# ng-ai-engineering-skills-map-2026-08-14

## Veille

X post by **Andrew Ng** from **August 14, 2026** (16:29 UTC), reprising the "Dear friends" letter from ***The Batch* #366** (DeepLearning.AI, same date), ~900 words. Ng presents **The AI Engineering Skills Map** and publishes **four skills** held to be the most important. **(1) Building and deploying AI applications** — the specificity is named: *« The key difference between AI and non-AI applications is that the former has unpredictable outputs »*, hence the emphasis on *evals* and error-analysis loops. **(2) Software engineering fundamentals**, because *« Understanding software fundamentals allows you to recognize what tradeoffs even exist »* — the inexperienced developer fails *« because they don't know what context to give their coding agent »*, hence the goal of *« steering coding agents using the precise language of software engineering »*. **(3) Using coding agents**, in an operational formulation: *« help the agent autonomously close loops by providing verifiers or evals »*, and *« knowing how much to intervene and how much to leave them alone »*. **(4) *Shaping the build***: *« Given a clear spec, coding agents are rapidly improving at delivering to it. Thus, our work as engineers is shifting toward deciding what should be in the spec »*, paired with *« Engineers should no longer expect to be given a pixel-perfect design and asked only to implement it. »* A **terminology note** carries most of the framing: Ng talks about **skills** in AI engineering and **not the role** "AI Engineer", with an explicit analogy — *« All developers today should know how to work with the cloud, and only a smaller number have a "Cloud engineer" title. »* The whole is backed by *« an analysis of more than 10,000 job postings, dozens of structured interviews with experts, hiring managers, and recruiters, surveys, and other online data »*, of which **no numeric results are published**: Ng describes his process as *« informally… akin to running clustering »* and announces a detailed map in future posts. He states the interest in the second-to-last sentence: *« DeepLearning.AI's principal focus is to help developers gain these AI engineering skills. »*

## Titre Article

The AI Engineering Skills Map

## Date

2026-08-14

## URL

https://x.com/AndrewYNg/status/2088302050706686198

## Keywords

AI Engineering Skills Map, skills map, Andrew Ng, DeepLearning.AI, The Batch #366, skills vs role, AI Engineer, skills baseline, requalification, reskilling, cloud analogy, full-stack, data engineer, DevOps, machine learning engineer, AI applications, unpredictable outputs, unpredictable outputs, LLM, context engineering, RAG, agentic workflows, evals, error analysis, software engineering fundamentals, tradeoffs, tradeoffs, cost, scalability, reliability, security, privacy, vibe coding, what context to give the agent, coding agents, agentic coding, when to intervene, verifiers, verifiers, autonomous loop closing, clear spec, shaping the build, product sense, business context, ownership, MVP, continuous learning, tool-trial routines, 10,000 job postings, structured interviews, informal clustering, detailed map to come, employability, upskilling, training provider, declared conflict of interest

## Authors

**Andrew Ng** — fondateur de **DeepLearning.AI**, general partner d'**AI Fund**, cofondateur de **Coursera** et de **Google Brain**, ancien chief scientist de Baidu. Texte signé, à la première personne, écrit *« with my team »* sans qu'aucun collaborateur soit nommé. Publié le **14 août 2026** sur X et dans ***The Batch* n°366** — même texte aux deux endroits ; préférer *The Batch* pour toute citation durable. Quatrième fiche Ng du corpus, après les lettres n°350 (24 avril), n°352 (8 mai) et n°359 (26 juin).

## Ton

**Profile**: a founder-pedagogue letter, magisterial and benevolent register — ~900 words, four blocks announced then developed, one synthesis sentence per block. Audience: developers **and** employers, addressed explicitly and separately from the first paragraph (*« (i) developers prioritize what to learn, and (ii) employers hire skilled developers »*). A study program presented as a state of play.

**Style**: the opening is a deliverable presentation (*« I am delighted to present »*) — the map is laid out before being justified, the empirical backing arriving in the next paragraph as a single four-item sentence. The framing is set against ambient noise, without naming anyone: *« But with the noisy, hype-filled, information environment around AI, what are the most valuable skills for you to learn? »* — the service rendered is curation, and the authority rests on the byline, with no competitor or contrary source cited. The **terminology note** (*« A note on terminology »*) is presented as a vocabulary clarification when it in fact carries the post's main framing. There is a single concession in the whole text, and it is the most nuanced sentence: *« knowing how to work with a clear spec **(and when not to bother doing so)** »*. The prose is clean but visibly written fast — *« avoid pitfalls like risk an agent messing up your production database »* is ungrammatical in the original, and the methodology account fits in a single sentence with four semicolons. It is a weekly letter, not a proofread report.

