Split View: 2026 떠오르는 직무별 필요 지식 지도 — 무엇을 공부해야 하는가

A Knowledge Map for the Rising Tech Roles of 2026 — What to Actually Study

• Introduction — Draw the Map First, So You Never Get Lost
• AI / LLM Engineer
• Platform · DevOps / SRE
• Security Engineer
• Data Engineer
• Frontend Engineer
• The Skills That Cut Across Every Role
• How to Start
• References

Introduction — Draw the Map First, So You Never Get Lost

Open any tech job posting and you will find dozens of required skills listed. Try to study that entire list and you will certainly burn out. What you need is not a list but a map — the structure of what the core axes of the role are, what counts as fundamentals, and what is genuinely emerging.

This article draws knowledge maps for the five roles with the clearest demand in 2026: AI/LLM engineer, platform/DevOps/SRE, security engineer, data engineer, and frontend engineer. Each role is organized in three layers — fundamentals → core → emerging — connected to tools on this site you can practice with immediately and to well-regarded references. This is part 1 of a trilogy: part 2, the tech interview playbook, covers proving this knowledge in interviews; part 3, studying with AI, covers acquiring it as fast as possible.

AI / LLM Engineer

The hottest role right now, but behind the flashy demos the actual work is surprisingly solid engineering. Most of it is not building models — it is integrating models into products safely.

Fundamentals

• How transformers work — tokenization, embeddings, self-attention, why the context window is finite
• Python and data-handling basics, API design
• A working feel for probability and statistics (what temperature and top-p sampling actually do)

Core knowledge

• RAG (retrieval-augmented generation): chunking strategies, choosing embedding models, vector databases, reranking — plus the judgment call "consider RAG before fine-tuning"
• Prompt engineering: system prompt design, few-shot, structured output (JSON schemas), tool use
• Evals: how to measure LLM output quality — golden datasets, the limits of LLM-as-judge, regression tests
• Serving economics: token costs, latency budgets, caching, streaming — the gap between demo and production mostly lives here

Emerging topics

• Agent architectures (plan-execute-verify loops, multi-agent orchestration)
• LLM security — prompt injection, data-exfiltration prevention (an emerging overlap with the security role)
• Small models, on-device inference, and cost optimization

On this site, build visual intuition for attention and backpropagation in the neural network lab, and practice prompt patterns with the prompt engineer tool.

Platform · DevOps / SRE

The definition is converging on "the role that lets developers ship fast and safely." It is a hybrid of infrastructure knowledge plus software engineering.

Fundamentals

• Linux — processes, filesystems, networking, shell scripting (being able to read htop is a good litmus test)
• Networking basics — DNS, TLS, HTTP, load balancing
• At least one programming language (Go or Python is the de facto standard)

Core knowledge

• Containers and Kubernetes: image build optimization, pod scheduling, services/ingress, resource requests and limits
• IaC (infrastructure as code): Terraform state management, modularization, drift detection
• CI/CD: pipeline design, progressive delivery (canary, blue-green), rollback strategy
• Observability: the three pillars — metrics, logs, traces; designing SLOs and error budgets; managing alert fatigue
• Incident response: on-call culture, blameless postmortems

Emerging topics

• eBPF-based observability and security (service mesh without sidecars)
• FinOps — treating cloud cost as an engineering metric
• Platform engineering — internal developer platforms (IDPs) and self-service golden paths

Practice hands-on with the Kubernetes playground, Linux terminal, eBPF playground, and container lab, and check certification-level knowledge with the Kubestronaut quiz.

Security Engineer

As every company becomes a software company, security has shifted from a specialty to a precondition of every system. It is one of the roles with the worst supply-demand gap.

