--- title: AI engineering is not a vibe, it's a skill tree | daily.dev description: A skill map from Andrew Ng highlighting what effective AI-assisted coding actually requires - context window management, task decomposition, output evaluation,... canonical: https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI engineering is not a vibe, it's a skill tree | daily.dev og:description: A skill map from Andrew Ng highlighting what effective AI-assisted coding actually requires - context window management, task decomposition, output evaluation,... og:url: https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q og:image: https://api.daily.dev/og/posts/JO1bZx86q.png og:image:alt: AI engineering is not a vibe, it's a skill tree og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI engineering is not a vibe, it's a skill tree **[Trends](https://daily.dev/sources/trends)** · 3 min read · 1 upvotes · 0 comments ## Summary A skill map from Andrew Ng highlighting what effective AI-assisted coding actually requires - context window management, task decomposition, output evaluation, debugging agent loops - is being used to push back on the idea that working with AI coding agents is just about learning to prompt. Santiago Valdarrama argues the craft has become more nuanced, not less, while Ethan Mollick contends domain expertise remains critical since it lets developers judge AI output quality and spot its limits. The emerging consensus: AI adds a layer on top of existing engineering skill rather than replacing it, so experienced developers retain an advantage over those who assume the learning curve is flat. ## Content GitHub's data is stark: agent-generated PRs have grown 9x in eight months, with no sign of slowing. The dev community is now reckoning with what that actually means, and the picture isn't pretty. A team lead writing about "broken windows theory" describes going from 100% PR review coverage to 2% in five months, while output rose 8x. One experienced engineer on the team flagged the real danger: a single mediocre agent-written implementation was being copied as a "golden standard" by other agents, compounding bugs across the codebase. Quality erosion that used to take years now takes months. GitClear's analysis of 623 million code changes puts numbers on the vibe: heavy AI tool users gained 25% velocity over their own prior output (not the promised 10x), while code duplication rose 81% and refactoring dropped from 21% to 3.8% of changed lines. Stack Overflow's 2025 survey found 84% of developers use AI tools, but 46% distrust the results and 45% report *more* debugging time. Veracode found AI code hits 95%+ syntax correctness but only ~55% passes security testing. The failure modes are specific and recurring: hardcoded secrets, swallowed exceptions, hallucinated packages, N+1 queries, blocking calls inside async functions. Real Python published a five-step review workflow specifically for AI-generated code because the standard review process wasn't catching these. The discourse is splitting along a predictable fault line. @copyconstruct's take is getting widely shared: "coding agents can often *amplify* weaknesses, both individual and organizational" — what she calls the "empty calories" problem, where hyperproductivity driven by your team's lowest common denominator just makes the bad stuff faster. @housecor put it bluntly: "If I don't know what 'right' looks like, I can't tell when AI does it wrong." On the other side, @charliermarsh pushes back: "Respectfully (to humans), I think we really over-estimate the quality of human code." The maintainer burnout angle is getting less attention but might be the most concrete harm. NLnet Labs (12 developers maintaining critical DNS infrastructure) published an LLM policy banning AI-generated code contributions after AI-assisted PRs arrived as large, plausible-looking code dumps nobody actually understood. Linus Torvalds has flagged the same problem in the Linux kernel: a flood of AI-generated bug fixes and reports overwhelming maintainers. The emerging consensus from teams that are handling this well: keep some code review (the "broken windows" piece suggests ~5%), invest heavily in reviewing agent *plans* before they write code, and give agents strong guardrails — linters, type systems, CI checks — rather than trusting their output. @rauchg's framing: "Agents are only as good as the proof-checkers, compilers, type systems, and linters you give them." The teams doing best with AI tools, per GitClear, are the ones that already had strong technical discipline before AI arrived. Which is either reassuring or deeply inconvenient, depending on where your team started. ## Questions this post answers ### Does using AI coding tools help junior developers close the skill gap with senior developers? Not necessarily. Practitioners like Ethan Mollick and Cory House argue AI rewards existing expertise because judging output quality and knowing what changes to request requires already