<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Bryan Lee Gregory on Medium]]></title>
        <description><![CDATA[Stories by Bryan Lee Gregory on Medium]]></description>
        <link>https://medium.com/@bryan_lee_gregory?source=rss-552f85099e8a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Bryan Lee Gregory on Medium</title>
            <link>https://medium.com/@bryan_lee_gregory?source=rss-552f85099e8a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 00:30:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bryan_lee_gregory/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[An experiment in writing analysis]]></title>
            <link>https://medium.com/@bryan_lee_gregory/an-experiment-in-writing-analysis-139151651bc0?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/139151651bc0</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[writing-style]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Tue, 31 Mar 2026 17:33:03 GMT</pubDate>
            <atom:updated>2026-03-31T17:33:03.037Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Two stick figures look at an AI-driven robot asking questions" src="https://cdn-images-1.medium.com/max/1024/1*HsiSCr8vRQNWRJHd8mlFmw.png" /></figure><p>There’s been a lot of virtual ink spilled over the last couple of years around how an author’s voice can get lost in AI — how over time it gets easier and easier to let LLM-generated wording choices take over as we build different habits.</p><p>I decided to run an experiment on ChatGPT to see what it identifies about my personal “voice” to see how well it matched up with my own expectations and ideals around my stories. I’m not particularly interested in the concept of “ask ChatGPT what it thinks about you”, but more “how would you identify whether a given article was written by me.” And hey — prospective future employers and partners, you can now get a sneak peek of my writing style!</p><p>Some notes:</p><ul><li><strong>I don’t run any of my Medium blog articles or novels through ChatGPT.</strong></li></ul><p>Before asking my prompt, I verified that ChatGPT doesn’t have access to my Medium stories. It’s using only my ChatGPT topics and questions for input. I *am* curious about how its answers might have changed if, in the future, I ran some of my more professionally-focused writing through it, but for now it (purposely) doesn’t have that source of input available. My writing that I publish is 100% my own; it has no LLM-generated input whatsoever (except where explicitly denoted, such as the analysis at the core of this article).</p><p>Similarly, I don’t run work-related emails, analysis, or other topics through this account — this is a collection of purely personal topics and conversations.</p><p>(as a side note, I interrogated ChatGPT in followups about its sources for its analysis below; it gave an interesting breakdown that’s generally believable around how it got to its responses without actually reading my articles)</p><ul><li><strong>My past ChatGPT conversations are on a wide variety of topics, but do include the <em>titles</em> of my Medium articles.</strong></li></ul><p>Some revealing patterns came out just from asking for AI-generated title images for my articles. My conversations generally fall into 3 categories:</p><p>Practical information research — “RSU tax implications Oregon”, “Copying formulas in Sheets”, “Resume analysis request”</p><p>Personal hobbies —“Reproducing specific shade of purple paint”, “Mushroom identification”, and “Head Protection in 1400s Germany”.</p><p>Creation requests (almost entirely image/art)-”Create a header image for an article” (this is where ChatGPT sees the topics I write about here), “Convert to sketch style”, “Image Creation Request”</p><ul><li><strong>Most of my conversations run through multiple related requests for refinement</strong></li></ul><p>It’s rare that I keep to a single question in a tightly focused conversation. There are pros and cons to this approach, but I find that I run a significant amount of <em>concept refinement </em>through my conversations, and some of them can wander significantly from their initial topic.</p><p>Here’s the prompt that I used:</p><blockquote>Review my previous conversations and give me a report on what specific aspects you consider part of my personal “writing voice”. What commonalities stand out? What would you use to “identify” that I wrote an article in the future?</blockquote><p>The response was very interesting. Just from the conversations I’ve had on ChatGPT’s platform, it pulled out a fairly significant set of points that I feel fit my personal voice well — certainly better than I expected without actually digging into my “real writing”.</p><p>The “one sentence summary” of my voice it generated:</p><blockquote>You write like a pragmatic systems thinker explaining how things actually work, gently correcting how people <em>think </em>they work.</blockquote><p>The main points it outlined were:</p><ol><li>Core Voice: Measured Contrarianism</li><li>Structured Thinking, Softened Delivery</li><li>Pragmatic Realism (Anti-Idealism Bias)</li><li>Focus on Incentives and Behavior (Not Just Ideas)</li><li>Calm, Non-Dogmatic Authority</li><li>Title Style: Conversational + Analytical Hybrid</li><li>Preference for Framing Over Prescription</li><li>Recurring Themes Across Articles</li><li>Subtle Empathy Without Sentimentality</li><li>Low “Performative Writing” Signal</li></ol><p>Each of these had a level of detail broken down below it (in a very LLM-specific pattern of framing, as you’d expect; you can find the full response at the end of this article). Overall, I was pleasantly surprised to see the AI-driven <em>description </em>of my voice from generally personal topics mirrors the goals that I set for my writing — low on the “clickbait” scale, illustrating balancing around real-world behavior instead of ideals, and focusing on calm, confident reframing of ideas.</p><p>Of course, that “pleasant surprise” comes with a tradeoff of its own. The response ended with:</p><blockquote>If you want, I can go a level deeper and:</blockquote><blockquote>Reverse-engineer your sentence-level patterns (syntax, cadence, vocabulary)</blockquote><blockquote>Or generate a “style guide” that someone else could use to imitate your voice convincingly</blockquote><p>That “style guide” suggestion outlines some real problems with LLM approaches, and leads to the discussions we’re now seeing around “voice” — how unique is your voice really, when an LLM can simply produce an imitation of it quickly and easily?</p><p>The reality is that, as we all know, humans can (and have) done very faithful imitations of “voice” and “tone” in the past without any help from LLMs. But lowering the barrier to entry here has its own consequences. Like any tool, it can be useful (<em>do you like my style? Here’s how to develop your own version of it into your work!</em>) or harmful (<em>Yeah, this article is *totally* by Bryan, and look at the garbage he’s spewing out these days! But it’s faithful to his voice, so you should believe it’s from him!).</em></p><p>One footnote:</p><p>1. I asked ChatGPT, as a followup, to create a suitable title image (seen above). I don’t actually like this one as much as I have the previous ones that I’ve more heavily refined and curated, but it seemed appropriate in this specific article to use the ChatGPT-generated one almost directly (I did expand it to a more suitable canvas size).</p><p>For the morbidly curious, I’ve reproduced the full ChatGPT response below. Now you too can write like me! Imitation, after all, is the highest form of flattery.</p><blockquote>I can give you a high-confidence analysis based on what you’ve shared about your topics, titles, and prior writing patterns in this conversation. Your titles alone are unusually revealing — they carry a lot of your voice.</blockquote><blockquote>What follows is a synthesis of your likely “writing fingerprint” — the traits that would let someone recognize your work even without your name attached.</blockquote><blockquote>1. Core Voice: Measured Contrarianism</blockquote><blockquote>You consistently write from a position that:</blockquote><blockquote>- Pushes back on prevailing narratives</blockquote><blockquote>- But avoids being inflammatory or absolutist</blockquote><blockquote>Examples from your titles:</blockquote><blockquote><em>- “A Partial Defense of AI Art”</em></blockquote><blockquote><em>- “People Won’t Deliver What You Ask For — And That’s OK”</em></blockquote><blockquote><em>- “Why Your ICs Need to Be Involved in Decisionmaking”</em></blockquote><blockquote>Signature trait:</blockquote><blockquote>You don’t argue <em>against</em> something — you <em>reframe it</em>.