and guaranteed, you’re not building real-time systems, you’re building hope-and-prayer systems. We’ve all seen it, RTOS-based projects where developers claim "real-time" capability but have never actually measured interrupt latency or task jitter. They’ll proudly show you their 1ms task scheduling but can’t tell you what happens when a cache miss coincides with a DMA transfer. I’ve debugged "real-time" systems that missed deadlines by 100ms under heavy load because someone used malloc() in an ISR. I’ve seen medical devices that claimed 10ms response times but actually had 200ms spikes when garbage collection kicked in. And I’ve watched aerospace systems fail certification because their "real-time" claims were based on average-case timing, not worst-case. Real-time isn’t about speed, it’s about determinism. It means knowing exactly how long every operation will take in the worst possible scenario. It means measuring, not assuming. It means understanding your hardware’s timing characteristics at the cycle level. 🔥 What's the most ridiculous "real-time" claim you've ever seen? Share your WCET horror stories below, bonus points for systems that failed spectacularly in the field! #EmbeddedSystems #RealTime #RTOS #WCET #Firmware #EmbeddedC #Determinism #TechTruth #EmbeddedEngineering

250

71 Comments

"MCP was a mistake. bash is better." That's like saying "HTTP was a mistake. Google is better." I get what he's trying to say. And he's not entirely wrong: if you want an effective agent, don't stuff 200 tools into it. Doesn't matter if they're connected via MCP, JSON-RPC, or classic tool calling — too many tools make any agent dumb. But that has nothing to do with the protocol. MCP is a transport layer. Bash is a tool. They don't compete. And here's the thing: bash NEEDS one of these protocols to be connected to an agent in the first place. Classic tool calling, MCP, JSON-RPC — pick one. Bash doesn't replace any of them. It USES one of them. That's like saying "phones were a mistake, my voice is better." Sure — but you still need something to carry it. Now let's talk timing. Peter Steinberger posts this right as he joins OpenAI. The same OpenAI that fought MCP for months before eventually adopting it — and now sits on the MCP Steering Committee. So he joins a company that actively uses MCP… and declares MCP a mistake? Did he not get the memo? In Langertha, my AI framework in Perl, I can bind an MCP server directly to classic tool calling. Or run it as JSON-RPC. Because the protocol is flexible enough. You just have to understand it before declaring it dead. The right take would have been: "Give your agents fewer tools and make them really good." Instead we got a hot take that confuses protocol with usage — and people are reposting it like gospel. Question the tool overload. Not the protocol. (Oh, and MCP::Run::Bash exists. Bash over MCP. In Perl. Turns out they get along just fine. Also by me)

16

27 Comments

I use paid versions of ChatGPT, Claude and Grok and am constantly switching back and forth between them. Here's when/how I use each one: Document Writing: Claude. I find Claude's MCP integrations are way better and it spends more time researching topics in Slack, email, etc. It even asks me clarifying questions before writing. The writing it produces is just better - more concise, less AI-ish, warmer, etc. I only use AI for internal writing - PR/FAQs, contracts, performance reviews, etc Slides: Claude. Not even close. The PowerPoint add-on is as good as it gets Spreadsheets: Claude. Gemini and ChatGPT straight up do not work. Claude's Excel add-on truly understands what I'm asking and correctly responds Research: Claude, but I use ChatGPT more often because of the better voice mode, better desktop client and speed. If I have to do deep competitive analysis, I'll pull up Claude because it is materially better Legal document review: ChatGPT. 5.2 Pro is nearly flawless, given a solid baseline and prompt. Custom GPTs are great, though it's annoying that I have to attach custom instructions due to prompt size limitations Shopping: ChatGPT. It has in-app carousels, images, and even some "buy it now" buttons. The ChatGPT client is much better than Claude, so I use it more often News / analysis / sentiment: Grok. It's natively plugged into X and as a result, its take on current events / news / trends is accurate within a few minutes. I can ask "What's going on in [whatever country the U.S. is currently bombing]?" or "What does VC Twitter think about [the fad of the day]?" Grok also happily discusses politically sensitive topics that the others won't even go near Code: Claude. My software development these days consists of writing scripts to enrich data. Claude Code is nearly flawless. I've tried it for far more advanced use cases and even inspected the code it produces. It's magnificent. ChatGPT is good for troubleshooting but that's about it Ok...now on to the non-functional: Clients: ChatGPT. The Claude desktop app is just a thin web wrapper and it (frustratingly) shows. ChatGPT has great native apps for iOS and Mac Voice mode: ChatGPT. The voice is so much more natural. I can interrupt and change guidance quickly without the conversation getting awkward (I wish humans worked that way haha). I just wish it was more concise. The voice is incredibly fluid Speed: ChatGPT. Claude regularly takes > 3x longer to respond Memory: None are great at actually remembering history. I'm sick of having to type "I'm a 40 year old male, my measurements are..." when asking for help finding clothes or whatever Bottom line is: - Grok is great for breaking news / sentiment - OpenAI is good enough for consumer use cases - better Google, shopping, etc - Claude is what I'm using day to day for business. Opus 4.6 is just so good! - I haven't really found a use for Gemini. I'm sure it's great, but I'm set between Grok, OpenAI and Claude