**Marker phrases**:
- ***« I talk about AI Engineering skills rather than the "AI Engineer" role »***
- ***« All developers today should know how to work with the cloud, and only a smaller number have a "Cloud engineer" title. »***
- ***« The key difference between AI and non-AI applications is that the former has unpredictable outputs »***
- ***« knowing how to drive disciplined evals and error analysis loops »***
- ***« Understanding software fundamentals allows you to recognize what tradeoffs even exist »***
- ***« they don't know what context to give their coding agent »***
- ***« steering coding agents using the precise language of software engineering »***
- ***« help the agent autonomously close loops by providing verifiers or evals »***
- ***« knowing how to work with a clear spec (and when not to bother doing so) »***
- ***« our work as engineers is shifting toward deciding what should be in the spec »***
- ***« Engineers should no longer expect to be given a pixel-perfect design and asked only to implement it »***
- ***« Underlying all these skills is a mindset of continuous learning »***

**Epistemic stance**: an authority synthesis backed by unpublished data. The four categories are argued and a practitioner will recognize them, but nothing in the text lets one distinguish what comes from the 10,000 postings from what comes from the author's experience. Cite the four skills as **Andrew Ng's thesis**; do not cite them as a **study result**. The author states himself that DeepLearning.AI teaches these skills.

## Pense-betes

- **Date / source**: **August 14, 2026**, X post by **Andrew Ng** and "Dear friends" letter from ***The Batch* #366** (DeepLearning.AI) — identical text. ~900 words.
- **Key framing**: the post talks about **skills**, not the **role** of AI Engineer. *« All developers today should know how to work with the cloud, and only a smaller number have a "Cloud engineer" title. Similarly, all developers — full-stack engineers, data engineers, DevOps engineers, machine learning engineers, and, yes, AI engineers — will need AI engineering skills. »* ### The four skills | # | Skill | Key phrasing from the post | |---|-----------|---------------------------| | 1 | Building and deploying AI applications | *« the former has unpredictable outputs »* → evals and error analysis | | 2 | Software engineering fundamentals | *« allows you to recognize what tradeoffs even exist »* | | 3 | Using coding agents | *« help the agent autonomously close loops by providing verifiers or evals »* | | 4 | *Shaping the build* | *« deciding what should be in the spec »* | A fifth, unnumbered skill: *« Underlying all these skills is a mindset of continuous learning »*, whose operational version is more precise than the slogan — *« having routines to keep trying new tools and evolve your workflows as best practices change »*. The useful word is **routines**: a planned practice, to be budgeted as time. ### What the cloud analogy implies depending on the reader | Reader | Implication | |---------|-------------| | Developer | the skill adds onto the existing job, it does not replace it | | HR / manager | the realistic path is **requalifying the existing workforce**, not hiring "AI engineers" | | Training provider | the addressable market shifts from a niche to the whole developer population | Same shift from role to skill as the one described in [[sfeir-ia-frontieres-metiers-skill-based-organisation-2026-08-01]], applied to engineering. ### Skill #2, a three-step mechanism 1. *« Understanding software fundamentals allows you to recognize what tradeoffs even exist »* — cost, scalability, reliability, speed, security, privacy. 2. The inexperienced developer who *vibe codes* fails not because they code poorly, but because they don't know the tradeoffs being made on their behalf: *« which will often be poor ones, because they don't know what context to give their coding agent »*. 3. Hence *« steering coding agents using the precise language of software engineering »* — domain knowledge becomes an interface again. Management consequence: AI shifts the senior engineer's value from execution to tradeoff-making. Converges with the comprehension debt from [[osmani-cognitive-surrender-comprehension-debt-2026-05-05]]. ### Continuity across four letters, April → August 2026 | Date | Letter | What it establishes | |------|--------|-----------------| | Apr. 24 | #350 | agents accelerate unevenly depending on the type of work | | May 8 | #352 | no jobs apocalypse, but a requalification "with evolving skills" — which ones is not said | | Jun 26 | #359 | engineers take on a partial product-management role | | Aug 14 | #366 | four skills, including *shaping the build*, which names June's role | The position is cumulative over four months, not opportunistic — which makes it sturdier than an isolated trend post. The point to carry into any recap: the same author states that jobs will be requalified, publishes the list of requalification skills, and notes that his organization teaches them. Everything is declared. Counterpoint: [[ng-the-batch-352-no-ai-jobpocalypse-2026-05-08]]. ### What the map leaves out | Absent from the four skills | Where it appears | |---|---| | Security | an aside in §2 — *« Security and privacy add further complexity »* | | Cost, token budget | an aside in §3 — *« without wasting excessive time or tokens »* | | Code review, testing | "testing" cited once in an enumeration | | Operations, observability, incidents | absent | | Teamwork, handoff | absent | | Compliance, data governance | absent | Most likely reading: the map describes what the **job market demands** (10,000 postings), not what an organization **should** require. The gap between the two lists is itself a result, not noted by the post. On security, see [[valente-zalewski-beyond-zero-enterprise-security-ai-era-2026-07-20]]. ### Reception (as of August 16, 2026) **3,074,365 views**, **17,271 likes**, **3,028 reposts**, **311 replies**, **35,343 bookmarks**. The bookmarks/likes ratio is ≈ **2.05**: the text was archived twice as often as it was applauded — a documentary-use signal, consistent with the "list of four" format. The replies/views ratio (311 for 3M) indicates it was saved rather than discussed. None of this measures agreement with, or the validity of, the four categories. ### Citation hygiene 1. Do not write "according to an analysis of 10,000 postings, the four most important skills are…": the post states a **method**, not a **result** — no percentage, no weighting, no time period or geography, and a clustering the author himself calls *"informal"*. Correct phrasing: *« Andrew Ng, drawing on an unpublished analysis of more than 10,000 job postings and on interviews, identifies four skills… »* 2. This post is **the announcement of a map, not the map itself**. Reopen the sheet when the detailed version is published. 3. Always mention that DeepLearning.AI teaches these skills — the author states it himself. 4. Prefer *The Batch* #366 over the X post for any durable citation; reserve the X link for reception data.