</blockquote><blockquote>Identifying marker:</blockquote><blockquote>If an article:</blockquote><blockquote>- Acknowledges a common belief</blockquote><blockquote>- Then carefully reshapes it rather than rejecting it outright</blockquote><blockquote>…it’s very likely your voice.</blockquote><blockquote>2. Structured Thinking, Softened Delivery</blockquote><blockquote>Your topics are managerial, technical, or organizational — but your tone avoids rigidity.</blockquote><blockquote>You tend to:</blockquote><blockquote>- Start with a premise or tension</blockquote><blockquote>- Walk through reasoning step-by-step</blockquote><blockquote>- Land on a practical, grounded conclusion</blockquote><blockquote>Signature trait:</blockquote><blockquote>- Logical scaffolding without sounding academic.</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- Clear internal structure (often invisible but present)</blockquote><blockquote>- Arguments unfold progressively, not rhetorically</blockquote><blockquote>3. Pragmatic Realism (Anti-Idealism Bias)</blockquote><blockquote>A strong throughline: systems don’t behave ideally, and that’s normal.</blockquote><blockquote>Examples:</blockquote><blockquote>- Hiring is messy</blockquote><blockquote>- Communication breaks down</blockquote><blockquote>- People don’t execute perfectly</blockquote><blockquote>- Tools (including AI) are uneven but useful</blockquote><blockquote>Signature trait:</blockquote><blockquote>You normalize imperfection rather than trying to eliminate it.</blockquote><blockquote>Identifying marker:</blockquote><blockquote>Phrases or ideas that:</blockquote><blockquote>- Accept constraints</blockquote><blockquote>- Emphasize tradeoffs</blockquote><blockquote>- Reject “silver bullet” thinking</blockquote><blockquote>4. Focus on Incentives and Behavior (Not Just Ideas)</blockquote><blockquote>You consistently anchor arguments in:</blockquote><blockquote>- How people actually behave</blockquote><blockquote>- Organizational incentives</blockquote><blockquote>- Real-world dynamics</blockquote><blockquote>Signature trait:</blockquote><blockquote>You don’t just discuss <em>what should happen</em> — you explain <em>what actually happens and why</em>.</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- Frequent references to roles (ICs, leadership, hiring managers)</blockquote><blockquote>- Analysis of behavior under constraints</blockquote><blockquote>5. Calm, Non-Dogmatic Authority</blockquote><blockquote>You write with confidence, but not dominance.</blockquote><blockquote>You:</blockquote><blockquote>- Avoid extreme language</blockquote><blockquote>- Rarely sound outraged or moralizing</blockquote><blockquote>- Don’t overstate conclusions</blockquote><blockquote>Signature trait:</blockquote><blockquote>“Quiet authority”</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- Few absolutes (“always,” “never”)</blockquote><blockquote>- Frequent nuance qualifiers (“often,” “in many cases,” “tends to”)</blockquote><blockquote>6. Title Style: Conversational + Analytical Hybrid</blockquote><blockquote>Your titles are especially distinctive.</blockquote><blockquote>They often follow patterns like:</blockquote><blockquote>- “X — Why Y”</blockquote><blockquote>- “X — And That’s OK”</blockquote><blockquote>- “X: Why You Should Pay Attention”</blockquote><blockquote>Signature trait:</blockquote><blockquote>A mix of:</blockquote><blockquote>- Casual phrasing</blockquote><blockquote>- Explicit analytical framing</blockquote><blockquote>Identifying marker:</blockquote><blockquote>Titles that:</blockquote><blockquote>- Contain a thesis</blockquote><blockquote>- Sound like a spoken explanation rather than a headline hook</blockquote><blockquote>7. Preference for Framing Over Prescription</blockquote><blockquote>You rarely say:</blockquote><blockquote><em>“Here’s the right way to do this.”</em></blockquote><blockquote>Instead, you say:</blockquote><blockquote><em>“Here’s how to think about this.”</em></blockquote><blockquote>Signature trait:</blockquote><blockquote>You shift mental models rather than give step-by-step advice.</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- Emphasis on perspective shifts</blockquote><blockquote>- Less emphasis on checklists or frameworks</blockquote><blockquote>8. Recurring Themes Across Articles</blockquote><blockquote>You seem to revisit a consistent set of domains:</blockquote><blockquote>A. Decision-making</blockquote><blockquote>- Deciders vs doers</blockquote><blockquote>- Ownership</blockquote><blockquote>- Involvement</blockquote><blockquote>B. Organizational dynamics</blockquote><blockquote>- Hiring</blockquote><blockquote>- Internal tooling</blockquote><blockquote>- Leadership tracks</blockquote><blockquote>C. Human limitations in systems</blockquote><blockquote>- Miscommunication</blockquote><blockquote>- Misaligned expectations</blockquote><blockquote>- Imperfect execution</blockquote><blockquote>D. Technology realism</blockquote><blockquote>- AI (balanced view)</blockquote><blockquote>- LLM coding (practical, not hype-driven)</blockquote><blockquote>Signature trait:</blockquote><blockquote>You explore <em>systems where humans and structure intersect</em>.</blockquote><blockquote>9. Subtle Empathy Without Sentimentality</blockquote><blockquote>You:</blockquote><blockquote>- Show understanding of people’s constraints</blockquote><blockquote>- But don’t drift into emotional or motivational writing</blockquote><blockquote>Signature trait:</blockquote><blockquote>Respectful realism about people</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- You explain behavior rather than judge it</blockquote><blockquote>10. Low “Performative Writing” Signal</blockquote><blockquote>You avoid:</blockquote><blockquote>- Clickbait</blockquote><blockquote>- Dramatic storytelling</blockquote><blockquote>- Overuse of anecdotes</blockquote><blockquote>Signature trait:</blockquote><blockquote>Content-first writing</blockquote><blockquote>Identifying marker:</blockquote><blockquote>- Minimal fluff</blockquote><blockquote>- Ideas carry the piece, not narrative theatrics</blockquote><blockquote>If I Had to “Fingerprint” Your Writing</blockquote><blockquote>If I were building a classifier to identify your work, I’d look for:</blockquote><blockquote>Structural signals</blockquote><blockquote>- Argument-driven essays</blockquote><blockquote>- Gradual development of a thesis</blockquote><blockquote>- Clean, logical progression</blockquote><blockquote>Linguistic signals</blockquote><blockquote>- Moderate sentence length</blockquote><blockquote>- Limited rhetorical flourish</blockquote><blockquote>- Controlled, precise wording</blockquote><blockquote>Conceptual signals</blockquote><blockquote>- Focus on tradeoffs</blockquote><blockquote>- Emphasis on real-world behavior</blockquote><blockquote>- Reframing of common assumptions</blockquote><blockquote>Tone signals</blockquote><blockquote>- Calm, measured, non-combative</blockquote><blockquote>- Confident but not declarative</blockquote><blockquote>- Analytical but accessible</blockquote><blockquote>One-Sentence Summary of Your Voice</blockquote><blockquote>You write like a pragmatic systems thinker explaining how things actually work, gently correcting how people think they work.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=139151651bc0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A (partial) defense of AI art]]></title>
            <link>https://medium.com/@bryan_lee_gregory/a-partial-defense-of-ai-art-1f65a3c57a29?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/1f65a3c57a29</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[novel-writing]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[art]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Thu, 12 Feb 2026 18:15:24 GMT</pubDate>
