domain · engineering

Engineer for the system that actually runs

Architecture, AI codegen, evals, agents, infra. From engineers who shipped at scale.

47 cards · 32 operators · 29 tier-A claims · 5 synthesis patterns.

Strongest claims

1. Agent-first content has six platform layers, access, discovery, capability, format, token, UX bridge Addy Osmani
2. When the agent isn't doing what you want, fix the context, not the model Sherwin Wu
3. Build for the model six months out, the current model will eat your scaffolding Sherwin Wu
4. AI capability is not evenly distributed, it spikes where labs have data, rewards, and verification loops Andrej Karpathy
5. Continuous Calibration, Continuous Development (CCCD) is the operating loop for AI products Aishwarya Naresh Reganti
6. Code and media are the only forms of leverage that don't require asking, labour and capital both come gated Naval Ravikant

Adjacent domains

• ai-native · 36 co-occurrences
• leadership · 13 co-occurrences
• product · 13 co-occurrences
• research · 6 co-occurrences
• growth · 2 co-occurrences
• founder-craft · 2 co-occurrences
• gtm · 2 co-occurrences
• founder-operator · 1 co-occurrences

Synthesis patterns in engineering

• Build for the next model, not the current one
• Context, not capability, is the bottleneck
• Evals are data analysis, single judge, binary rubrics, error analysis first
• Principal/staff IC as force-multiplier archetype
• Verification, not execution, is the irreplaceable human job

47 insights in engineering

• 10-80-10: human direction, AI execution, human polish · Arvid Kahl
• An automation that works 95% of the time is not an automation · Cat Wu
• Advisor-tool replaces ensemble-of-3 stability hacks at near-Sonnet rates · Cat Wu
• Agent-first content has six platform layers, access, discovery, capability, format, token, UX bridge · Addy Osmani
• At agent scale, security governance must shift from humans approving humans to automated agent-identity and policy systems · Jonathan Jaffe
• AI adoption creates an asymmetric information problem where wins are public and costs are private, requiring deliberate feedback loop design · Charity Majors
• AI has crossed the threshold to something indistinguishable from judgment and taste, winners will know what to build, not how · Matt Shumer
• Above a certain level, every problem is a people problem · Silvia Botros
• Building production ML systems at scale · Anand Karunan
• Security scanning via data flow tracing finds business logic flaws that pattern-matching tools miss for years · Anthropic
• When the agent isn't doing what you want, fix the context, not the model · Sherwin Wu
• Build for the model six months out, the current model will eat your scaffolding · Sherwin Wu
• AI capability is not evenly distributed, it spikes where labs have data, rewards, and verification loops · Andrej Karpathy
• Continuous Calibration, Continuous Development (CCCD) is the operating loop for AI products · Aishwarya Naresh Reganti
• Code and media are the only forms of leverage that don't require asking, labour and capital both come gated · Naval Ravikant
• The dark factory: nobody reads the code, gated by a simulated QA swarm · Simon Willison
• Test automation that can't adapt to product changes creates a maintenance burden worse than manual testing. · Dileep Krishna
• Give the model tools and a goal; do not hard-code the workflow · Boris Cherny
• Evals are systematic data analysis on your LLM application, start with error analysis, not tests · Hamel Husain
• Glue work is technical leadership, not a tax on the IC · Tanya Reilly
• n8n + MCP + Claude lets GTM teams build workflows in plain English. · Ilya Azovtsev
• Enterprise AI buyers are shifting from benchmark rankings to intelligence per dollar as budget pressure reshapes model selection · Tomasz Tunguz
• You can outsource thinking, but not understanding, verification is the new human job · Andrej Karpathy
• Cheap external model for grunt work; Claude only sees judgment · Kesava Mandiga
• Build LLM-as-judge as binary true/false, one judge per pesky failure mode, and validate against human labels · Hamel Husain
• An LLM should maintain a wiki, not re-derive knowledge per query · Andrej Karpathy
• The middle is hollowing out, execution gets automated, leaving spec-writing and verification as the high-value human tasks · Eugene Yan
• Close the feedback loop by mining session transcripts for patterns to promote into config · Eugene Yan
• November 2025 was the qualitative threshold, coding agents now almost always do what you tell them · Simon Willison
• Sample 100+ traces, write one free-form note per trace, let an LLM cluster the notes, humans first, machines second · Hamel Husain
• Hoard a personal repository of things that worked, coding agents will recombine them · Simon Willison
• The planning fallacy guarantees every launch timeline is optimistic, the fix is the outside view · Daniel Kahneman
• A principal IC is a force multiplier, not a more-senior senior · Silvia Botros
• Encode jargon shorthand once, save tokens forever · Simon Willison
• When a new model lands, re-read the system prompt and remove crutches · Cat Wu
• Simple agents reading rich, specific context outperform complex agents reading thin context · Maja Voje
• Don't ask "how long will this take?", ask "how much time do we want to spend on this?" · Jason Fried
• Frontier models can write procedural SKILL.md files that cheaper local models execute, decoupling reasoning cost from inference cost · Tomasz Tunguz
• Prompts are code, Skills deserve testing, documentation, dependency mapping, performance profiling · Nate (Substack)
• We are in the transition from Software 2.0 to Software 3.0, AI Engineers will build the majority of new applications · swyx
• Staff engineering has four archetypes, two of them rotate across teams by design · Will Larson
• Iatrogenics, when the intervention causes more harm than the disease, most "fixes" in complex systems are net-negative · Nassim Taleb
• Agents come in three classes, tag each loop or under-resource it · Hamza Farooq
• A trace alone teaches nothing; learning requires feedback attached to the trace · Harrison Chase
• Underfund teams deliberately so AI substrate, not headcount, absorbs the work · Boris Cherny
• Use new tools as new tools, not as old tools, be ambitious and retry from scratch · Benjamin Mann
• Running MicroPython inside a WebAssembly sandbox gives AI agents safe code execution without host system access risk · Simon Willison