## RésuméDe400mots

X post by **Andrew Ng** from **August 14, 2026**, reprised from the "Dear friends" letter of ***The Batch* #366** (DeepLearning.AI).

**What is announced.** *The AI Engineering Skills Map*: **four skills** presented as the most important for a developer, backed by *« an analysis of more than 10,000 job postings »*, dozens of structured interviews with experts, hiring managers, and recruiters, surveys, and other online data. Explicit dual audience: helping developers **prioritize what they learn** and employers **hire**.

**The four skills.** **(1) Building and deploying AI applications** — their difference lying in the **unpredictability of outputs**, one must know the building blocks (LLM, context engineering, RAG, agentic workflows, machine learning, deep learning) and above all the **statistical techniques to measure, steer, and govern**, including *« disciplined evals and error-analysis loops »*. **(2) Software engineering fundamentals** — understanding them lets you *« recognize what tradeoffs exist »* (cost, scalability, reliability, speed, security, privacy) and thus steer the agent *« in the precise language of software engineering »*; the inexperienced vibe coder fails because *« they don't know what context to give their agent »*. **(3) Using coding agents** — a mental model of their limits, context management, planning/execution tradeoffs, **providing verifiers or evals so the agent closes its loops on its own**, working with a clear spec *« and when not to bother doing so »*, multi-agent orchestration, guardrails. **(4) *Shaping the build*** — since agents deliver well against a clear spec, the work shifts toward **deciding what the spec should contain**: product sense, business context, project ownership. **Underlying all four: a mindset of continuous learning**, with *« routines for trying new tools »*.

**The real thesis, slipped into a terminology note.** Ng talks about **skills** in AI engineering and not the **role** "AI Engineer": *« all developers should know how to work with the cloud, only a small number carry the "Cloud engineer" title »*. **AI engineering becomes a baseline, not a specialty** — you don't hire, you requalify.

**Two caveats.** **No numeric results are published**: no weighting, no sub-skills, a clustering described as an *« informal »* analogy, and the detailed map deferred to future posts. **This is the announcement of a map, not the map.** And the author states his interest: *« DeepLearning.AI's principal focus is to help developers gain these AI engineering skills. »*