            <atom:updated>2026-02-12T18:15:24.955Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_OJ71QCmBOvj9esOeAnEaQ.png" /></figure><p>I’m going to preface this article with a warning — there’s some topics in this one that some folks might consider offensive, focused around theft, consent of use, and what constitutes originality. There’s some thoughts in here that <em>I</em> consider somewhat offensive. Fair warning has been given.</p><p>I was thinking yesterday about writing a post about traditional 3-tier enterprise architectures (database, backend, and frontend) and how specializations across one tier or another have blurred into cross-tier generalization as our technology stacks have grown and matured over the last decades. There’s a future evident here that’s related to LLM usage and learnings and how it’s changed our need for software expertise again.</p><p>Interesting to write about? Maybe. But at the same time it’s a topic that’s been written about incessantly by others better suited than I over the last couple of years. The article that came together in my mind was (at least to me) a clearly derivative work. That doesn’t mean it wouldn’t have value, and I might still write it in the future as an iterative work, but I don’t know that anything truly “novel” would have come out of it.</p><p>Later yesterday evening I had the pleasure of a long conversation with a good friend of mine on the topic of AI art and the ethics around it. Much of the conversation had to do with the topic of derivation versus originality — how much of human art is simply copy/pasting from other human art, and how much is truly novel work that required a leap from existing works to create? When thinking over my original idea for an article, it struck me that there’s a lot of parallels between the two topics.</p><p>I’ll first reference a <a href="https://theoatmeal.com/comics/ai_art">recent comic</a> by the esteemed Matthew Inman of “The Oatmeal”. In it he gives a cartoonist’s view of AI-generated art, including an entertaining takedown of “the people who are excited about AI art” (spoiler alert: He’s not kind in the rest of the quote). This already had my mind thinking about the topic, in addition to the fact that I’m using AI generated art in a <a href="https://medium.com/@bryan_lee_gregory/some-thoughts-on-llm-coding-268406763536">personal side project</a> (though whether those will stay long term or not is to be determined).</p><p>My conversation with my friend last night focused on 3 main AI art related topics:</p><h3>1. Building the corpus</h3><p>It’s no secret at this point that the majority of public LLMs were trained on materials without the owners’ consent, and even if you argue that it’s “fair use”, it’s certainly done without proper credit/attribution.</p><p>This is problematic, to say the least.</p><p>However, it hits home a little differently with me than with many. I’ve written <a href="https://www.bryanleegregory.com/books.html">some self-published novels</a> and continue to write as a hobby. I even sold a handful of copies, though not enough to even pay back the artist for the cover image. As part of my efforts, I put out ebook versions online of the novels in addition to producing physical copies. Because of this, there’s an extremely high likelihood that my work — that I poured an immense amount of time, effort and proverbial “blood, sweat, and tears” into — is part of the training materials for LLM models.</p><p>And… I personally think that’s <em>awesome</em>.</p><p>It’s somewhat awe-inspiring for me to think that my original works are (an infinitesimally small) part of what has helped build some of the most incredible technological achievement in the last 50 years. Somewhere, an LLM has produced work due to specific a specific word, phrase, or topic linkage from <em>my efforts</em>. That’s really, really cool to think about.</p><p>But I recognize I’m in the minority here. I have a number of friends who produce art professionally to one degree or another, and they’re pretty much universally pissed off that their work was <em>stolen without their consent </em>for the purpose of producing derivations from. And I get that, even if I don’t feel the same way personally.</p><p>For this initial point — the training data used — it’s pretty firmly in the “seriously problematic” zone, and I think there’s still a reckoning to be had in the future. It’s <a href="https://thewaltdisneycompany.com/news/disney-openai-sora-agreement/">already started</a> at the top of the pile, and continues to ripple downwards over time.</p><h3>2. Cost of usage</h3><p>AI has had <em>immense</em> resources poured into it. It’s no secret that datacenters produced almost entirely for the purpose of running LLMS are using resources at a hideous rate. The investment to date in this branch of technologies is staggering — I shudder to think how that investment might have been otherwise spent for the betterment of humanity.</p><p>And yet I remain optimistic. The initial investment into microprocessors was massive at the time. Luminaries in the industry thought it was a dead end, a technology that would go nowhere — yet today they’re ubiquitous, relatively cheap, and are used in a wide variety of ways to vastly improve the human experience.</p><p>I am hopeful that LLMs will go the same route. We’re already seeing progress — you can run a small model on your smartphone <em>today</em> that would have been inconceivable five years ago. We’re still in the R&amp;D phase of generative AI, and I have strong hopes that in 20 years we will have achieved efficiencies and value that today is difficult to imagine. I might be proven wrong, and we won’t have a Moore’s Law equivalent for GenAI models; but I think there’s a strong likelihood that the current growing pains are temporary.</p><p>That doesn’t make them any less painful today, and vast swathes of humanity have yet to see any payoff for the astronomical quantities of wealth poured into AIs yet. We live in a world where it can be a struggle to remain optimistic about the long term; I think it’s entirely reasonable to feel the pain today and cry out that these resources are being wasted or at best ill-apportioned.</p><h3>3. Novelty vs wholly derivative works</h3><p>This morning, driven by the conversation of the night before, I challenged myself to think of a truly novel image. The first thing that came to mind was a duck (why? Who knows!), and sure enough the first things that my brain put together were very familiar images — we have a small wetlands behind our house (I’m looking out at it now while typing this!) where we often see ducks. A picture of that is what came to mind initially, then images of ducks from when I was a child and we had some on our farm, then other images that I’ve seen before. While I’m sure there were minor errors in recollection, none of these were <em>new</em>.</p><p>So I tried to image a duck in a situation I hadn’t seen before. How about a duck in a bathtub? Well, I know what a bathtub looks like, and I know what a duck looks like. My mind first put a log in the tub too because I associate logs in lakes with ducks, thanks to the wetlands outside my back door. The result was surely a new image, something that hadn’t ever been done in quite exactly the same way before — but nevertheless it was <em>derivative</em>. Duck + bathtub + log — they’re all associations my mind made that strike me as very similar to what an LLM is doing when it builds an image.</p><p>Just like the article I originally planned, my brain was putting topics together in a variety of associations — but none of it was truly <em>creative</em>. It’s all copy and pasting in different ways.</p><p>Here’s the potentially offensive part: In my opinion, the vast majority of human art is <em>derivative</em>. It’s cutting and pasting things we’ve witnessed or observed and already associate together, in a way that’s <em>incredibly </em>similar to AI-generated art. With apologies to Schulz and Davis, it’s hard to argue that the latter half of their cartoons respectively were anything but derivative of their own work. There are only so many ways you can rationally have Garfield love lasagna and hate Mondays (to the point where Garfield parodies often center around just <em>how repetitive</em> the cartoons got).</p><p>LLMs have the advantage of being (compared to humans) the proverbial “infinite monkeys with typewriters”. They can produce and iterate incredible quantities of material and associations quickly. I love the books that I’ve written — but could an AI have authored them itself? Eventually, sure. Given enough time, effort, and iterations, I believe so. There are massive benefits to iterating in bulk — the ability to give an AI a prompt and ask for a hundred variations on a specific clipart goal, and iterate over and over on the top 5, can rapidly produce a result that rivals a human artist for the same goal.