45 original guides across 5 shelves — the full index of everything on Fieldguide.
Understanding the actual mechanism behind AI writing tools changes what you ask them to do, and how much you trust what comes back.
For a structured perspective on using AI responsibly across different tasks, the NIST AI Risk Management Framework provides useful external guidance.
The skill AI writing tools displaced wasn't writing. It was staring at a blank page. A different skill matters more now.
Brainstorming is one of the few writing tasks where an AI tool's tendency to produce plausible-but-imprecise output is actually an advantage.
The same discussion also raises questions about transparency and workplace data; this reference provides related context for evaluating those trade-offs.
Default AI writing has a recognizable, generic quality. Getting past it takes more specific input than most people initially give it.
A model's context window is a hard, specific limit, and it explains a lot about why long AI-assisted writing projects get harder, not easier, as they grow.
A good AI summary and a misleading one can look identical in structure. The difference is usually in what got cut, not what got kept.
AI tools can genuinely speed up research. The speedup comes from a specific role in the process, not from treating the tool as the source itself.
AI tools made it dramatically cheaper to produce content at scale. That advantage has largely already been priced in, and not in the direction most publishers expected.
The difference between AI-assisted writing that stays genuinely yours and writing that quietly drifts out of your control is mostly about workflow, not willpower.
Understanding the diffusion process behind most AI image tools explains a lot about why prompts behave the way they do.
The specific, well-known weaknesses in AI image generation share a common underlying cause worth understanding directly.
AI video generation is genuinely capable and genuinely earlier-stage than AI image generation. Calibrating expectations to that gap matters.
Brand consistency is one of AI image generation's harder problems by default. A few specific techniques close most of the gap.
AI design tools genuinely lower the floor for non-designers. They don't raise the ceiling nearly as much, and confusing the two produces real disappointment.
Legal questions around AI-generated images remain genuinely unsettled in several jurisdictions. This is a general orientation, not legal advice.
Full-image generation gets most of the attention. Targeted editing tools are where a lot of the practical, everyday value actually lives.
The choice between stock photos and AI-generated images isn't a simple upgrade in either direction. Each has a specific, different set of trade-offs.
How you frame AI-generated work to a client shapes their expectations more than the work itself does. A few specific habits keep that framing honest.
Most no-code automation runs on a simple pattern. Understanding it makes both plain automation and AI-enhanced automation much easier to design well.
“AI agent” gets used to describe a wide range of genuinely different systems. Knowing which one a specific tool actually is matters before you rely on it.
AI-assisted support automation works best when it's honest about what it is, and clearly bounded about what it can actually resolve.
Most of the value of AI automation comes from connecting tools together, not from any single tool in isolation.
Debugging a no-code automation is a genuinely different skill from debugging code, and it's worth learning deliberately rather than picking up by accident.
Email automation is one of the most common first AI automation projects. It's also one of the easiest to get subtly wrong.
Scheduling is a genuinely good fit for AI automation, for a specific, structural reason worth understanding.
Turning one piece of content into several formats is one of the more genuinely reliable, high-value AI automation projects for a small team to start with.
Building an automation is a one-time event. Keeping it working is an ongoing commitment most people underestimate badly when they first build it.
Some prompting knowledge is durable. A lot of it is specific to a particular model generation and quietly goes stale.
The single highest-leverage change most people can make to their prompting isn't a trick — it's just being more specific.
The single interaction most people never learn to do well with AI tools is the second message in a conversation, not the first.
Showing rather than describing remains one of the most reliable ways to get a specific, consistent output format.
A single, sprawling prompt asking for too much at once tends to produce a worse result than the same work broken into a sequence.
A handful of specific, avoidable mistakes account for a large share of the frustrating first experiences people have with AI tools.
One of the more genuinely useful, underused prompting techniques is asking the AI tool itself to help you write a better prompt.
Building a small library of reusable prompts is one of the more underrated productivity habits for anyone using AI tools regularly.
The term gets used loosely. The specific mechanism behind it explains why it hasn't simply been fixed by newer, better models.
A new AI tool launches almost weekly. A short, consistent evaluation process is more useful than trying to test everything.
The gap between a free AI tool and a paid one usually isn't primarily about raw capability. It's worth knowing what the actual difference is before choosing based on price alone.
What happens to the text, images, or documents you feed into an AI tool varies significantly between products, and the details are usually checkable.
Adopting every promising new AI tool has a real, cumulative cost that's easy to underweight against each individual tool's apparent benefit.
Whether and how to disclose AI involvement in your work is an evolving norm rather than a settled rule, and worth a deliberate policy rather than an ad hoc decision each time.
AI bias is a real, well-documented issue. Understanding its actual mechanism helps you spot it in practice rather than treating it as an abstract concern.
A good AI tool stack for a small team looks different from a scaled-down version of what a large enterprise uses, and the difference matters.
Not every task benefits from AI assistance. A short, honest list of when the slower, unassisted approach is actually the better choice.
The pace of AI tool releases makes total awareness impossible. A deliberate, bounded approach to staying informed works better than trying to keep up with everything.