98

21 Comments

JUST IN: Claude Code hit $2.5B ARR in 12 months. That makes it the fastest growing software ever. Anthropic buried the Claude Code ARR news in its announcement today of raising $30B at a $380B valuation, but I think it might just be the most important news of the day. Here's your guide to getting started: 1. Understand what you get It's not just for coding. It's an LLM with the best agentic harness: file system access, the ability to connect to MCPs like your analytics and docs tools, and autonomous planning. Karpathy called it "a little spirit/ghost that lives on your computer." 2. Setup in 5 minutes Install with npm install -g @anthropic-ai/claude-code, run "claude," log in with Claude Pro ($20/mo), navigate to your project. You can also download the native app if you prefer skipping the terminal install. 3. Start by asking questions, not giving orders Ask "what does this codebase do?" before requesting changes. Let Claude map the architecture first. Then try "find the bug in UserAuth.js" and watch it pinpoint the exact line and fix. 4. Update your CLAUDE.md This is your secret weapon. It gives Claude your project context, coding standards, common commands, team conventions, and architecture notes. Claude reads it first every session. Run /init to auto-generate one, then keep customizing it for your workflow. 5. Level up with advanced features - Hit Shift+Tab to make Claude think before coding. Review the approach, then execute. Use message queuing to stack multiple requests. - Use /clear to reset context without losing your CLAUDE.md. - Spin up subagents for specialized tasks like testing, security, and docs. 6. Master the use cases for you For PMs, these include: • Draft PRDs from templates • Parse CSVs for feature requests • Prep interview questions by company and level • Run competitive research across 5 pricing pages • Build working React prototypes from wireframes When you engineer the context properly, you'll find the output is 10x regular Claude or ChatGPT. Here's everything you need to go from 0 to 1: Claude Code basics: https://lnkd.in/eUyPEAma Claude Code advanced: https://lnkd.in/gvA6rZtX Claude Code operating sys: https://lnkd.in/gsH9RSfJ Claude Code workflows: https://lnkd.in/eadQ5fN6 Claude skills: https://lnkd.in/eZyJzMnH 💬 Comment "CLAUDE" + DM for my Claude.md. ➕ Follow Aakash Gupta for daily AI PM insights. Claude Code is the fastest growing product of all time. You're not too late to learn it.

199

97 Comments

More and more organizations are turning to internal LLMs, especially in data sensitive industries. The future of intelligence platforms is giving companies choice not locking them in… #llm #mindzie #openarchitecture

7

After years working in API-first universes, and now that we live in the era of MCPs, I can't get my head around the fact that availability / booking systems of hotels "only" work on GDSs (Amadeus, Sabre, etc.) or proprietary extranets (Four Seasons Preferred, MO Fan Club, etc.) and don't offer a kind of unified API layer. Or am I missing something?

9

Word on the street: Imprivata (the backbone of digital identity, access, and security across countless health systems) is reportedly being explored for a sale by Thoma Bravo. Potential valuation? 💰 Up to ~$7B Original purchase price? 😮 ~$544M That’s not a typo. For those who may not live in the weeds of healthcare IT, Imprivata isn’t a “nice-to-have.” It’s the invisible infrastructure clinicians rely on every single day: • Secure EHR access • Single sign-on • MFA • Identity governance • Endpoint security • Regulatory compliance Translation: If Imprivata sneezes, hospital workflows catch a cold. What makes this especially interesting is why now. Identity, security, and clinician access aren’t side projects anymore, they’re strategic assets. In a world racing toward AI-driven care, zero-trust security models, and interoperability mandates, the companies that sit between clinicians and data suddenly become incredibly valuable chess pieces. And yes… this feels familiar. We just watched speculation swirl around Oracle and Cerner. Now we’re seeing another core healthcare infrastructure platform potentially move hands. This time from private equity to… well, that’s the billion-dollar question. Strategic buyer? Another PE roll-up? Big Tech sniffing around healthcare identity? Whatever happens next, this isn’t just an M&A headline. It’s a signal. Healthcare IT is no longer just about growth, it’s about who controls access, trust, and workflow at scale. And that’s where the real power sits. More to come. #HealthcareIT #HealthTech #DigitalHealth #CyberSecurity #HealthcareLeadership #PrivateEquity #HealthSystemOperations #EHR #ClinicianExperience #HealthcareInnovation #MergersAndAcquisitions #Jobot #JobotHealthcare