## GrapheDeConnaissance

- Andrew Ng —publie→ AI Engineering Skills Map (DOCUMENT, 0.98)
- Andrew Ng —dirige→ DeepLearning.AI (ORGANISATION, 0.96)
- AI Engineering Skills Map —fait_partie_de→ The Batch (DOCUMENT, 0.94)
- AI Engineering Skills Map —affirme_que→ les quatre compétences d'AI engineering les plus importantes sont construire et déployer des applications IA, les fondamentaux du génie logiciel, l'utilisation des agents de codage et le fait de façonner la construction (AFFIRMATION, 0.97)
- AI Engineering Skills Map —est_basé_sur→ une analyse non publiée de plus de 10 000 offres d'emploi, des dizaines d'entretiens structurés avec experts, hiring managers et recruteurs, des sondages et d'autres données en ligne, le clustering étant qualifié par l'auteur d'analogie informelle (MESURE, 0.93)
- Andrew Ng —affirme_que→ il faut parler de compétences d'AI engineering plutôt que du rôle d'AI Engineer, parce que tous les développeurs devraient savoir travailler avec le cloud alors que seul un petit nombre porte le titre de Cloud engineer (CITATION, 0.96)
- compétences d'AI engineering —s_applique_à→ tous les développeurs — full-stack, data, DevOps, machine learning et AI engineers — et non à une spécialité identifiée par un titre de poste (AFFIRMATION, 0.95)
- compétences d'AI engineering —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.95)
- construction et déploiement d'applications IA —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.96)
- construction et déploiement d'applications IA —affirme_que→ la différence clé entre une application IA et une application classique est l'imprévisibilité de ses sorties, d'où le recours à des techniques statistiques pour mesurer, piloter et gouverner le système (AFFIRMATION, 0.95)
- construction et déploiement d'applications IA —utilise→ des boucles d'évaluations et d'analyse d'erreurs disciplinées, présentées comme la compétence centrale pour rendre un système IA plus prévisible (AFFIRMATION, 0.94)
- fondamentaux du génie logiciel —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.96)
- fondamentaux du génie logiciel —permet→ de reconnaître quels arbitrages existent entre coût, scalabilité, fiabilité, vitesse, sécurité et vie privée, et donc de mieux choisir sa stack, son architecture, son magasin de données et ses tests (AFFIRMATION, 0.95)
- fondamentaux du génie logiciel —permet→ de piloter un agent de codage dans le langage précis du génie logiciel (AFFIRMATION, 0.94)
- vibe coding —s_oppose_à→ la qualité du résultat lorsqu'il est pratiqué par un développeur inexpérimenté, qui ignore les arbitrages faits par son agent et ne sait donc pas quel contexte lui donner (AFFIRMATION, 0.93)
- utilisation des agents de codage —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.96)
- utilisation des agents de codage —recommande→ de fournir des vérificateurs ou des evals pour que l'agent ferme ses boucles de façon autonome, de gérer son contexte, d'arbitrer entre planification et exécution, de savoir travailler avec un spec clair et quand ne pas s'en donner la peine, et d'orchestrer plusieurs agents (AFFIRMATION, 0.95)
- utilisation des agents de codage —recommande→ de tenir des routines d'essai de nouveaux outils et de faire évoluer ses workflows, la pratique du codage agentique changeant trop vite pour que la connaissance des pratiques de pointe suffise (AFFIRMATION, 0.94)
- façonner la construction —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.96)
- façonner la construction —affirme_que→ les agents livrant de mieux en mieux un spec clair, le travail de l'ingénieur se déplace vers la décision de ce que le spec doit contenir, et un ingénieur ne doit plus s'attendre à recevoir un design pixel-perfect à seulement implémenter (CITATION, 0.95)
- façonner la construction —utilise→ le sens produit, la compréhension du contexte métier et des objectifs client, ainsi que la conduite de projet — savoir quand livrer vite un MVP à tester auprès d'utilisateurs et quand ralentir pour construire plus soigneusement (AFFIRMATION, 0.94)
- façonner la construction —converge_avec→ le rôle partiel de product management que Ng attribuait déjà aux ingénieurs dans sa lettre sur les trois boucles du développement produit (AFFIRMATION, 0.9)
- apprentissage continu —fait_partie_de→ AI Engineering Skills Map (DOCUMENT, 0.93)
- apprentissage continu —s_applique_à→ les quatre compétences comme socle commun, l'IA continuant de changer vite (AFFIRMATION, 0.92)
- DeepLearning.AI —affirme_que→ son objectif principal est d'aider les développeurs à acquérir ces compétences d'AI engineering (CITATION, 0.96)
- AI Engineering Skills Map —prédit→ la publication d'une carte plus détaillée et de billets développant chacune des quatre compétences, non parus à la date de la fiche (AFFIRMATION, 0.94)
- AI Engineering Skills Map —mesure→ 3 074 365 vues, 35 343 bookmarks, 17 271 likes, 3 028 reposts et 311 réponses sur X au 16 août 2026, soit un rapport bookmarks sur likes d'environ 2,05 (MESURE, 0.9)

---
Canonical: https://www.thekb.eu/en/fiches/ng-ai-engineering-skills-map-2026-08-14/