</p><p>I don’t think that’s inherently a bad thing, personally. I do think it will change the face of human art moving forward. I think “repetitive” art, or art that is <em>designed</em> to be derivative — which frankly is most of what’s used commercially these days - will be produced by AIs moving forward. And I think there’s a place for <em>novel </em>AI-driven art, which (though derivative of its corpus) is built from a sheer quantity of associations that humans couldn’t make in their lifetimes. Infinite monkeys, infinite typewriters, a new Shakespeare results.</p><p>That doesn’t mean human artists are going away, but they’ll have to change, just like the world of software development is changing rapidly. I think human artists will be forced to change their own style very differently than they have in the past. There’s artists who are famous for reinventing themselves — Madonna, for example — and there’s those that spend a lifetime making variations on their own style — the Jim Davises of the world. I think one of those types has a future — and the other will struggle in the upcoming flood of AI art that’s already begun.</p><p>It won’t stop me personally from writing. I enjoy the process, and — even knowing that the topics and stories I build are based on a lifetime of sources and observations, many of which are in common with what LLMs pull from — the odds of an LLM putting together the same string of things is vanishingly small. But it’s there. There remains a (very small) possibility I could read an AI-built novel tomorrow and realize it’s exactly what I would have written myself. I’m OK with that, in the same way that another human could potentially write what I’d have written as well.</p><p>I’ll leave this article with one more thought. LLM-driven art is, as of today, one hundred percent derivative. LLM knowledge comes entirely from human knowledge. AI art can make new links that we wouldn’t have thought of, but all of the basic information comes from us.</p><p>But that won’t be true forever. Someday soon, an LLM-driven machine will use sensors to observe information that <em>no human can ever have seen</em>. Imagine an interplanetary rover, equipped with efficient LLMs and a wide variety of sensors, witnessing something that humans have never before observed.</p><p>It will make its own <em>truly novel</em> associations. Instead of drawing from an entirely human-built corpus of knowledge, it will now <em>add to that knowledge</em>. Some specific association will be made to new information for the very first time ever, that humans could not have made ourselves at that point.</p><p>And when that happens, LLMs will no longer be making entirely derivative art. They’ll make something truly new. And we as humans will have to challenge our own understanding of ‘art’ and its value yet again.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1f65a3c57a29" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Some thoughts on LLM Coding]]></title>
            <link>https://medium.com/@bryan_lee_gregory/some-thoughts-on-llm-coding-268406763536?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/268406763536</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[game-development]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Tue, 10 Feb 2026 17:45:35 GMT</pubDate>
            <atom:updated>2026-02-10T17:45:35.200Z</atom:updated>
            <content:encoded><![CDATA[<p>In my downtime recently, I’ve been playing around with building a video game (let’s say it’s heavily inspired by an early 1990s game I loved, but attempting to update and extend it for the modern era). It makes for a convenient use case to brush up on my Java/React/SQL skills, play around further with AI-driven coding, and evaluate different approaches hands-on. I’m collecting some notes here so it’s a little more stream-of-thought than my usual articles. Top on my mind: Models continue to show significant differences, strengths, and weaknesses.</p><p>Most of my efforts so far are using Claude Code and Copilot in Visual Studio Code. I’ve tinkered with some other competitors as well, but these are the two I’ve used the most.</p><p><strong>Copilot:</strong> You get a lot of usage for free! Its inline suggestions are quite good and quickly fill out functions and help with repetitive data entry. On numerous occasions, I’d write a comment before starting a function, and Copilot would read the comment and magically suggest just the right code that I was thinking of (or something very close to it). Quite impressive in that context.</p><p>For bigger projects, though, it falls short. When using the Copilot Chat to plan through a bigger change, Copilot can flail… a lot. At one point it entirely screwed up a set of files and basically threw its hands in the air and apologized for messing them up. Fortunately I’m a stickler for proper code control even in personal projects and it was easy to revert.</p><p>When using Copilot for a bigger plan, the code it generated often missed some important context elsewhere in the code. I gave it quite detailed requirements and architectural expectations, but it still missed key points regularly.</p><p><strong>Claude Code:</strong> You don’t get much usage in a given session at the $20 USD/month Pro option. I’ll regularly run through my allocated session usage (resets every 5ish hours) within 30 minutes of an intensive session. I’ve learned to stretch this further with experience, and the more I can point to very specific files or usage the better — less context gets used up — but it still feels quite stingy compared to Copilot’s generous allocation even at the free level.</p><p>Claude isn’t as quick to suggest specific edits, but its overall high-level planning and implementation is <em>much</em> better than Copilot’s. I can give it the same detailed requirements and architectural expectations as I give Copilot, and the quality difference is stark. The generated code is significantly more thorough and takes the surrounding project into view much better.</p><p>Comparing the two, I find that I use Copilot regularly for small functions and inline code suggestions, and it’s quite good at those, but I’m using it less over time to sketch out bigger feature implementations. I hoard my Claude session allocations for those features, and it does a stellar job at putting together a first pass at a significant enhancement.</p><p>Both need human review and eyeballs to work properly. Both have given plenty of corner cases that I need to debug or fix up. Neither is a substitute for an experienced developer — yet.</p><p>This experience has made me greatly value my experience in software architecture and skills in leading projects. High-quality requirements and specifications are a must to get the most out of an LLM. It’s also driven home the necessity to leverage multiple approaches and understand the strengths and weaknesses of each competing model and company. The same is true when looking at generating placeholder art assets — ChatGPT, Grok, Midjourney, and others each have very clear strengths and weaknesses when interpreting my instructions to generate an image (my 2 cents: Grok falls short of the others by a significant margin — but it’ll let me bang my prompt against it hundreds of times in a row for free).</p><p>Models continue to evolve. Just in the last couple of weeks that I’ve been working on this project, Claude released a new model (and gave a bunch of free credits to try it out, which I appreciated!). I don’t expect the pace to slacken, and I’ll continue working with other models and competitors as I work through the bigger project on everything from code to art to (eventually) sound and music design. I fully expect to have to change my approach multiple times over the next few months as different models reveal new patterns.</p><p>I continue to worry about the impact on the software engineering pipeline. Developers <em>have</em> to continue to evolve their skills, but most importantly, we have to continue providing a pipeline for entry-level engineers to find their way in the field and grow to where they can oversee LLM-driven coding with experience under their belts. There’s a rapidly developing hole where new entrants don’t have the architectural experience in the real world to know when an LLM gets something structurally wrong, and fewer and fewer companies are giving them the chance to learn. Something’s got to give, or in 5 to 10 years we’ll have a massive gap in developer knowledge and no pipeline ready to fill it.</p><p>I doubt I’ve written anything here that hasn’t been explored more fully by others, but the experience continues to be illuminating — at my previous position I oversaw teams going through this same journey, and tinkered with it myself more directly, but there’s nothing like trying to implement an entire project from scratch that really drives home the pros and cons of these approaches.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=268406763536" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Internal tooling: How AI can improve everyone’s least favorite workflow]]></title>
