All roadmaps

Prompt Engineering

The highest-leverage writing skill of the decade.

How LLMs interpret prompts, the core and advanced techniques, output control, safety awareness and evaluation - prompting as an engineering discipline, not a bag of tricks.

Your progress0/31 topics · 0%

Progress is saved in this browser.

4 stages31 topics₹0 to follow

Filled nodes are the focus for this track; outlined ones can wait. Tap a topic to open it and mark it done, learning or skipped.

Prompt Engineering in 2026Related roadmapsAI Agents roadmapAI Engineer roadmapClaude Code roadmapProve it as you goReading is half of it. This roadmap isbacked by a free, scored PromptEngineering certification with a publicverification link and a AI Engineer jobsimulation on real tickets - evidence arecruiter can check, not a claim. Marktopics done, learning or skipped as yougo; your progress carries into theclassic view.Get certified - freeJob simulationClassic viewKnow the demand before thesyntaxPrompt Engineering in 2026What to learn in 2026Prompts asspecsEvals andstructureContext engineeringWhich roles use itEvery knowledgeroleAI product andengFresher vs switcherWhat companies expectThe hiring barWhat interviewstestProof they verifyWhat gets you hiredA 90-day planProjects thatget readCertification and proofUnderstand the machine you'reinstructing.FoundationsHow LLMs read promptsRoles & system promptsAnatomy of a good promptThe patterns that fix 80% ofbad outputs.Core TechniquesZero-shot vs few-shotChain-of-thoughtOutput formattingDecompositionFrom good prompts to robustsystems.Advanced PracticeSelf-critique & revisionloopsLong-context strategiesPrompt injection awarenessEvaluating promptsFresher or experienced?Both tracks hold every topic. On the Fresher track thefoundations are filled and the later stages are outlined - dothem when a project needs them. On the Experienced track thefoundations are outlined and the later stages are filled. The2026 section at the top applies to both.Prefer a list?The classic view shows the same 3 stages as a sequentialchecklist with time estimates. Your marks are shared betweenthe two views.Open the classic view

Everything on this roadmap

Every stage and topic on the map, in reading order. Click any topic to open it.

1.Know the demand before the syntax

2.Understand the machine you're instructing.

3.The patterns that fix 80% of bad outputs.

4.From good prompts to robust systems.

The Prompt Engineering roadmap, topic by topic

Every topic on the map in reading order, with what to learn and the resources we recommend. Main topics are open; open any subtopic to read it. Select a topic name to open it in the topic panel, where you can track your progress.

  1. 1. Know the demand before the syntax

    Prompt Engineering in 2026

    2026

    Prompt engineering went from a curiosity to a listed skill in two years: the Stanford AI Index 2025, using Lightcast data, found job postings mentioning generative AI skills rising almost fourfold in a year to more than 66,000, and postings naming prompt engineering rising from about 1,400 to nearly 6,300; LinkedIn's Skills on the Rise 2025 put AI literacy first and found it appearing in job descriptions six times more often than a year earlier. In 2026 the skill has matured into something closer to specification writing and evaluation. The four nodes answer what to learn, which roles use it, what companies expect and what gets you hired.

    What to learn in 2026

    Learn to write a prompt as a specification, to structure outputs and measure them with evaluations, and then the 2026 discipline of context engineering - deciding what the model sees. The sub-nodes order the three.

    Prompts as specs

    A good prompt is a specification: the role and the task, the inputs and their format, the rules and the edge cases, examples that show the shape of a good answer, and the output format a system can parse. Clear instructions, few-shot examples, step-by-step reasoning where it helps and explicit refusal cases are the techniques; the discipline is writing for a reader that takes everything literally. The Stanford AI Index 2025's fourfold rise in generative AI postings is demand for people who can do this reliably, not for people who have tried a chatbot.

    Evals and structure

    Prompts that matter are measured: an evaluation set of real inputs with expected outputs, a scoring rule - exact match, a rubric, a model judge with its limits understood - and a number that moves when the prompt changes. Structured outputs through schemas turn a model into a component other code can call, and validation catches the cases where it does not comply. The AI Engineer roadmap's evaluation stage carries this in depth; without it, prompt engineering is guessing with confidence. The Stanford AI Index 2025 counted nearly 6,300 postings naming prompt engineering; the ones that pay well want the evaluation, not the guess.

    Context engineering

    AI

    Context engineering is the 2026 form of the skill: deciding what a model sees - the retrieved documents, the tool results, the conversation history, the memory - and in what order, within a budget, so that the answer is grounded and the cost is bounded. Prompt caching, compaction of long conversations and retrieval with permissions are its tools. LinkedIn's Skills on the Rise 2025 ranked AI literacy first, and the Work Trend Index found 75 percent of knowledge workers using AI at work; the people who get the most from models are the ones who manage their context, not just their wording.

