This is Nate's core discipline, and reading all 32 posts in order shows a real evolution rather than a static bag of tricks — the same way a craftsman's tool list changes as the material changes. Late 2024 through early 2025 is fundamentals: stop treating ChatGPT like a search box, and use a six-part skeleton (role, task, context, format, examples, constraints) that shows up, relabeled, in nearly every later post. February 2025's "Spellbook" organizes technique into four pillars — self-review/iteration, multi-stage/multi-model chaining, logic/precision, perspective/style — and by March he's formalizing multi-model orchestration explicitly: use GPT-4o to brainstorm broadly, a structure-oriented model to outline, a reasoning model to fill in logical depth, and a research-grade model to synthesize and cite, switching deliberately whenever a model shows logical errors, quality drift, or a knowledge wall rather than sticking with one model out of habit.
Spring 2025 is where the craft gets genuinely systematic. He builds a "four-block" blueprint (Purpose, Instructions, Reference, Output) plus two meta-switches — MODE (reflection/action/agentic) and EFFORT (how hard to think) — specifically because reasoning models like o3 default to bias-for-action and anchor hard on your first framing; you have to explicitly declare intent before the model will explore instead of execute. He also derives, from actual transformer mechanics, why prompt order matters as much as content: early tokens steer the whole probability space, output-format specs act as soft stop-tokens and belong last, and long chats need guardrails re-pinned every dozen or so messages because attention decays toward recency.
By June 2025 the single biggest turn happens: prompting itself gets reframed as one layer inside a bigger discipline he calls context engineering — the deterministic context you write versus the probabilistic context an agent autonomously discovers via search and tools, where your literal prompt can be a fraction of a percent of what the model actually processes. That reframing recurs and sharpens for the rest of the run. August's GPT-5 system-prompt teardown reveals a deliberate "bias to ship" (one clarifying question, then execute, no permission-seeking) that rewards front-loaded assumptions and tool-policy declarations over conversational back-and-forth. September's meta-prompt framework (role, objective, methodology, output contract, constraints, uncertainty protocol, self-check) and the "prompt doctor" post's six diagnosable failure modes (underspecification, regeneration-scope confusion, reasoning collapse, hallucination, consistency drift, context overload) both push toward treating prompts as enforceable specs, with structured-output schemas as the escalation path when prose isn't reliable enough.
Late 2025 keeps decomposing prompting into more specialized moves: Claude Skills package reusable methodology as loadable files rather than re-typed instructions; "Goldilocks prompting" argues most tasks want a ~500-token middle ground of principles and heuristics, not procedures, reserving fully deterministic prompts for machine-to-machine handoffs; the Gemini-3-vs-GPT-5.1 piece frames model choice as an entropy tradeoff (feed messy multimodal chaos to the context-entropy specialist, feed clean inputs and a hard task to the task-entropy specialist); and the six-stage prompt-lifecycle piece names the field's actual gap as intent formation, not wordsmithing — nearly all tooling serves drafting-through-deployment, almost none helps you figure out what you want before you start typing.
The most recent posts (Feb–Mar 2026) are the payoff of that whole arc. "Why your AI output feels generic" names the mechanism directly — RLHF optimizes for a median rater, so default output is calibrated to nobody — and pushes past prompting entirely into memory, persistent instructions, tool/app connections, and style profiles as compounding steering levers. "Prompting split into 4 skills" then reframes the entire cluster's history as a stack: Prompt Craft (the 2024-era conversational skill) is now table stakes; Context Engineering curates the environment; Intent Engineering encodes organizational values so agents don't optimize the wrong thing (the Klarna speed-over-relationship failure is the cautionary case); and Specification Engineering — writing complete, self-contained, verifiable specs that survive multi-hour unsupervised agent runs — is the frontier skill, because synchronous chat-style correction stops working once agents run for hours without you watching. The latest post pushes one layer further still: your rejections of AI output, not your prompts, are the compounding asset, because the constraint you articulate when you say "no, this is wrong because X" is the only part of the interaction that doesn't evaporate.
For an operator running a fleet of agents, the through-line is: prompting-as-clever-wording was always a bridge skill. What survives every model generation is structure, explicit intent, enforceable specs, and a captured record of what you rejected and why.