            <link>https://medium.com/@bryan_lee_gregory/internal-tooling-how-ai-can-improve-everyones-least-favorite-workflow-4570a50d6b4a?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/4570a50d6b4a</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[workflow]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[internal-tools]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Mon, 02 Feb 2026 15:47:52 GMT</pubDate>
            <atom:updated>2026-02-02T15:47:52.578Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A robot working to fix a broken car" src="https://cdn-images-1.medium.com/max/1024/1*-bYMo_iTQCmomocAT5H-Ug.png" /></figure><p>Every reasonably mature company (and most that aren’t) end up with a conglomerate of miscellaneous tooling that was built to fit a “need of the moment”. Maybe it’s a script that creates fully-built test users in your database. Perhaps it’s the admin functionality — with a messy, we’ll politely say “minimalistic” UI — which allows you to change data parameters you don’t normally allow. Or it could be the supposedly short-term lookup tool used by your support personnel to answer critical questions from customers which was intended to be replaced years ago.</p><p>We’ve all seen them. Many of us have built or expanded upon them. They end up in repositories nobody has the keys to, or running in corners of our network that are poorly documented. They’re often in languages that the rest of our codebase moved on from or never even used. They’re poorly maintained because it’s always hard for leadership to justify spending more time and money on them. They’re a SOC nightmare and everybody complains about the access rights and documentation thereof when time for your audit rolls around.</p><p>At the same time, these tools are often some of the greatest sources of friction for your teams. The technical team hates working on them as much as they complain about their necessity. Non-technical support personnel absolutely rely on their functionality, and it can grind things to a halt when nobody pays enough attention to a critical tool that breaks inadvertently. There’s good reason they were built in the first place, and that reason often stretches on into perpetuity.</p><p>The need for oft-neglected internal tooling isn’t going away. I’ve had a copy of <a href="https://xkcd.com/1205/">this xkcd comic</a> about “how much time is it worth spending to automate a function?” hanging on my office wall for years; it’s a great way to visualize (and explain) the tradeoffs involved in upgrading and maintaining tooling for manual tasks. I’ve evangelized the importance of internal tooling and its maintenance, but it’s always an uphill struggle to justify prioritizing it over customer-facing (and potentially income-generating) work instead.</p><p>Fortunately for all of us, we have good news in this area. Here’s the thing: The UIs, testing, and documentation for these internal tools are often <em>already terrible</em>. They’re rarely put together with any sort of intentional design. Similarly, internal tooling often neglects basic audit and validation work.</p><p>Well, guess what — LLMs are <em>fast </em>at this sort of thing.</p><p>The reality is that “vibe coding”* often creates messy code behind the scenes, or misses some elements of a design, or needs a ton of work put in to create high-quality test suites. For these types of sadly neglected internal tools, however, <em>anything</em> can be an improvement. If your boutique, designless half-broken UI in a language and framework nobody on your team knows completely falls over and someone needs to fix it, is it really introducing a risk to ask your AI assistant to take a shot at coding a new one from scratch? Getting to that 80% or 90% threshold quickly in a language or framework that matches the rest of your infrastructure can be a <em>far </em>better investment than trying to detangle existing spaghetti code that survives on hopes and prayers at the best of time.</p><p>In many examples, this becomes a case of “quick attempts with AI can’t really make it much worse.” Embrace that fact! Use your old internal tooling as the poster child for learning when and how to effectively use LLMs to add to your codebase — not just as an assistant, but as a <em>primary coding mechanism</em>. Humans will still need to be involved in ensuring functionality, specifying requirements, and building guiderails, which is where the industry is headed anyways. When internal tooling often takes a significant time investment for little gain due to missing standards and documentation, maybe it’s time to leverage that existing “brokenness” and make it a feature, not a bug.</p><p>It’s also a way to start building up your clear “definitions of work” for your LLM for other work along the way. Maybe your team hasn’t built out a standard claude.md (or the equivalent) that could cover your existing projects, and this is a good way to start. Use these internal tools — especially where they’re already broken and needing some serious polishing up or complete rewrites — as your starting ground for giving LLMs free reign to start a new project or transform something from codebase A to codebase B. This is one time where you can unironically ask yourself “can it really get any worse?”</p><p>Have you tried throwing a pile of ugly internal tooling code at an LLM yet? How did it do to rebuild it? What tradeoffs did you find, and was the end product better than what you started with?</p><p>*For today, I’m going with a <em>very rough </em>definition for “vibe coding” of “asking your LLM assistant to do work without having strong system architecture and design documentation in place, overseen carefully by a human”. Sheafs of virtual paper have already been spent on the topic and carefully defining the phrase isn’t really the point here.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4570a50d6b4a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[New year, new prospects!]]></title>
            <link>https://medium.com/@bryan_lee_gregory/new-year-new-prospects-2553288c0f97?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/2553288c0f97</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[people]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Mon, 05 Jan 2026 19:57:39 GMT</pubDate>
            <atom:updated>2026-01-05T19:57:39.127Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A group of stick figures celebrating the new year" src="https://cdn-images-1.medium.com/max/1024/1*fDomC60hNpmxuW6yvAfj2w.png" /></figure><p>As we enter 2026 I’ve been thinking about how both people and technology “roll over” into the new year. Spirits rise and tech changes — it’s a regular cycle that we should celebrate, not dread, as we (metaphorically) flip our digital calendars to look into something new.</p><p><strong>People:</strong> Hiring picks up in Q1. Companies are considering their yearly plans or starting to execute on changes they put together at the end of the previous year. In some cases, yearly budgets get refreshed — individuals are asked to put together plans for expansion, training, travel, and the like. It’s the best time of year to be looking to hire or get hired, before fortunes change and budgets get clamped down.</p><p>More ephemerally, spirits generally go up at the start of a new year. This ranges from “New Year Resolutions” (that people haven’t yet dropped out of) to hopes of wiping the board clean and making a fresh start. Economic indicators often rise, especially if Q4 results come in well. It’s common for individuals or companies to take time off at the end of the year, leaving everyone ready to come back refreshed and ready for something new.</p><p>Finally, there’s plenty of companies that either delay or freeze hiring and spending late in the year, pushing it into the start of the new year. I talked about this <a href="https://medium.com/@bryan_lee_gregory/hiring-patterns-why-you-should-pay-attention-456d44938d4e">in my last post</a>, and I’ll remind everyone that it’s worth paying attention to what your own company is doing this time of year to get ahead of the game.