    Which roles use it

    Every knowledge worker now prompts; AI product managers, engineers and operations specialists do it as a craft. The sub-nodes name both groups and lay out the fresher and switcher routes.

    Every knowledge role

    Analysts, marketers, writers, lawyers, recruiters, teachers, support staff and managers all prompt daily now; the Work Trend Index found 75 percent of knowledge workers using AI at work and 66 percent of leaders unwilling to hire without AI skills, and LinkedIn's data shows AI literacy in job descriptions six times more often than a year before. For these roles the skill is a multiplier on the job you already have: faster drafts, better analysis, fewer errors when the output is checked. The English and Excel roadmaps show the same pattern in their own domains.

    AI product and eng

    AI product managers, AI engineers, applied scientists and AI operations specialists do prompting as a craft inside products: designing the prompts and context, building the evaluation sets, running experiments and owning the quality of a model-backed feature. The Stanford AI Index 2025's rise in prompt engineering postings from about 1,400 to nearly 6,300 is this group, and most of those postings sit inside engineering and product titles rather than under the name prompt engineer. The AI Engineer roadmap is the path for the engineering version.

    Fresher vs switcher

    A fresher should build one measured feature - a prompt, an evaluation set, a result - rather than a collection of clever prompts, because the measurement is what employers cannot get from a chatbot; HackerRank's 2025 report found early-career hiring the hardest hit, so a documented evaluation is the differentiator. A switcher from any knowledge role should apply the skill to their own domain, where they know what a good answer looks like. On this map freshers get the basics filled and the advanced stages outlined; experienced learners get the reverse.

    What companies expect

    Companies expect prompts that work reliably on real inputs, measured against an evaluation set, with structured outputs a system can use and an honest account of where the model fails. The sub-nodes cover the bar, the interview and what gets verified.

    The hiring bar

    The bar is reliability: a prompt that produces the right output on real inputs most of the time, with the failure cases known and handled, outputs structured so code can use them, an evaluation set that proves it, and a cost per call you can state. The Work Trend Index's 71 percent of leaders who prefer AI-skilled candidates means the skill is expected; the evaluation is what makes it credible.

    What interviews test

    Expect to improve a prompt live on a real task and explain each change, to design an evaluation for a given feature, to discuss where a model will fail and what to do about it, and to show a workflow or feature you built. For engineering roles expect structured outputs, retrieval and cost questions; for product and operations roles expect judgement about what AI should and should not do in the workflow. The strongest answers come with numbers from an evaluation you ran.

    Proof they verify

    Hiring teams read evaluation results: a repository or document with the prompts, the inputs, the scores before and after, and the failure analysis beats any list of techniques; a working feature or workflow someone else uses beats a demo. Pair it with a verified OneRoadmap certification - the Prompt Engineering and AI Engineer certifications are free and carry a public verification link - and a short account of your own AI workflow.

    What gets you hired

    A prompt library with evaluation results, a working feature or workflow built on it, and a clear write-up of what the model does well and badly. The sub-nodes give the ninety-day plan, the projects that get read and the proof that counts.

    A 90-day plan

    Month one: the techniques applied to one real task from your work or study, with twenty real inputs and a scoring rule. Month two: structured outputs and an evaluation set of a hundred inputs, a prompt improved against it with the numbers recorded, and retrieval added where the task needs facts. Month three: a working feature or workflow built on it that someone else uses, a write-up of the failure cases, the Prompt Engineering certification, and twenty targeted applications a week. LinkedIn's Skills on the Rise 2025 put AI literacy first; a measured workflow is the proof of it.

    Projects that get read

    Projects that get read are measured workflows: a classification or extraction prompt with an evaluation set and a confusion matrix; a document-answering tool with retrieval, grounding checks and a cost note; a support or operations workflow that drafts and a human approves, with adoption numbers; a prompt library for a team with the evaluation behind each entry. The Stanford AI Index 2025's demand numbers are for exactly this - prompting that is engineered.

    Certification and proof

    Proof is verifiable: OneRoadmap's free Prompt Engineering certification with a public link, a repository of evaluated prompts, a feature or workflow in use and a short account of how you work with models. With AI literacy the fastest-rising skill on LinkedIn and in six times more job descriptions than a year before, the candidates who stand out are the ones who can show a number.

  2. 2. Understand the machine you're instructing.

    Foundations

    Understand the machine you're instructing.

    Time: about 1 week. Topics: How LLMs read prompts · Roles & system prompts · Anatomy of a good prompt.