knowing what correct code looks like. Non-experts tend to get stuck with default outputs and cannot reliably catch subtly wrong architecture, plausible-looking bugs, or security holes the model introduces. _For teams weighing how to upskill juniors on AI-assisted workflows, daily.dev surfaces ongoing discussion on where AI coding help actually pays off._ ### What skills does Andrew Ng's skill map for coding agents cover? It covers context management, agent orchestration, evaluation loops, and knowing when to trust versus verify AI output. Santiago Valdarrama cites this map to argue that becoming an effective AI engineer is not a quick weekend project, pushing back against claims that it takes almost no time to learn. _Developers mapping out their own AI-agent learning path can follow this kind of framework discussion on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 3 comments across x (as of 2026-09-17). **TL;DR:** There's very little substantive discussion available; the few replies present are brief and mostly speculative about review automation, with no real debate captured. **Sentiment:** 33% positive · 34% mixed · 33% skeptical **The case for** - Agents handling review prep could save teams significant time. **By community** - x (mixed): A handful of short, low-engagement replies speculate that AI could speed up code review, with one comment questioning if the post itself was AI-written. **Highlights** > @augmentcode This could make code review much faster while keeping humans in control. > — [razaaitech on x](https://x.com/razaaitech/status/2100368369010532856) > @augmentcode Agents handling the review prep could save teams serious time > — [EviaAiTech on x](https://x.com/EviaAiTech/status/2100370640570405118) > @augmentcode This post sounds like Opus > — [alexisgrrd on x](https://x.com/alexisgrrd/status/2100309107739643980) **Source threads** - [x](https://x.com/steipete/status/2100282363087560924) · 0 points · 0 comments - [x](https://x.com/augmentcode/status/2100302138387943617) · 0 points · 3 comments ## Similar posts on daily.dev - [AI-coding agents kill team collaboration](https://daily.dev/posts/ai-coding-agents-kill-team-collaboration-ovin9khjp) · LeadDev · 2 upvotes · 0 comments - [The 80% Problem in Agentic Coding](https://daily.dev/posts/the-80-problem-in-agentic-coding-9nacygamm) · Elevate · 120 upvotes · 7 comments - [AI-authored code has nearly doubled, but so has PR size](https://daily.dev/posts/ai-authored-code-has-nearly-doubled-but-so-has-pr-size-mxu7xpjjr) · Software Engineering Research · 3 upvotes · 1 comments - [AI Code Review Bottleneck 2026: Why PRs Are Stuck in Review](https://daily.dev/posts/ai-code-review-bottleneck-2026-why-prs-are-stuck-in-review-o7qcycgq0) · Alex CloudStar · 2 upvotes · 1 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#prompt-engineering](https://daily.dev/tags/prompt-engineering), [#ai-engineering](https://daily.dev/tags/ai-engineering) [View this post on daily.dev](https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AI engineering is not a vibe, it's a skill tree","url":"https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q"},"datePublished":"2026-09-11T17:14:11.074Z","dateModified":"2026-09-17T22:40:03.849Z","description":"A skill map from Andrew Ng highlighting what effective AI-assisted coding actually requires - context window management, task decomposition, output evaluation,...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,prompt-engineering,ai-engineering","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"AI engineering is not a vibe, it's a skill tree"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-engineering-is-not-a-vibe-it-s-a-skill-tree-jo1bzx86q#faq","mainEntity":[{"@type":"Question","name":"Does using AI coding tools help junior developers close the skill gap with senior developers?","acceptedAnswer":{"@type":"Answer","text":"Not necessarily. Practitioners like Ethan Mollick and Cory House argue AI rewards existing expertise because judging output quality and knowing what changes to request requires already knowing what correct code looks like. Non-experts tend to get stuck with default outputs and cannot reliably catch subtly wrong architecture, plausible-looking bugs, or security holes the model introduces. For teams weighing how to upskill juniors on AI-assisted workflows, daily.dev surfaces ongoing discussion on where AI coding help actually pays off."}},{"@type":"Question","name":"What skills does Andrew Ng's skill map for coding agents cover?","acceptedAnswer":{"@type":"Answer","text":"It covers context management, agent orchestration, evaluation loops, and knowing when to trust versus verify AI output. Santiago Valdarrama cites this map to argue that becoming an effective AI engineer is not a quick weekend project, pushing back against claims that it takes almost no time to learn. Developers mapping out their own AI-agent learning path can follow this kind of framework discussion on daily.dev."}}]} ```