Key moves
Use a consistent prompt skeleton — role, objective, methodology/process, output contract, constraints, uncertainty handling, self-check — rather than ad hoc phrasing; treat vague instructions ("be concise," "be direct") as guaranteed to fail because they don't move the model off its default.
For reasoning models (o3-class and later), declare MODE (reflection/action/agentic) and EFFORT (how hard to think) explicitly at the top of the prompt — these models default to bias-for-action and will anchor hard on your first framing unless you tell them otherwise.
Order matters mechanically: put non-negotiable rules first, task-specific detail in the middle, and the exact output format/schema last — the format acts as a soft stop-token and the early tokens steer the whole rest of the generation.
In long-running chats, re-pin your core constraints every 10-15 messages; attention decays toward recent tokens, so old system rules silently lose weight even though they're still technically "in context."
Diagnose failing prompts against six recurring patterns before rewriting from scratch: underspecification, regeneration-scope confusion (fix one thing, everything regenerates), multi-step reasoning collapse, hallucination from lack of permission to say "I don't know," consistency drift from hidden ambiguity, and context overload from undifferentiated dumps of information.
Match models to the kind of uncertainty you need resolved: feed messy multimodal chaos (logs, screenshots, transcripts) to a context-entropy specialist with a tight output schema; feed clean, curated input to a task-entropy specialist for complex reasoning, planning, or narrative work.
Use "Goldilocks" mid-length prompts (roughly 500 tokens: named anti-patterns, categorical alternatives, decision heuristics) for the ~80% of tasks where you want the model to exercise judgment; reserve long, deterministic, schema-locked prompts for the ~20% where downstream systems or other agents depend on an exact output contract.
Treat context engineering as the layer above prompting: design what you directly control (system prompts, curated docs, retrieval rules) to responsibly guide what the agent will autonomously discover via search and tools, rather than assuming your literal prompt is most of what the model sees.
Package recurring methodology (brand voice, analysis frameworks, review checklists) as a loadable skill/file instead of re-explaining it every session — composable, reusable across models, and cheap in context until actually invoked.
Move beyond prompting for anything you do regularly: encode durable preferences in memory/custom instructions, connect the tools (search, files, calendar) that change what kind of answer you get, and build a style profile from real writing samples rather than describing tone in words.
For agent-delegated work that will run unsupervised for hours, write a full specification (acceptance criteria, explicit constraints on what must/must-not happen, decomposition into independently verifiable subtasks) rather than a conversational prompt — you can't course-correct in real time once the agent is running alone.
Capture your rejections, not just your prompts: when you reject AI output, write down exactly why in language precise enough to reuse — that articulated constraint is the durable, compounding asset; the output itself is disposable.
2026-02-27 — Prompting just split into 4 different skills — Prompt craft is table stakes now; context, intent, and specification engineering are the layers that actually differentiate results.
2026-02-05 — Why your AI output feels generic — RLHF optimizes models for a median rater; memory, instructions, tools, and style controls are the real levers beyond prompting.
2025-11-24 — The Complete Prompt Tooling Ecosystem — Prompt work has six stages and the tooling ecosystem almost entirely ignores the first one: intent formation.
2025-11-20 — My Lab Notes on Gemini 3 vs. ChatGPT 5.1 — Pick models by which kind of uncertainty (messy context vs. open-ended task) you need resolved, not by benchmark score.
2025-11-13 — 10x Your Prompt Power With a 100 Word Prompt — Mid-length "Goldilocks" prompts that give principles and heuristics beat both one-liners and 10-page spec dumps for most creative/judgment work.
2025-11-06 — The Prompt Doctor Is In — Six diagnosable failure patterns explain most broken prompts, each with a chat-level fix and a schema-enforced API-level fix.
2025-11-05 — ChatGPT 201: Advanced Prompting Made Easy — Five technique tiers (self-correction, meta-prompting, reasoning scaffolds, perspective engineering, specialized tactics) mapped to when each actually helps.
