For those who created the original FEP
Crosswalk: what happened to each topic what you studied
The original FEP (2025) had 8 modules and 47 topics. This table shows which bucket each one fell into with the September 2026 models and which FEP 2026 lesson it appears in again. Nothing was lost: the “destination” half of what you learned matters more than ever.
diedonly when it helpscontinuesnew / rewritten
| FEP module | Original topic | Bucket | In FEP 2026 |
|---|---|---|---|
| 1 | Introduction to generative AI · What is an LLM | continues | Lesson 1, condensed: what AI in 2026 does on its own and what it doesn’t |
| 1 | Tokens and context window | continues | Lesson 4, as a concept: the table has a size; the medium gets less attention |
| 1 | Clarity · Anatomy (the 4 ingredients) | continues | Lesson 3: became the six parts of the brief |
| 2 | Effective contextualization | continues | Lesson 4: a five-line context sheet, with a home of its own |
| 2 | Zero-shot | continues | Lesson 3: is the default; it stopped being a “technique” |
| 2 | Few-shot | only when it helps | Lesson 5: when the default can’t be put into words |
| 2 | Role prompting | only when it helps | Lesson 5: role as a perspective with a criterion, in one sentence |
| 2 | Chain-of-Thought · Skeleton of Thought | died | Lesson 2 (why) and lesson 7 (what remains: effort) |
| 2 | 24 Persona Dimensions · EXPERT Framework · Ultimate Prompt | died | Lesson 2: a persona template doesn’t change the result; perspective in one sentence, in lesson 5 |
| 2 | Text humanization | died | Lesson 2; voice goes in the context sheet (lesson 4) |
| 2 | Interactive prompts | only when it helps | Lesson 5: ask first, only when the brief is unclear |
| 3 | Task decomposition · Prompt chaining | only when it helps | Lesson 2 (manual chaining is dead) and lesson 5 (steps only with human approval) |
| 3 | Formatting and structuring | continues | Lesson 3: delivery format, within the definition of good |
| 3 | Negative instructions | died continues | Lesson 2: style lists are dead; business boundaries remain (lesson 3) |
| 3 | Parameter tuning (temperature, top_p) | died | Lesson 2; replaced by model and effort selection (lesson 7) |
| 3 | Output prefilling | died | Lesson 2: request the format or native structured output |
| 3 | Iterative refinement | continues | Lesson 3: iterate on the brief, not the micro-instruction |
| 4 | Long context management | only when it helps | Lesson 4: rules at the edges, only the necessary material |
| 4 | Multimodal prompting | only when it helps | Lesson 5: only when the actual input is an image, audio, or document |
| 4 | Model-specific optimization | rewritten | Lesson 7: which model for which task; read a new model’s guide in ten minutes |
| 4 | Structured outputs (JSON/XML) | only when it helps | Lesson 5: only when another program will read it |
| 4 | Prompt testing and A/B testing | continues | Lesson 7: weekly sample with an observable criterion |
| 5 | Context architecture | continues | Lesson 4 |
| 5 | Embeddings · Vector databases · RAG | died | Lesson 4, one section: when the foundation doesn’t fit in the project. Building RAG became the builder’s job |
| 6 | Tool design · MCP · Memory · ReAct · Planning · Error handling | rewritten | Lesson 6: gates, permissions, and approval, for those who delegate and don’t program |
| 7 | Extended thinking | rewritten | Lesson 7: effort level in three steps |
| 7 | Multi-agent systems | rewritten | Lesson 6, briefly: two human checkpoints. In-depth coverage in the Super-Agentes course |
| 7 | Claude Skills | promoted | Lesson 4: skill as the house manual, for anyone |
| 7 | Voice AI and multimodal agents | died | What’s out: it became a finished product, outside the requester’s scope |
| 8 | Security and prompt injection | continues | Lesson 6: became more important for agents that read third-party content |
| 8 | Ethical AI and alignment | continues | Lesson 7: responsibility—who answers for what came out |
| 8 | Observability · Production Systems | died | Lesson 7, one section: measure with a sample. The rest is a programmer’s course |
| — | 100 Essential Prompts | died | 12-brief kit, in four adaptable blocks |
| — | GPT-5.5 prompting guide | died | Lesson 7: how to read any new model’s guide using three questions |
| — | Certification by level | died | Each lesson ends with an artifact you can show |
How to use this table
Mark the three topics you used most. If they landed in “dead,” Lesson 2 explains what to use instead. If they landed in “only when useful,” Lesson 5 gives you the condition. If they landed in “still going,” you already know the main idea, and the course shows you what changed places. There’s one criterion behind all the buckets: did the technique describe what you want, or compensate for a weakness in 2024 AI?