7

Claude just dropped Skills. MCPs, sub-agents, slash commands... where does it all connect? Let’s break down how agent skills actually work — and what’s coming next. Plus a few surprises 👀 https://lnkd.in/epMu5RSe

5

Title: Your No‑Code Stack Is Taxing Your Margins. Here’s the Fix. If you’ve moved past POC volume, no‑code/iPaaS costs, latency, and fragility start eroding margins. Validation = no‑code. Scale = code‑first, event‑driven. What to measure (put these on a live dashboard) - Cost per automated task (all‑in: infra + LLM + vendors). - P95 latency end‑to‑end (conversion and CSAT sensitive). - Success rate and reprocess rate (timeouts, retries, DLQ). - Automation coverage/deflection rate (human bypass). - MTTR (minutes to fix incidents). Real pricing snapshot (October 2025) - No‑code: Zapier 5k tasks ≈ $139/mo; at higher volumes + multi‑step flows + polling + SaaS add‑ons, 100k tasks/month commonly lands ≈ $1.6k–$2.8k/mo. - Code‑first: Hetzner VPS ≈ $4–$15/mo; Supabase Pro ≈ $25–$50/mo; Upstash Redis ≈ $10–$25/mo; CloudAMQP RabbitMQ ≈ $0–$19+; Inngest free tier (50k exec) or ≈ $75/mo; Groq LLMs ≈ $0.05–$0.30 per 1M tokens with caching; typical 100k tasks ≈ $75–$250/mo. What changes in practice - Cost per task drops ~85–95% when moving from multi‑hop iPaaS to single‑hop queue + cache + API on VPS. - P95 latency improves from ~6–15s to ~0.8–2.5s. - Success rate rises toward 99.5%+ with retries, idempotency, DLQs; MTTR falls to 30–90 minutes with unified tracing. When to choose what - Use no‑code to validate in under 2 weeks, low volume (<20–30k tasks/month), or short‑lived projects. - Go code‑first for recurring workloads, growing volume, SLAs/compliance, and when unit economics matter. Minimal, cost‑efficient blueprint - API + workers on a low‑cost VPS; Supabase Postgres for auth/state; Upstash Redis for cache/ratelimits/idempotency; CloudAMQP RabbitMQ for durable jobs; Inngest for orchestration; Groq for fast, value‑for‑money LLMs with caching and small‑model default; Sentry + correlation IDs for observability. 30‑day proof plan - Weeks 1–2: Instrument current no‑code flow with the 5 KPIs; capture baselines. - Weeks 3–4: Ship code‑first mirror; shadow then ramp to 50% traffic. - Week 5+: Cut over, publish side‑by‑side metrics, show crossover point and annualized savings. #AI #AIAgent #Automation #AiAgency #nocode #codefirst #FullStackDevelopment #NicheAI #AIbusiness

3

Tech Titan Thursday: Jim Gray If you’ve ever run a database transaction, you’ve touched Jim Gray’s work. He was the architect of ideas we now take for granted: ACID transactions, recovery, and fault tolerance. His “Notes on Database Operating Systems” (1978) and the “Five-Minute Rule” (1987) are classics. But what inspired me most was his way of thinking: bridging deep theory with practical engineering. Gray was a prolific writer. His papers turned complex ideas into accessible knowledge, and that played a huge role in his influence. A good reminder for all of us: doing great work is step one, but sharing it with the world is how it changes the field. Reading Gray taught me key ideas about databases. Maybe the most important lesson: Writing to disk isn't enough. Reliable systems anticipate crashes, conflicts, and the messy reality of distributed systems. Durability isn’t a feature you tack on, its a way of thinking. And writing about your work makes it go farther and wider that you can imagine. #Databases #DistributedSystems