</p><p>What about reviews? Fewer companies hold to the concept of the “yearly review” cycle these days, but there’s still plenty who do. It’s not uncommon to see annual reviews and refreshes happen early in the year; if you’re in this group, make sure you’re prepared and have a complete recounting of 2025 for you and your org ready to go when it’s needed.</p><p><strong>Technology: </strong>After the slowdown for end of year holidays, people come back into action looking for change at the start of a new year. This is especially true in the technology landscape.</p><p>Looking over the last couple of years, we’re firmly past the beginning of LLMs making a major impact on every aspect of technology. There’s a number of knock on effects from this, of which many others have written more thoroughly — the end of the “AI honeymoon”, continued evolution of LLMs, ever-increasing costs, and rapid change on which model is “best” for your particular usage.</p><p>In the development landscape, at the start of 2026, it’s pretty well expected that you’ll have LLMs participating in your process. Maybe it’s something as simple as integration of one or more models into your code development process as “helpers” — in many ways this can be as small as an evolution of code completion tools that have been embedded in our IDEs for years. Maybe you’ve moved onto something bigger — you might have included automated code review steps in your pipelines, for example, or have entirely LLM-driven test suites by now. No matter where you are in your process, though, we’re at the phase where nobody wants to be “left out”.</p><p>At the same time, productivity gains are being questioned and examined with more scrutiny. If you’re still just “playing around” with AI-assisted development, expect to be asked for data in 2026. Can you measure a difference pre-AI tools and post-AI tools? Can you demonstrate how much these added costs have lowered your development time, improved your resiliency, or improved deployment iteration? Metrics matter, and everyone needs to be prepared to show data on why their choices are the right ones moving forward.</p><p>No matter what you’re looking at in the new year, I hope everyone had a great 2025 and looks forward to a better 2026!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2553288c0f97" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hiring Patterns: Why you should pay attention]]></title>
            <link>https://medium.com/@bryan_lee_gregory/hiring-patterns-why-you-should-pay-attention-456d44938d4e?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/456d44938d4e</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[hiring]]></category>
            <category><![CDATA[careers]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Wed, 19 Nov 2025 20:25:02 GMT</pubDate>
            <atom:updated>2025-11-20T16:12:31.899Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NPNOSURsRO2qoEDHvMO9Eg.png" /></figure><p>I’ve done a decent amount of hiring in my career (and a lot more asking for hires!) across a number of different positions, primarily Software Engineering. Even in a ‘stable’ company there’s still boom and bust cycles to be aware of, and a savvy people manager should pay attention to what that means for their own asks.</p><ol><li><strong>When asking to add someone to your team, do your homework first.</strong></li></ol><p>Pretty much any people manager in the Software world wants to add people to their team. There’s always more work to be done, always more goals and KPIs to hit, and always more tech debt that your team <em>could</em> be tackling.</p><p>Here’s the thing, though: You <em>should</em> always have more work to do than people to do it (though by a reasonable amount!). Not all of the work in your team’s queue is <em>actually</em> vital; it’s important for every level to be able to balance needs of the moment and overall business requirements. Sometimes projects get cancelled or moved around; if you don’t have a runway in front of you, your team is stuck until you can rebuild that backlog. In the same vein, if you don’t know what’s coming a few weeks or months down the line, you’ll have a difficult time making long-term architectural decisions to support where your work is <em>going</em>, not just where it’s currently at.</p><p>If you talk to your boss and report that your team is perfectly 100% balancing work and time, you’ll probably get a skeptical look and some hard questions about why you haven’t built up a useful backlog to look forward to.</p><p>So — back to the original topic- before asking for more people, you should <em>always</em> know what exactly you want those people to do. More than that, you should have very clear data supporting your ask. Will adding someone to your team increase your capacity? Prove it by showing that more work could be done in parallel. Will adding someone to your team add a missing skill? Show why that skill is vital and can’t be learned (or learned quickly enough) by someone on the team today. Use hard data to show why a team add is vital to the company interest, not just your own — what specific company objectives will adding this person help with?</p><p>Finally, always remember that your team not meeting deadlines doesn’t necessarily mean that you need more people. You’ve better make damn sure that your ask isn’t instead showing that you’re just not managing the team efficiently enough.</p><p>2. <strong>Be aware of your company’s planning process and budget.</strong></p><p>This is a mistake I made as a newly minted manager. I didn’t pay any attention to the company’s planning. I knew it happened, but only vaguely; and I always thought “that’s way above my pay grade”.</p><p>I was wrong.</p><p>How your company operates should be in everyone’s view all the time. When does your company give out promotions and compensation changes? That’s too late to be thinking about the process, because decisions about those are often made months beforehand. When is your company most open to hiring — after quarterly results are in? After the calendar year is over?</p><p>It’s very common for little hiring (and a number of layoffs) to happen in Q4 as companies work to make their balance sheets look as good as possible going into the new year. I can’t count how many times I got told to cancel or delay requisitions as we came into Q4.</p><p>Always assume that other managers, and other departments, <em>are</em> keeping track of hiring and firing cycles within your company and the industry. Be open with your manager about what you’re looking for and give them reasons to support you — and keep you up to date with the right time and place to officially put in your ask.</p><p>3. <strong>Keep track of the industry.</strong></p><p>If everyone in your industry is hiring in a ‘boom’ cycle, you’re going to have to wait longer to get a quality candidate in even if you get approval for a req. However, the reverse isn’t always true — big companies might do a wide round of layoffs, but that doesn’t mean you necessarily want those candidates.</p><p>How will your company tolerate a long hiring cycle? Would they rather get someone more junior in who can be trained up as a long term investment, or do they prefer senior folks who can hit the ground running? What’s the industry availability — and pricing — on the type of people you want?</p><p>Skills can also be problematic — even if the industry in the whole is at a low point, specific specialties might prove hard to find. Pay attention to what it takes to upskill your current team. It might be easier — and more valuable- to give someone on your squad time and flexibility to learn something new rather than trying to hire an expert, especially if it’s in a rapidly developing field.</p><p>What hiring tips and tricks have you discovered in these areas? Are there company-specific quirks you’ve run into? Have you ever made a mistake in your asks, and what did you learn from it?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=456d44938d4e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[People won’t deliver what you ask for. And that’s OK.]]></title>
            <link>https://medium.com/@bryan_lee_gregory/people-wont-deliver-what-you-ask-for-and-that-s-ok-6f66a92eee4c?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/6f66a92eee4c</guid>
            <category><![CDATA[delegation]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[process-improvement]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Mon, 13 Oct 2025 21:02:50 GMT</pubDate>