2025-10-17 — NEW: Claude Just Made Prompting 10x Easier — Claude Skills package reusable expertise into loadable files instead of re-typed mega-prompts, and the format is portable across models.
2025-10-15 — AI Has Solved the Oldest Problem: Scaling Expertise — The same four-part prompt formula (role, audience, goal, constraints) scales professional documentation across 27 completely different job families.
2025-09-26 — Quick Start: Make AI Optimize Your Prompts — Have the model score and rewrite its own prompt against your examples instead of manually guessing at better wording; DSPy formalizes this for production.
2025-09-09 — The Only Prompt Engineering Guide You'll Need — Consolidated reference: the CLEAR framework (Context, Length, Examples, Audience, Role) as the beginner on-ramp to everything else in this cluster.
2025-09-05 — Make ChatGPT-5 Write Like a Real Human — Forbidding corporate buzzwords and requiring concrete specifics overrides GPT-5's trained preference for AI-pleasing, jargon-heavy prose.
2025-09-04 — The ChatGPT-5 Prompting Manual — Full meta-prompt anatomy (role, objective, methodology, output contract, uncertainty protocol, self-check) with nine production templates for GPT-5's routing behavior.
2025-08-12 — Cracking the Agent Code — GPT-5's leaked system prompt reveals a deliberate "bias to ship": declare assumptions and tool policy up front or the model silently guesses.
2025-08-07 — ChatGPT-5 Is Coming — Argues most people are still prompting GPT-4o techniques on reasoning models and getting a fraction of available capability as a result.
2025-08-04 — Stop Letting AI Guess — Contract-first prompting: have the model interrogate you to 95% confidence and echo back its understanding before it generates anything.
2025-06-26 — Ready for ChatGPT-5: 139 Page Prompting Guide — 39 complete "thinking system" prompts spanning work and life decisions, built for frontier models rather than constrained to narrow job tasks.
2025-06-20 — Beyond the Perfect Prompt: Context Engineering — The pivotal reframe: your literal prompt is a sliver of what an agentic model actually processes; design the environment, not just the message.
2025-06-16 — How to Use AI When Your Brain Is Oatmeal — When you can't think clearly, dump the mess plus 2-3 "don't do X" guardrails and let the model ask the clarifying questions.
2025-05-29 — Surfing the Guardrails — Claude's leaked system prompt is ~90% failure-mode handling, not capability description — professional prompting is defensive programming.
2025-05-09 — How to Prompt with Wake Words — Save short named micro-rules to ChatGPT memory as one-word macros that apply only when explicitly invoked.
2025-05-02 — The Universal Anatomy of the Prompt — Seven load-bearing prompt sections and the token-level mechanics (causal attention, recency bias) explaining why their order changes accuracy.
2025-04-30 — Don't Just Think, Focus: Chain-of-Refinement — A Think→Critique→Refine loop on reasoning models turns structured, "tokenizable" inputs into self-edited drafts in two passes.
2025-04-24 — Your Prompt Is the Product — o3/o4-mini need explicit MODE and EFFORT declarations up front or they default to fast, action-biased, table-heavy answers.
2025-04-02 — Nate's Secret Sauce — A single 47-page research report required 19 iterated prompts across four models, arguing durable structure beats elegant one-shot wording.
2025-03-06 — Prompt Chaining Masterclass — Multi-model orchestration playbook: think in tokens, match model strengths to sub-tasks, and switch models on visible signs of struggle.
2025-02-14 — How to Escape AI Doom Loops — Endless re-prompting loops come from merging conceptual architecture and tactical implementation into one oversized ask; separate the phases.
2025-02-11 — The Prompt Spellbook — Twelve advanced techniques organized into four pillars: self-review, multi-stage/model collaboration, logic/precision, and perspective/style.
2025-01-31 — Your Pocket Guide to Prompt Engineering — Foundational six-element structure (role, goal, context, format, examples, constraints) cross-referenced against every major model maker's own docs.
2025-01-11 — The Art of AI Whispering — Condensed cheatsheet distilling a year of prompting lessons into quick, memorable top tips.
2024-12-14 — Stop Asking Your AI for Answers — The foundational reframe: treat ChatGPT as a conversation partner to think with, not a search engine to query.