AI Engineering Cheatsheets: Three Decision Tables for Model Selection, Agent Design, and Writing

What These Cheatsheets Are For

AI engineering decisions often involve picking among options that are functionally similar on the surface but meaningfully different in cost, latency, or complexity. Choosing between a workflow and an agent, or between a vector search setup and a hybrid retrieval approach, involves tradeoffs that depend on the specific task. This repository gives three prebuilt decision tables for the most common of those choices.

The goal, as stated in the README, is to let an engineer open a cheatsheet, find their situation in the table, and follow the recommendation. The author frames these as tested starting points from August 2026, with the explicit note that model and tool preferences are a snapshot and should be validated on the actual workload. That framing sets a realistic expectation: these are not permanent answers but calibrated first guesses.

The Three Cheatsheets and Their Coverage

The repository contains three files in addition to the README.

AI_Engineering_Playbook.md covers choosing the right AI technique, model, effort level, modality workflow, prompting strategy, RAG setup, memory pattern, evaluation method, and production configuration. It is the broadest of the three and addresses the most common decisions an AI engineer faces when building a system from scratch.

Agent_Architecture_Guide.md addresses the choice between a plain workflow, a single agent, and a multi-agent system. It then covers operating an agent system with approaches to context management, evidence stores, phased review, portable skills, feedback loops, and scheduling. This file is specifically for engineers building systems where an LLM takes actions, not just generates text.

Anti_Slop_AI_Writing_Guide.md is narrower. It targets situations where AI-assisted writing produces generic, low-quality output. The README describes it as covering a 7-section prompt structure, a set of checkable anti-slop rules, an evidence-first review process, a single targeted rewrite step, and platform checks. Unlike the other two, this one is sequential: the README specifies that sources should be assembled first, then the template filled, then the draft completed, and only then the evidence-first review run before a human edit.

How to Use These Files

The README gives a three-step workflow: open the relevant cheatsheet, find your situation in the decision tables, follow the recommendation. That is the intended interaction: table lookup, not linear reading.

For the Anti-Slop guide specifically, the README adds a more explicit sequence. Assemble sources before filling the template, complete the full draft, run the evidence-first review, then perform the human edit. Skipping the review step or the source assembly step defeats the purpose of that particular guide.

The repository has no install step. These are Markdown files in a public GitHub repository, and the fastest access is to open them directly on GitHub or clone the repository with a single command. The README does not reference a CLI, a web app, or any programmatic interface.

The Snapshot Problem: When Cheatsheets Date

The README is explicit: model and tool preferences are a snapshot from August 2026, described as tested starting points that require validation on the user's actual workload. This is not a disclaimer added out of caution. It reflects a real characteristic of AI engineering reference material.

Models that perform well on a benchmark in August 2026 may be superseded by newer releases within months. Recommended prompting strategies shift as model behavior changes with new training runs. The preference tables in AI_Engineering_Playbook.md will become progressively less accurate as the field moves. Engineers using these cheatsheets more than a few months after August 2026 should treat the specific model names and version references as potentially outdated, while the structural guidance on technique selection and agent architecture is likely to age more slowly.

Limitations for Teams vs. Individual Engineers

These cheatsheets are designed for individual engineers making implementation decisions. They do not cover organizational processes, cost forecasting, compliance constraints, or the kind of multi-stakeholder evaluation that large teams need before adopting a model or architecture pattern.

The Anti-Slop guide in particular is written for a single author producing AI-assisted long-form content, as suggested by its reference to a Towards AI long-form starting point. It is not a style guide for teams or a framework for content quality at scale. The Agent Architecture Guide also assumes a single engineer deciding on their own architecture, not a team standardizing patterns across multiple services.

Relationship to the Towards AI Courses

The README notes that the cheatsheets come from Towards AI courses and that those courses cover the same frameworks in more depth, with full lessons, code, and projects. The listed courses are Full Stack AI Engineering, AI for Business Professionals, and Agentic AI Engineering, all at academy.towardsai.net. A free agents webinar on YouTube (linked from the README) covers the workflows-versus-agents topic in more depth.

This means the cheatsheets function as condensed reference cards derived from paid instructional content. Engineers who find the tables useful and need more context behind any given recommendation have a documented path to fuller explanations through the linked courses, though those involve separate enrollment.

Alternatives and What Sets This Apart

Official documentation from model providers like OpenAI and Anthropic covers model selection and agent design, but it is organized around each provider's own products rather than cross-provider decision tables. Community resources like the LangChain documentation cover agent architecture in depth but are specific to that framework.

The distinguishing quality of this repository is its format: three opinionated, cross-provider decision tables that take a position rather than surveying all options. An engineer who wants to skip the survey and go directly to a concrete recommendation will find it faster to consult one of these cheatsheets than to read through vendor documentation. The trade-off is that the tables reflect one practitioner's tested preferences as of August 2026, not a consensus across the field.

Editorial conclusion

Engineers who regularly make AI technique, model, or agent architecture decisions will find these cheatsheets useful as a starting reference. The content is explicitly dated August 2026, so anyone using it after that point should confirm that model preferences and tool recommendations still apply to their context. The repository is not a course or a framework. It works best as a quick-reference layer on top of hands-on experience, not as a substitute for it. Start with the AI_Engineering_Playbook.md to find the decision table relevant to your current problem.

Frequently asked questions

Official sources

1. Issues
2. louisfb01/ai-engineering-cheatsheets on GitHub
3. Project website
4. README