Fundamentals

• OWASP Top 10 — understanding injection, cross-site scripting, and broken access control at the code level
• Practical cryptography sense — hash vs. encryption, symmetric vs. asymmetric, the TLS handshake (leave implementation to libraries; keep the judgment yourself)
• The attack surface of networks and operating systems

Core knowledge

• Authentication and authorization: how OAuth 2.0 and OIDC flows differ, JWT signature verification/expiry/the storage-location debate, sessions vs. tokens
• Cloud security: least-privilege IAM design, secrets management (vaults, rotation), network isolation
• Supply-chain security: dependency scanning, SBOMs, build integrity — the fastest-growing area of recent years
• Secure SDLC: threat modeling, the security lens in code review, SAST/DAST

Emerging topics

• Zero-trust architecture — from perimeter defense to identity-centric defense
• Securing LLM/AI systems — prompt injection, data exfiltration through models, vulnerabilities in AI-generated code
• Passkeys and the spread of phishing-resistant authentication

Touch the concepts directly in the auth & security lab, crypto lab, JWT decoder, and hash generator.

Data Engineer

"Data is the new oil" is a tired phrase, but demand for the people who build the refineries has never been higher. The AI boom is also a data-pipeline boom.

Fundamentals

• SQL, deeply — window functions, reading execution plans, how indexes behave. In this role SQL is both the fundamental and the final weapon
• One programming language (usually Python) and a feel for data structures
• Data modeling — normalization, star schemas, slowly changing dimensions (SCD)

Core knowledge

• Batch and streaming: ELT patterns, dbt-style transformation layers, Kafka-based streaming, what exactly-once processing really means
• Lakehouse: Parquet, Iceberg/Delta table formats, catalogs
• Data quality and governance: tests, lineage, handling late-arriving data
• Orchestration: DAG design, idempotent pipelines, backfill strategy

Emerging topics

• Embedded analytics à la DuckDB — the rediscovery of "small data"
• Data preparation for AI workloads — embedding pipelines, vector index management
• The convergence of streaming and batch (unified processing engines)

Experiment with queries and streaming concepts in the SQL playground, PostgreSQL playground, DuckDB playground, and message queue playground.

Frontend Engineer

While the "AI will replace frontend" talk circulated, the actual bar rose. Stamping out components shrank as work; judging performance, accessibility, and architecture is what remains.

Fundamentals

• HTML/CSS taken seriously — semantic markup, layout (grid/flex), responsive design
• JavaScript deep dive — closures, the event loop, promises, modules
• TypeScript — no longer optional; it is the dialect

Core knowledge

• React today: designing the server/client component boundary, Suspense, the trend toward simpler state management
• Performance: Core Web Vitals (LCP/CLS/INP), bundle analysis, image optimization, the trade-offs of rendering strategies (SSR/SSG/ISR)
• Accessibility (a11y): keyboard navigation, correct (and restrained) ARIA, color contrast
• Testing strategy — integration over unit, user-perspective tests

Emerging topics

• Edge rendering and partial prerendering
• AI-assisted development workflows — the new differentiator is the review skill to verify and refactor generated code
• Design systems and design-code synchronization

Build the fundamentals with the JS sandbox, HTML/CSS playground, CSS quiz, and regex quiz.

The Skills That Cut Across Every Role

Overlay the five maps and a common terrain appears. Whichever role you choose, these four compound.

• Writing: design docs, postmortems, PR descriptions — the more senior you get, the more writing out-leverages code. It is also the core vehicle of "visible work" discussed in the compounding career.
• Using AI tools: a productivity multiplier regardless of role. But commanding the tool rather than being consumed by it is a skill of its own — covered in part 3, studying with AI.
• English: the latest docs and discussions still appear in English first. The roadmap in what fluency really is helps.
• A feel for distributed systems: networks drop, retries create duplicates, clocks drift — every role works on top of this reality.

How to Start

No need to be overwhelmed by the maps. There is an order.

1. Pick one role — combining roles is a story for after the fundamentals are in place.
2. Start from the fundamentals layer — fundamentals do not go out of fashion, and interviews test them first.
3. Learn the core layer through projects — knowledge that is only read does not stick. Build something small and ship it.
4. Pick exactly one emerging topic and go deep — chase them all and everything stays shallow.
5. Record and publish — writing up what you learned compounds both memory and career.

The next part covers proving this accumulated knowledge in the interview room — concrete preparation for coding tests, system design, and behavioral interviews.

References

• roadmap.sh — community roadmaps by role
• Google SRE Book (free online)
• OWASP Top 10
• Martin Kleppmann, "Designing Data-Intensive Applications"
• web.dev — Core Web Vitals
• Stack Overflow Developer Survey