Data, Analytics and AI
Generative AI, LLMs and agents
Generative models write, draw and summarise instead of classifying or forecasting, and they form the newest layer of the field rather than a replacement for the rest of it. What follows maps foundation models and the ways they are adapted, from prompting and retrieval to fine-tuning and agents, what the AWS and Microsoft AI exams now test about them, and the checks that decide whether a clever demo becomes a dependable feature.
Why this topic exists: Foundation models, prompting, retrieval, fine-tuning and agents: the newest layer of the field and the one AIF-C01 and Azure AI Fundamentals now test (the AI portal reads the concepts in order).
Generative AI is a label for models that produce content. For practitioners the more useful unit is the foundation model: a large network, usually a transformer, trained once at great expense on broad data and then adapted to many tasks. Large language models are the text-shaped members of that family and the ones most current work involves. The skill in this topic is not training such models, which a handful of laboratories do, but choosing, adapting and constraining them for a particular job.
Four ways to adapt a model, cheapest first
- Prompting. Instructions, context and worked examples placed in the request itself. Prompt engineering costs nothing to change and is the right first attempt for almost every task.
- Retrieval. Search your own documents at the moment a question comes in and hand the relevant passages to the model. Retrieval-augmented generation is how a general model answers from your policies rather than from memory, and most of its quality lives in the search step.
- Fine-tuning. Further training on your own examples to fix a format, a tone or a narrow skill. It changes behaviour more than knowledge, and it needs curated data plus a way to measure improvement.
- Agents. A loop in which the model calls tools, reads what they return and decides the next step. Agentic AI turns a text generator into something that acts, which raises the price of every error.
The order matters because each step multiplies cost and risk. Teams that reach for fine-tuning to fix a problem better retrieval would have solved spend weeks assembling training data for no gain. The AWS AI Practitioner exam gives applications of foundation models a domain of their own and asks candidates to choose among exactly these options; Azure AI Fundamentals and Oracle's OCI AI Foundations cover the same vocabulary on their platforms. Microsoft's Applied Skills lab for Foundry agents goes a step further and has you deploy a model and build a working agent in a live environment.
Decide who checks the output
Every generative feature needs a named reviewer or an automatic check for the cases where the output is wrong, because it will sometimes be wrong with total fluency. Where no person reads each result, constrain the format, validate it against rules and log enough to reconstruct what the model was shown. Prompt injection, meaning instructions smuggled into the content a model reads, is the security form of the same problem.
What this layer leans on
The generative layer depends on the rest of the silo more than its marketing admits. Retrieval is a data engineering and search problem, evaluation is statistics, and running an application at scale is MLOps. Copyright, personal data and disclosure belong to responsible AI. People arriving from software development tend to do well here because much of the work is systems integration; people arriving from analysis bring the measurement discipline the area most lacks. The AI portal walks through the thirteen underlying concepts in reading order.
Next to this topic
- Data analysis fundamentalsAsk, prepare, process, analyse, share: the analyst's workflow as Google's programme and DP-900's core data concepts teach it, before any tool (SFIA Data analytics).
- SQL for analysisQuerying, joining, aggregating and windowing data is the analyst's first real tool and the language of every warehouse (DP-900 relational concepts; DMBOK Data Storage and Operations; the programming field owns database design).
- Data visualisation and business intelligenceDashboards, reports and the semantic models behind them: PL-300's four skill areas and DMBOK's Data Warehousing and BI area (Tableau, Power BI, Looker).
- Statistics and experimentationProbability, inference, regression and A/B tests separate a chart from a claim (Google's Advanced Data Analytics; SFIA Data science).
- Python for data workpandas, NumPy, notebooks and plotting: the working environment of analysis and machine learning (SFIA Data science; the AI pillar's argument for Python).
- Data engineering and pipelinesIngestion, transformation, warehouses, lakehouses and orchestration (DMBOK Data Integration and Interoperability; DEA-C01; DP-700; Databricks; Spark, Airflow, dbt).
- Data governance, quality and privacyOwnership, definitions, metadata, master data, quality rules and lawful use are DMBOK's centre and half its wheel; nobody hires for them until the numbers disagree.
- Machine learningSupervised and unsupervised learning, model evaluation, deep learning and the discipline of not fooling yourself (SFIA Machine learning; AIF-C01; Azure AI Fundamentals).
- Natural language processing and computer visionText and images are the two applied fields where classic machine learning met deep learning first, and where most production AI still lives (SFIA Machine learning; Azure AI Fundamentals AI-901 workloads).
- Responsible AI and AI governanceBias, transparency, privacy, safety and the frameworks that make them work (NIST AI RMF; AIF-C01's responsible-AI domain; SFIA 9's AI and data ethics skill; the EU AI Act).
- MLOps and AI engineeringGetting a model or an LLM application into production and keeping it there: deployment, monitoring, evaluation, cost (SFIA Machine learning and Data engineering; Applied Skills in Foundry; MLOps in the AI portal).
Concepts to know
Glossary entries with the reason each one matters here.
- GenAI
The topic's own concept.
- LLMs
The model class at the centre.
- Prompt Engineering
How the models are instructed.
- RAG
Grounding a model in your own data.
- Agentic AI
Models that act, not only answer.
- Transformer
The architecture underneath.
Certifications that test it
Vendor exams and free certificates; facts, cost and the preparation path are on each page, and the certifications hub has them all.
- AWS · AIF-C01AWS Certified AI PractitionerFoundation models are a fifth of the questions.
- Microsoft · AI-901Microsoft Certified: Azure AI FundamentalsGenerative AI workloads on Azure.
- Microsoft · Lab-based assessment · freeMicrosoft Applied Skills: Get started developing agents in Microsoft FoundryBuilding an agent in a lab, free.
Tools of the trade
Frequently asked
- Do I need machine learning to work with LLMs?
- Not to build useful applications on hosted models, which is mostly software work. You do need the evaluation instincts machine learning teaches: held-out test cases, suspicion of scores that look too good, and the understanding that performance on your own inputs is the only number that counts.
- What is a hallucination?
- Output that is fluent and wrong, such as an invented citation or a confident answer to something the source never addressed. It follows from how language models generate text, one plausible continuation at a time. Grounding answers in retrieved documents, asking for sources and checking them reduces it without eliminating it.
- When is fine-tuning worth the effort?
- When prompting and retrieval have been tried properly and the remaining gap concerns form or a narrow skill: a strict output format, a house style, a specialised classification run at high volume. It is rarely the right way to add facts, which change faster than you can retrain.
Courses in the directory
485 courses are filed here; the top 6 by our ranking, details and the provider link on each course page.
Last reviewed 26 September 2026 · Getting Digital