Adam Bender's Google I/O talk argues every developer ecosystem faces a 10x transformation from AI-assisted development. The thesis is unsettling: AI acts as an amplifier, not a fixer. Teams with strong fundamentals scale beautifully. Teams with weak fundamentals scale their dysfunction at equal speed. Here's what that means for how you should prepare.
I used to go to sleep tired, wake up tired, and stay tired through weekends and holidays. The problem wasn't sleep deprivation. It was misunderstanding what recovery actually requires. After mapping the research on psychological detachment, mastery experiences, and the four dimensions of recovery, the fix isn't doing less. It's doing different.
David Heinemeier Hansson now works agent-first. But his philosophy of craft didn't flip — the tools finally got good enough to preserve taste while dramatically widening what feels economically possible to build. After mapping his latest thinking against my own experience with Claude Code, the inflection point wasn't model quality alone. It was the shift from intrusive autocomplete to delegation-based harnesses.
Garry Tan claimed he ships 37,000 lines of AI-generated code per day. Scott Tolinski examined what happens when AI agents are let loose without proper oversight. The audit reveals what volume without review actually produces: digital landfill.
Andrej Karpathy argues that coding has shifted from typing code to orchestrating multiple agents. After mapping his latest thinking against my own experience with Claude Code and Cursor, the most valuable pattern is no longer a single helpful model, but autonomous systems that can keep working without a human in the loop. Here's what that means for engineers in 2026.
Anthropic's paper on AI and skill formation has been weaponised by both sides of the AI debate. After reading the study, watching Theo's breakdown, and cross-referencing Sean Goedecke's critique, the actual finding is narrower and more useful than the headlines suggest. Here's what it actually measures, and what it means for how you should use AI.
Strong software engineers in large tech companies need more ego than the industry usually admits. Not arrogance in the bully sense, but a durable self-belief that survives constant confusion, incomplete knowledge, and organisational friction. Sean Goedecke's essay on ego and humility is the most useful career framing I've read in years.
Code in production doesn't mean a project is shipped. After watching Sean Goedecke break down how big tech actually delivers, I've reframed shipping as an organisational outcome, not a technical milestone. Here's what that means for engineers who want to see their work actually land.
Engineers try AI tools 2-3 times, hit a wall, and walk away. They declare the tool useless and go back to manual work. After watching this pattern repeat across dozens of teams, I've realised the single biggest blocker to AI adoption isn't model quality or tool selection. It's the inability to close the feedback loop.
The highest-leverage move in agentic engineering isn't longer prompts or elaborate orchestration. It's treating your coding agent like a capable teammate you can interrupt, redirect, and ask for options. When the model is strong and the feedback loop is tight, conversation outperforms ceremony.
Being surrounded by colleagues content with the status quo doesn't have to limit your trajectory. Steve McConnell's career pyramid framework provides a way to navigate stagnant environments through deliberate self-development and strategic project selection.
Competitive moats in AI products now last weeks, not months. After mapping how Granola broke into the saturated AI note-taking market, the pattern is clear: sustainable advantage comes from how you assemble standard AI blocks with your unique data and workflows, not from the blocks themselves.
The real AI transformation isn't mass unemployment. It's job degradation — breaking complex skilled work into simpler, faster tasks performed at increased speed. After mapping how this pattern is playing out at major tech companies, the warning signs are clear. Engineers aren't being replaced. They're being turned into assembly-line workers.
The terminal is evolving from precise commands to conversational AI collaboration. After hands-on testing with Claude Code, OpenAI's Codex CLI, and Amazon Q Developer on real projects, I explore how these tools are transforming development workflows. From reviving legacy Laravel codebases to addressing GitHub issues, each tool brings distinct strengths and costs. The verdict? AI CLI tools excel at the "first 90%" but still need human oversight for production-ready code.
Success in enterprise environments isn't just about brilliant code or technical skills. It's about understanding power dynamics, building social capital, and knowing when to dive into problems others avoid. This framework covers strategic onboarding, relationship building, and sustainable performance.
After five years away from regular web development, I'm diving back in with this new digital garden. Here's why I built it, who it's for (hint—maybe no one), and what you might find here. A rambling journey through my return to the front-end wilderness.
Technical interviews filter for coding ability. They don't filter for the attributes that separate good engineers from great ones. After studying how high-performing teams actually select and develop talent, the pattern is clear: rally capacity, communication, and learning agility matter more than algorithmic speed. Here's what to look for instead.