❌ Data breach
✔️ Surprise off-site backup

Baloo Uriza

What Actually Breaks at 30 People

• 🧠 Core claim — when engineering teams scale from 10 to 30 people, the first thing that breaks isn’t documentation or communication — it’s the leader acting as the hidden context router.
• 🔀 Invisible routing layer — small-team leaders often resolve ambiguity, remember decisions, connect people, and synchronize assumptions informally, making the org seem healthier than it really is.
• ⚠️ Scaling failure mode — once the team exceeds the leader’s cognitive bandwidth, questions stall, decisions become inconsistent, and delegated work keeps boomeranging back.
• 🧩 Four hidden functions — the leader was unconsciously handling decision history, assumption alignment, ambiguity absorption, and social cohesion.
• 📚 Documentation alone isn’t enough — static wikis decay quickly when no one owns them, creating a false sense that context has been captured.
• 🛠️ Better fix — externalize the hidden work into maintained systems: decision logs near the work, leads responsible for cross-team assumptions, and visible escalation of ambiguity.
• 🗣️ Name the shift explicitly — tell the team that earlier speed depended on one person holding context, and that the weight now needs to be shared.
• ❓ Best diagnostic question — ask: “What was I doing that nobody else knew I was doing?”

Why Are Experiences Of Vibe Coding So Polarised?

• 🎯 Polarized AI coding experiences — Same tools produce dramatically different results: some report 20x productivity gains while others encounter "AI slop" with architectural drift and technical debt
• 🧠 Intent-level operation matters — Unlike traditional tools that work at language/framework level, generative AI operates on intent, making architectural decisions without accounting for long-term impact
• 📋 Clear goals are essential — Author prioritizes two outcomes: negligible operational debt (ongoing risk/unplanned work) and highly malleable clean code with minimal complexity
• 🔬 Controlled experiment design — Used URL shortener service as test case with three approaches: true "vibe coding," guided development with templates/rules, and unattended generation
• ❌ Pure vibe coding unreliable — "Forget the code exists" approach proved inconsistent, with Claude ignoring instructions and drifting from intended structure despite explicit constraints
• ✅ Guided approach delivers gains — Using technical prompts, implementation notes, templates, and manual review produced clean system in ~1 hour with estimated 8-16x improvement over manual coding
• ⚠️ Unattended generates debt — Faster completion but with substantially more code, significant technical debt, and design regressions when left without guidance
• 🏗️ Templates solve bootstrap problem — Custom GitHub templates with "Wiring.md" files allow AI to merge complex architectural layers that traditional automation struggles with
• 🎛️ Constraint architecture works — Templates capture structure, rules capture behavior, stories capture trade-offs — allowing experience sharing through artifacts rather than constant review
• 🎯 Sweet spot identified — AI most valuable when user has strong high-level judgment but lacks detailed implementation knowledge, like CI/CD pipelines with Docker

What I Learned From Nearly 1,000 Interviews at Amazon

• 🔍 Bar Raiser Experience — Amazon Bar Raisers have veto power over candidates and are specially trained to ensure every hire raises the company's average talent level
• 📊 Key Pattern Discovery — After conducting nearly 1,000 interviews across all levels (intern to Principal Engineer), the biggest pattern emerged: candidates who didn't get offers rarely failed due to lack of technical skills
• 🎭 Presentation Over Technical — Most rejections happened because of how candidates presented themselves, not their technical abilities — revealing a massive blind spot most candidates have about non-technical interview aspects
• 🃏 Technical Skills as "Ante" — Technical skills are just the entry requirement to get into the interview game, but they're not what actually wins the job offer
• 💡 Where Decisions Are Made — The real hiring decisions happen in the non-technical realm, which most candidates completely overlook in their preparation
• 🎯 Blind Spot Problem — Candidates focus heavily on technical prep while ignoring the presentation and soft skills that actually determine hiring outcomes

Managing Up

• 🎯 "Managing Up" is often manipulative — The phrase describes people who selectively share information to create a specific narrative, rather than providing complete transparency to their manager.
• 🔍 Competent leaders verify information — Good managers will fact-check your reports with other sources, not because they don't trust you, but as healthy communication hygiene.
• ⚖️ Your boss isn't more important, just different — Managers handle broader responsibility (more people, products, scope) but that doesn't make their work inherently more valuable than yours.
• 📊 Three critical areas need constant vigilance — Projects (what you build), People (team dynamics), and Politics (organizational communication patterns).
• 🚨 Always report unexpected developments immediately — Strange, unfamiliar situations should be shared right away, even if they seem non-threatening, because your manager's context might provide clarity.
• 👥 Key people updates to share — Significant changes in key team members, major successes, progress on growth plans, and any mention of HR/Legal/People teams.
• 🎯 Critical project intel — Positive/negative developments on important projects, external gossip, achieved/missed milestones, and any "hint of impending doom."
• 🗣️ Political intelligence matters — External team developments, rumors about projects, and just plain weird behavior all provide valuable organizational context.
• ↔️ "Managing Sideways" is equally important — Your manager relationship only provides half the information you need — peer relationships contain the other crucial half that's often ignored.
• 🤝 Horizontal relationships are undervalued — Most people focus too much on managing up while missing critical context from colleagues at their level.

Say the Thing You Want

• 💭 The regret cycle — People often have career desires they want to discuss in 1:1s but stay silent due to fear, then rationalize not speaking up and promise to do it "next time" (but don't)
• 😰 Three core fears — Asking feels presumptuous ("who am I to ask?"), risky ("what if they think I'm not ready?"), and exposing ("someone will see me want something I can't have")
• 🚫 Invisible desires get nowhere — Thoughts kept to yourself have "no surface area" — nobody can react, build on them, or help you achieve them
• 🎯 Managers need direction — Your manager's job includes helping you grow, but they can't send opportunities your way if they don't know your goals — they have their own fires and multiple reports to manage
• 🗺️ Speaking up creates a roadmap — When you voice your goal, managers can provide specific feedback like "here's the gap I'm seeing" — the most useful guidance you can get
• 🔍 Self-assessment has blind spots — You need external perspective to see your strengths and weaknesses clearly, which only comes from putting your goals out there
• ⚡ Saying it makes it real — Voicing goals out loud gives them weight and changes your behavior — you start making different choices and having different conversations
• 🤷 The backfire risk is worth it — If saying "here's what I want for my career" gets you punished, that's valuable information about your workplace that's better to know now
• 💬 Simple phrases work — Try "I've been thinking about the path to senior. Can we talk about where I stand?" or "I'm interested in leading a project. What would I need to show you?"

That’s all for this week’s edition

I hope you liked it, and you’ve learned something — if you did, don’t forget to give a thumbs-up, add your thoughts as comments, and share this issue with your friends and network.

See you all next week 👋

Oh, and if someone forwarded this email to you, sign up if you found it useful 👇