AI Practices
Just as there are many different types of writing, so too are there many ways to work with gen AI content based on the goal, audience, and use case of the text.
Updated July 20, 2026.
The more the writing demands that I sound like my unique, human self, the less AI will be involved.
For interpersonal writing with friends and family, I don’t use AI at all. I use my heart and my head, as nature intended.
For creative writing, where it’s meant to be my voice with the addition of an audience who might not personally know me—think blog posts, Substack articles, or that book I keep meaning to write—I use AI as a workshopping partner.
This means I’ll present it with the goal of the piece and any other relevant parameters, show it my drafts, and ask for objective feedback, which I may or may not accept. Objective meaning any blatant grammatical or typographical errors, or anywhere I might be missing the mark of my stated goal. The actual editing is done by me in a google doc. Claude operates within a framework I've established for this purpose.
For personal branding, I’m still at the creative wheel, while the workshopping goes deeper.
My process for personal branding, like my LinkedIn bio and this website, is much the same as for my creative writing, only with more extensive workshopping.
In these contexts, some semantic ablation is acceptable, given just how wordy and poetic I can be. I’m still doing the writing and editing myself, but I’m considering Claude’s more subjective feedback and acting on it when I agree.
For these use cases, Claude helps me color inside the lines, while I choose what hues and shades I use.
As we get into the technical side of things, AI takes on more writing work while I take on the role of orchestrator.
Technical writing is a science, unlike the more artistic disciplines of creative and copywriting. In tech writing, rules are king. The aim is consistency and clarity, and with that comes the very homogeneity at which LLMs largely excel.
It’s in this space where AI becomes the greatest boon for my work, allowing me to expedite laborious tasks like organizing and standardizing, so that I can quickly dial in on making sure knowledge is intuitively structured, concepts are effectively conveyed, workflows are properly validated, and information gets into the hands of the users who need it now, not later.
Tools like Atlassian’s Loom are able to almost entirely eliminate the tedious grunt work of documenting step-by-steps. With tools like Claude and ChatGPT, brain dumps become documentation as if by magic. And with my quality assurance expertise and deep understanding of what makes technical writing work, I can focus on accuracy and strategy as I shape the end result.
I believe scrupulously vetted, well-structured documentation is the foundation of accurate AI outputs.
Now is the time for organizations to have documentation in place that AI can pull from to offer accurate answers to human queries.
If the documentation doesn’t exist, AI can’t reference it. If it exists but it isn’t complete, correct, or AI-ready, LLMs are more likely to hallucinate to fill in any gaps.
Any knowledge base that I’m responsible for is going to undergo meticulous, intentional, and strategic QA review to ensure it’s doing its part to make LLM responses as factually sound as possible.
Preventing the spread of misinformation is a core value I hold, and documentation is where I can make the biggest impact when it comes to AI.
What’s my overall attitude toward gen AI?
Like many people, my introduction to gen AI was ChatGPT in late 2022. I used it for much of my personal and professional work throughout 2023-2025, often impressed with its ability to help me do things I couldn’t have done without spending frustrated days sifting through online forums to understand things like VBA Macros or how to make a simple diff checker.
However, it was when I tried to use ChatGPT for tasks that I did know how to do—writing, editing, proofreading, even simple math ledgers or Excel formulas—that I realized just how poorly it handled those tasks. I found myself then scrutinizing every output and challenging every hallucination, to such an extent that it was less time-consuming and aggravating to just do the tasks myself.
In late 2025 I switched from ChatGPT to Claude, which is a much more sane experience than working with ChatGPT. However, Claude isn’t impervious to hallucinations, and as of July 2026 seems to be getting worse every day.
Copilot was equally vexing as ChatGPT, helping me achieve things I couldn’t have done on my own, while completely butchering the things I could do on my own but had hoped would be expedited by using Microsoft’s built-in AI.
In my corporate life I was tasked with testing Atlassian’s tools, Rovo and Loom. I found Rovo to be infuriatingly obtuse, but Loom to be quite a game changer, even in its Beta version.
As a Content QA responsible for reviewing text, audio, and video assets created using various gen AI tools, I can say with my whole chest that any time saved upstream creates double or more time lost in the QA phase downstream, and that entire workflows would need to be reworked to account for this reality.
AI evangelists will insist it’s all a matter of writing foolproof prompts. I can’t help but wonder though, if machine psychology is something we’re all required to master in order to reap real benefits from gen AI, is it truly as democratizing as they make it out to be?
More importantly, with LLMs being trained on materials created by humans and their responses rated by humans, working with AI has become essentially the same as working with a human (derogatory). AI carries the same implicit biases, erroneous assumptions, logical fallacies, and proclivity for misreading subtext as humans do, but with the shield of being un-fireable and ultimately having no accountability for the damage its inaccuracies cause. It’s a glaring double standard when someone handing off the same egregious outputs pre-AI would be disciplined, but when it comes from an LLM it’s accepted without question.
In an ideal world, AI would be an objective source of rigorously verified truth. The current reality as of 2026 is AI is at least as flawed as humans, makes more errors than humans, and causes real harm to people and organizations who don’t have robust QA processes in place for gen AI output, yet humans (and QA processes in general) are being increasingly cut out of the loop.
As you can see, my feelings on the topic are a mixed bag. Ultimately though, what I think doesn’t matter, because AI is here. The toothpaste is out of the tube. There’s no putting it back. In order to survive and stay relevant, we must embrace working with AI tools, and so I do.
And to that end, I think the most important thing any of us can do is to stay grounded in the reality of the tools’ strengths and limitations as they develop, keep our expectations reasonable, and work with AI accordingly. Prompt engineering is a vital skill, of course, but a perfect prompt won’t make a model do something it’s simply not able to do, and someone with skills and experience will always need to check the work.
tl;dr: Keep humans doing the work only humans can do, let the machines do the rest, and always have a human being reviewing gen AI output for accuracy and quality.
Note: All em dashes on this page were lovingly handcrafted by me, using my artisanal, free-range English degree.