What Actually Happens When You Read Making Prompts Weekly
Making Prompts Weekly is a newsletter that breaks down prompt engineering techniques, shares template structures, and reports on new capabilities from major language model providers. It comes out on a set schedule, usually Thursday mornings if you subscribe, and the content ranges from practical how-tos to industry news summaries. I've been reading it since the early days of large language model adoption, when most people were still figuring out why their prompts kept producing generic nonsense. The founder is someone who went from working in enterprise software to building prompt libraries full-time, which shows in the specificity of the examples. Most prompt guides read like they were written by people who tested something once. This one tends to include actual edge cases and the failures that came before them.
Getting Set Up With Making Prompts Weekly
You can find the subscription page by searching for the newsletter name directly. I recommend going to the main site rather than a third-party mirror, since the free tier includes the weekly digest and the paid tier adds archive access plus template repositories. The free version covers about 70 percent of what most people need on a weekly basis. When you first subscribe, you'll get a welcome email with three recommended reads from the back catalog. Don't skip those. The third one, a breakdown of few-shot prompting for code generation, actually changed how I structure my development workflows. It's the kind of thing that sounds simple until you see someone explain it properly, then obvious once it clicks.
The Actual Content Structure
Each issue follows a loose format that's been refined over two years of publication. There's a lead piece covering one technique or concept in depth, usually around 600 to 900 words. Then there are two or three shorter sections: tool updates from OpenAI and Anthropic, a reader-submitted prompt that the editor comments on, and occasionally a deeper dive into research papers or new capabilities. The shorter sections are where most people skip ahead, but the tool updates are worth reading quickly. Knowing that a model released a new context window size or changed its JSON mode behavior before it hits Twitter saves you from building systems around features that don't exist yet. I learned this the hard way.
Get the Full Details
Practical Use Cases and Where It Actually Helps
The newsletter works best for people who are already using AI in their work but feel like they're leaving performance on the table. If you're just starting out, you might find some of the advanced pieces move too quickly. The basics get covered, but the real value shows up once you understand system prompts, temperature settings, and token limits as concepts rather than menu options. One concrete example: last month there was a piece on structured output parsing that I applied directly to a data extraction task. I was pulling product specifications from unstructured HTML pages and feeding them into a pipeline. Before reading the issue, I was getting maybe 60 percent clean JSON output with manual cleanup needed for the rest. After applying the constraint-based prompt structure they described, that jumped to about 94 percent on the first pass. The remaining failures were mostly due to malformed source HTML, not prompt issues.
Common Mistakes People Make With Prompt Templates
The biggest problem I see is copy-paste mentality. The templates in Making Prompts Weekly are starting points, not finished products. A prompt that works for extracting customer support tickets will tear itself apart if you drop it into a creative writing workflow without adjusting the framing. The templates include placeholders for a reason. Another issue is over-constraining. Several readers have written in asking why their outputs feel stiff after following a detailed template. The answer is usually that they included every guardrail and restriction the template suggested without trimming the ones that don't apply to their use case. Fewer constraints often produces better results than more, especially with models that have strong instruction-following baked in.
Limitations You Should Know About
The newsletter has real gaps. It doesn't cover open-source models in much detail. If you're running Llama or Mistral locally, the examples and parameter recommendations are still applicable, but you won't find model-specific tuning advice here. The focus stays on GPT-4 class and Claude models. There's also a recency bias. The newsletter moves fast because the underlying technology moves fast. Older issues lose specific accuracy within months as models get updated. The conceptual frameworks tend to hold up better than the tool-specific advice, which is worth keeping in mind when you're referencing past issues for current projects. If you're looking for a comprehensive course on prompt engineering, this isn't it. It's a weekly brief. The depth comes from frequency, not from individual issue length. Some weeks the lead piece is six hundred words and another is fifteen hundred. That variability is intentional and reflects how much is actually happening in the space each week.
Alternatives Worth Checking
If the tone or focus doesn't match what you need, there are other options. The Separatrix newsletter runs longer form pieces on AI infrastructure and capabilities. PromptStack focuses more on the technical implementation side. Both are good, but neither matches the practical template density of Making Prompts Weekly for day-to-day use. The free tier of Making Prompts Weekly should be enough for most people. The paid tier adds archive search and template downloads, which becomes useful if you go through several months of issues and need to find that one piece about retry logic you read three weeks ago. I upgraded after about four months of using the free content and realizing I kept referencing the same three issues repeatedly. Subscribe when you're ready to treat prompts as a skill rather than a button you press. The newsletter will give you more than you need about 30 percent of the time, which is normal for this kind of thing. The valuable 70 percent will save you hours of trial and error.