            <atom:updated>2025-10-13T21:02:50.430Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*52rBa3N4zfLKPly7nCl_tA.png" /></figure><p>Even before I “officially” took on a leadership role, there were plenty of times where I worked to lead a project as an individual contributor (IC). In this role, I’d work up a design for a project, do a review with the team and update it based on feedback, then describe what I had in mind for each task and divvy out the work. I’d then find myself getting frustrated that the beautiful image I had in mind for what I’d get from each participant didn’t measure up to what I expected. Pretty normal stuff for anyone leading a group project, whether it’s in school, a hobby group, or at work, right?</p><p>It took me years to learn that the problem wasn’t really that things weren’t being done “right”. “Right”, in this case, was all in my head. Sure, there were quality issues here and there - the name of a method or table might have a typo in it, a critical unit test might have been missed, or some documentation didn’t get finished. Those problems were easy to talk through and address. What really got to me, and what took so long for me to understand, was when the work differed in some fundamental way from what I’d expected to get, and that <em>pissed me off</em>.</p><p>What took me so long to realize is that I was looking for something to be angry about. I blamed the other people for not understanding. I blamed them for disagreeing. Above all, though, I blamed myself for not communicating well enough. Somehow, I thought, I should have been able to take the vision in my mind and magically stick it in everyone else’s brain exactly the same way it appeared to me.</p><p>As you might expect, that’s not a particularly healthy view to have of workflow.</p><p>Communication skills are always ripe for improvement. You can work on documentation until your fingers bleed — or these days we can ask LLMs to help polish it up, which really can help. Maybe you can draw better diagrams (I <em>love</em> drawing on whiteboards!). No matter what, though, we’ll never truly meet others on a mind-to-mind basis. Something always gets lost. So what are we to do?</p><p>I’ve learned to embrace those differences that crop up in practice. Sometimes, they’re an opportunity to teach — for a concrete example, let’s say a junior developer builds out an API that only takes a single item and calls it N times for a collection of those items. This can be an opportunity to talk about the overall use case and how to expand a perspective to cover what‘s really needed. Other times, what actually got built might give a chance to consider whether it might be an improvement on your original design. Is it something that might be useful in the future, or for another use case that’s being built out? Don’t just throw it away because it’s not what you thought you needed at the start of the project.</p><p>Finally, every difference is a chance to improve those communication skills. If you don’t find out that what’s being built doesn’t meet your needs until the <em>end</em> of a project, you’ve definitely got some opportunities to improve communications along the way. If something doesn’t come across as you expected, don’t blame the other person for misunderstanding; instead, focus on how to make it better next time. Nobody’s perfect, but we can all strive to get closer to an ideal.</p><p>So next time that you ask someone to do something, think about -</p><ul><li>How do you verify that what’s coming across is what you have in mind? Ask followup questions and confirm.</li><li>How do you check in and ensure that the output will match what you asked for? You’ve got to find the right balance between not enough and too much.</li><li>At the end of the process, what can you do to improve it on the next go-round? Retrospectives are great as an idea, but often don’t get taken seriously or utilized to their full potential.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6f66a92eee4c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deciders and Doers: Why your ICs need to be involved in decisionmaking]]></title>
            <link>https://medium.com/@bryan_lee_gregory/deciders-and-doers-why-your-ics-need-to-be-involved-in-decisionmaking-eb26640a5ee1?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb26640a5ee1</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[decision-making]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Wed, 24 Sep 2025 17:18:00 GMT</pubDate>
            <atom:updated>2025-09-24T17:18:00.358Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*16Q333KJC-9Cb6mXLVIb2w.png" /></figure><p>In a <a href="https://medium.com/@bryan_lee_gregory/the-leadership-track-why-go-there-eefbb9e773dc#4834">previous post</a>, I wrote an aside about ICs (individual contributors) needing to be involved and invested in decisionmaking. From a leadership perspective it’s easy to forget or gloss over <em>why</em> it’s important over time. Few people would directly argue against that view, but I’ve seen plenty of cases where leaders lost that ability to directly connect with and understand why it’s vital for your ICs to remain part of your decisionmaking process.</p><p>The topic involves some complicated and at times conflicting inputs. The “deciders” (leadership across cross-functional teams; this might be anyone from the C-team to Engineering Managers and Product Managers) often have different information than the “doers” (ICs, the people who are actually building your product requirements and writing code). Maybe they know about different priorities coming down the line. Maybe they have a better understanding of overall company needs. We have a variety of levels in a corporate hierarchy for a reason (though there’s been plenty of words published on where those levels help or hurt; there’s something to be said for flatter structures but they can also struggle in this area); at the end of the day no one person should be expected to have all the context of every project across every part of a modern company.</p><p>That often translates to the “deciders” making a decision and pushing it down to the ICs without giving the full context of the decision. For example, let’s say Team A has 7 ICs and a full slate of priorities/projects. Somewhere in upper management, it’s determined that due to an expensive vendor contract ending next year, we need to devote some resources to migration away from that vendor to save significant $$ in the future. Perhaps we’re also (confidentially) building a better strategic relationship with a competing vendor. The decision trickles down through leadership layers until the manager of Team A is told “you need to give up 2 of your ICs for the next 6 months to this decommissioning work”.</p><p>That manager knows that some nebulous ‘decommissioning’ is happening. They might even know that it will “save some money”. But from their perspective, their own slate of projects is suddenly thrown into disarray for a reason they don’t really have the full context on. The ICs on their team know even less; they just know that suddenly two of their teammates are yanked to work on other projects and they have to clean up the mess while “doing more with less” (a phrase I <em>abhor</em> for a number of reasons — but maybe that’ll be the topic of a future post!)</p><p>As a result:</p><ul><li>Team A’s ICs and manager are demotivated and frustrated</li><li>Team A’s existing priorities and projects are thrown into disarray</li><li>The ICs who are moved might be curious about new projects, but they’re not happy about having left things half-finished, or might get interrupted by their prior teammates over and over for their knowledge</li></ul><p>What’s missing here? Two-way communication and expectation-setting. The ICs, and the manager of Team A, weren’t involved in the decision, which means they’re not invested in the results.</p><p>In theory, somewhere along the line, Team A’s priorities were examined and it was determined that they were the team who could most “afford” to have two ICs moved out. The problem is that — even if that’s true — they <em>never had the chance to speak on their own behalf</em>. That’s the “involved” part.</p><p>For every step you get in a hierarchy further from the ICs, you lose some context about exactly what’s happening “on the ground”. Part of a manager’s responsibility is to help summarize and advocate for their team or teams. In this case, if they’re involved at the start of the process, they can (and should!) write up what exactly the effect would be of pulling two ICs out of the team. A skilled manager should be able to quickly summarize the consequences.