    How LLMs read prompts

    Tokens, context windows and next-token prediction - why phrasing, order and format change results.

    Roles & system prompts

    System vs user messages, personas and behavioural constraints - setting the frame before the task.

    Anatomy of a good prompt

    Task, context, constraints, examples, output format - a repeatable template that survives new models.

  3. 3. The patterns that fix 80% of bad outputs.

    Core Techniques

    The patterns that fix 80% of bad outputs.

    Time: about 1-2 weeks. Topics: Zero-shot vs few-shot · Chain-of-thought · Output formatting · Decomposition.

    Zero-shot vs few-shot

    When instructions suffice and when 2–3 curated examples teach the pattern better than any description.

    Chain-of-thought

    Giving models room to reason before answering - and when it helps vs wastes tokens.

    Output formatting

    Forcing JSON, tables, or exact templates - outputs that downstream code and humans can consume.

    Decomposition

    Splitting big asks into chained small ones - reliability through smaller, checkable steps.

  4. 4. From good prompts to robust systems.

    Advanced Practice

    From good prompts to robust systems.

    Time: about 1-2 weeks. Topics: Self-critique & revision loops · Long-context strategies · Prompt injection awareness · Evaluating prompts.

    Build: Prompt library. Build a personal library of 10 tested prompt templates for your real work, each with example inputs and expected outputs.

    Self-critique & revision loops

    Draft → critique → revise patterns that trade tokens for quality where it matters.

    Long-context strategies

    Document placement, quoting and extraction-first patterns for 100k-token contexts.

    Prompt injection awareness

    How untrusted content hijacks instructions - writing prompts that degrade safely.

    Evaluating prompts

    Fixed test sets, A/B comparison and regression checks - prompts as versioned, tested artifacts.

Prove it as you go

Reading is half of it. This roadmap is backed by a free, scored Prompt Engineering certification with a public verification link and a AI Engineer job simulation on real tickets - evidence a recruiter can check, not a claim. Mark topics done, learning or skipped as you go; your progress carries into the classic view.

Fresher or experienced?

Both tracks hold every topic. On the Fresher track the foundations are filled and the later stages are outlined - do them when a project needs them. On the Experienced track the foundations are outlined and the later stages are filled. The 2026 section at the top applies to both.

Prefer a list?

The classic view shows the same 3 stages as a sequential checklist with time estimates. Your marks are shared between the two views.

Frequently asked questions

Is prompt engineering a real job?

It is a real skill inside many jobs and occasionally a title. The Stanford AI Index 2025 found postings naming prompt engineering rising from about 1,400 to nearly 6,300 in a year, most of them inside engineering, product and operations roles. LinkedIn's Skills on the Rise 2025 put AI literacy first; the craft version of it - prompts as specifications, measured with evaluations - is what pays.

Do I need to code to learn prompt engineering?

No, to use it; yes, to engineer it. Writing a prompt as a specification and judging outputs needs no code and helps every knowledge role. Building a reliable feature - structured outputs, an evaluation set, retrieval - needs enough programming to call an API and score results. The AI Engineer roadmap carries that half.

Is prompt engineering enough on its own?

No. Prompting is one tool and on its own rarely survives real users. What employers look for is the prompt turned into a component: structured output a system can use, an evaluation set that proves it, retrieval with permissions where facts matter, a cost per call you can state. That is why evaluation and context engineering are stages of their own here.

What is context engineering?

Deciding what the model sees - the retrieved documents, the tool results, the conversation history, the memory - in what order and within what budget, so the answer is grounded and the cost is bounded. It is the 2026 form of the skill, and it matters more than wording once a system has real data behind it.

Which model should I learn on?

Any of the major ones; the techniques transfer. Learn the differences by running the same evaluation set against two models and reading where they diverge. The skill is the specification and the measurement, not the vendor.

How long does it take to learn?

The techniques take days; reliability takes weeks of building evaluation sets against real inputs and improving prompts against the numbers; context engineering takes a real project with real data. Nobody can promise a timeline. A measured workflow someone else uses is the evidence.

Will better models make prompt engineering obsolete?

Better models made the tricks obsolete and the specification more valuable: the clearer the instruction, the examples and the format, the more a capable model does with them. What remains is deciding what the model should do, proving it does it and managing what it sees. Those are not going away.

How do I prove prompt engineering skills without job experience?

A repository of evaluated prompts: the inputs, the scoring rule, the numbers before and after, the failure analysis; a feature or workflow built on them that someone else uses; and OneRoadmap's free Prompt Engineering certification with its public verification link. A number from an evaluation you ran beats any list of techniques.