244

31 Comments

⚠️ If you’re about to build a web application or AI system… read this before you waste 2 months. Most teams don’t fail because of bad ideas. They fail because they start wrong. • Wrong architecture • Unscalable frontend and backend • Auth issues • Broken deployments • Missing real-world infrastructure By the time they fix it, they’ve already burned time, money, and momentum. After 6+ years as a Lead Full-Stack Developer, I built something to eliminate this from day one. A production-ready full-stack + AI system designed for scalable, real-world applications. Not a template Not a starter kit A complete foundation you can use for MVPs and scale into production. 🚀 What you get from day one: 🎯 Frontend ready for scale • Responsive setup • Scalable Sass architecture • Clean design system • Redux integrated 🔐 Authentication done right • Secure JWT auth • Role-based access • Google + email login • Security best practices ⚙️ Backend + infrastructure • Django + PostgreSQL • Redis caching • Celery background tasks • Fabric automation 💳 Business-ready features • Stripe integration • SendGrid email system • Pre-built product flows 🚀 DevOps & deployment • Dockerized setup • Nginx config • CI/CD pipeline • Deployment guide + scripts 🤖 AI-ready architecture • AI agent structure • OpenAI + Google Cloud ready 💡 This is not about starting fast. It’s about starting right and scaling without rebuilding later. What normally takes more than 2 months, you get instantly. That’s at least ~$20,000+ of senior engineering work. And I don’t just give you the system: • 5 hours hands-on training • Full documentation • Optional mentorship / lead support 💰 Package: $2,000 Built for: • Startups without strong tech teams •  Founders who want to save money on development in the first few months • Teams without a senior engineer yet • Freelancers/agencies who want to move 10x faster Instead of figuring everything out from scratch… Let’s do this properly. 👉 Book a call: https://lnkd.in/eBMDCNf8 Or DM me “SYSTEM” ⚡ You don’t need to reinvent the foundation. You just need the right one.

7

5 Comments

Billions will watch the #WorldCup ⚽ Millions will place bets. Behind every live odds update, bet validation and in-play experience is infrastructure that has to perform in milliseconds - even during massive traffic spikes. DraftKings Inc. uses Aerospike to support real-time betting workloads with ultra-low latency and high availability during peak sporting events like the Super Bowl 🏈 Learn more about how DraftKings achieves real-time betting with speed, safety and scale: https://lnkd.in/gN-pFTVS

4

Most CTOs don’t fail because of bad architecture. They fail because hiring can’t keep up with the roadmap. A CTO I recently spoke with was stuck: • Product features slipping • Core engineers overloaded • Local hiring taking 3–4 months • Agencies adding noise, not velocity Instead of hiring “more people,” he hired one remote dedicated developer. But here’s what he did differently 👇 1️⃣ Hired for ownership, not hours The developer worked full-time on one product. Daily standups. Real accountability. No context switching. 2️⃣ Tested in the real codebase Not abstract interviews. A real sprint task, real PRs, real collaboration. 3️⃣ Integrated like an internal hire Same tools. Same standards. Same expectations. Remote didn’t mean “external.” 4️⃣ Scaled only after value was proven Within 30 days: • Faster releases • Lower team stress • No compromise on code quality The insight most CTOs miss: Remote dedicated developers are not a cost hack. They are a velocity and focus multiplier—when hired correctly. If your roadmap is clear but your team is stretched, the problem may not be talent — it may be how you hire it. #CTO #EngineeringLeadership #RemoteTeams #StartupScaling #ProductDevelopment #TechHiring

11

2 Comments

Wix has acquired a Six-month-old, bootstrapped "vibe-coding" startup, Base44, for $80 million in cash. Base44 is part of a new crop of AI tools designed for non-programmers. It allows users to enter text prompts to build complete applications with database, storage, authentication, analytics, and integration, "helping everyone, technical or not, build software without coding at all." For Wix, the deal adds a proven, fast-growing, and profitable LLM vibe-coding product to its offerings. For Base44, the acquisition provides the resources to scale and meet the volume its rapid growth demands. The eight-person team will receive $25M as a retention bonus. Congratulations to founder Maor Shlomo and the teams at Base44 and Wix on this incredible outcome. TechCrunch: https://lnkd.in/dPygarpJ #Acquisition #MandA #VibeCoding #NoCode #AI #Bootstrap #SoloFounder #Wix #DeveloperTools

6

2 Comments