</p><p>Where I’ve seen things fall short is when people get <em>defensive</em> about their priorities and projects. A manager who gives a comprehensive, honest response about exactly which priorities and timelines will have to shift in this case will be in a better spot than a manager who says the team “can’t afford” to lose people and gets hyperbolic about the results. The reality is that everything has consequences, and good leaders know that; they just might not have all the context that the ICs themselves might have in this case.</p><p>Billions of dollars every year are spent to track projects and try to tie timelines, resources, dependencies, and milestones together. The reality is that something is <em>always</em> missed. Maybe an IC just discovered a bug that will cause another team’s deliverable to be delayed. Maybe a dependency link was missed for tracking because a key resource left before adding it in. Maybe your senior IC who’s been working on a service for years is the only one who actually understands that making decision X will result in Y consequences (which is its own problem, but again, a discussion for another time).</p><p>The other half to consider is <em>investment</em>. This translates to how much people care about what they’re working on. ICs who are always told what to do, and never feel like they have a chance to understand the context and/or contribute to discussions, will run out of motivation quickly. It might show up as lower productivity. It might show up as missed KPIs. It might reveal itself in exit interviews when you have higher turnover.</p><p>The reality is that everyone wants to feel they have a voice, however small. ICs — especially more experienced ones- generally understand that they don’t have a ton of sway in decisions that come down from above. That’s OK, as long as their voices are heard when they have something important to contribute. A leader who misses a crucial bit of information that was held by their teams, and makes the wrong decision because of it, will kick themselves down the line for not consulting where they should have.</p><p>The flip side of the argument is <em>speed</em>. You can’t afford to consult every person on every decision — that way lies madness! Leaders are <em>supposed</em> to know enough about their teams and projects to estimate impact of changes like this; that’s part of their job. But nobody’s perfect, and mistakes get made.</p><p>All this comes down to a balancing act. Leaders have to decide how much to consult downwards to get more details; sometimes there isn’t the opportunity to do so because of a need for speed, or sometimes the information critical to understanding a decision <em>can’t</em> be revealed below a certain level for one reason or another. Leaders who have built trust by reaching out when it <em>is</em> an option, and sharing context <em>as much as possible</em> to explain why decisions are being made, will get a pass from their teams when they can’t communicate as fully as they’d like.</p><p>When have you run into this balancing act? How have you dealt with the need for involvement and investment vs confidential information or a need to move a decision quickly?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb26640a5ee1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Leadership Track: Why go there?]]></title>
            <link>https://medium.com/@bryan_lee_gregory/the-leadership-track-why-go-there-eefbb9e773dc?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/eefbb9e773dc</guid>
            <category><![CDATA[leadership-development]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[management-and-leadership]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Wed, 03 Sep 2025 19:45:43 GMT</pubDate>
            <atom:updated>2025-09-03T19:45:43.148Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BBD_FaLiP8278xsoaQVVXw.png" /></figure><p>I spent the first ten years of my career avoiding the management side of the world like the plague. There were a few reasons for this, ranging from a <a href="https://medium.com/@bryan_lee_gregory/visibility-and-self-promotion-20a13ba43a74">lack of self-confidence</a> to an <a href="https://medium.com/@bryan_lee_gregory/decisiveness-and-ownership-6b7f7518b9f4">avoidance of accountability and ownership</a>. Like pretty much all of my peers, there was always a part of me that believed I’d make better decisions than the people above me in the hierarchy, but the thought of actually sticking my neck out and putting my own decisions on the line felt like a risk without any payoff.</p><p>Even more than this, though, I didn’t like the thought of <em>leading people</em>. My vision of management was dealing with constant HR nightmares, clashing personalities, disparate priorities, and unrealistic expectations. Why on earth would I want to walk into <em>that?</em></p><p>Finally, there was a real fear about putting my fate into the hands of others. If I wrote my own code, I was responsible only for that small fenced garden of work. If someone else on the team failed, well, that was their fault, right? (Note that my perspective on this changed <em>drastically</em> over the years!) But if you’re in a management position, you’re now beholden to everyone on your team — if they fail, you fail; you’re no longer solely in control over your own destiny. That thought was <em>terrifying</em>.</p><p>Let me give some context for those who aren’t familiar with the space. Software engineering careers traditionally go one of two directions: individual contributors (ICs) or management, i.e. the “Leadership Track”. There’s a dizzying array of IC skills and specialties, but at the end of the day even when they’re leading projects or pushing for change they have a direct and real impact on progress in those initiatives. They write code, they push fixes, they review changes, and at the end of the day it’s your ICs who make for the real success or failure of any given project.</p><p>ICs <em>don’t</em> generally have people reporting up through them, and they’re <em>usually</em> not responsible for the decisions affecting the direction of their org. There’s always exceptions and places where a particular personality can blur the lines, but in an <em>incredibly </em>oversimplified view you can think of most ICs as “doers” and most folks on the Leadership track as “deciders”. [Note: if you actually take that mindset to its full conclusion you’ll fall on your face, as ICs <em>need</em> to be involved and invested in decisionmaking, but that’s a topic for another time — we’re simplifying for the sake of explanation here!]</p><p>Back to where I was. I’d gotten to the position of Senior Engineer and was happy day to day writing architecture documents, pumping out code, finishing up sprints, and generally doing my own vision of what an IC should aspire to do. Sure, over a few years I started a lot of mentoring, helping with hiring, directing my team to how our development sprints should run, and so forth, but that was just a side part of things that I didn’t focus on. Or at least, that’s what I told myself.</p><p>At the time, I had an amazing manager and mentor who I still chat with regularly even now, many years later. He prompted me a few times to consider the leadership track, but I kept putting him off saying that I really didn’t want to <em>manage people</em>, because in my own mind it came with all the annoying stuff I noted above without any real benefits. Finally, one day he sat me down and essentially told me that he was officially putting me in charge of the team. I was pretty hesitant until he pointed out that I was already effectively managing the team without it being in my title, and that he thought that despite my reservations I would be a pretty good people leader.</p><p>One of my first “official” management tasks was to tell someone they were being let go. It wasn’t pleasant by any means, but for as much as I’d always feared approaching anything of the sort, I found that — to my surprise — I handled it OK. I did what I could to take care of the guy; I said the right things, I made sure the right steps were being taken, and was able to walk away from it without feeling completely devastated about it.</p><p>Starting from that low point, my management experience could only go up from there. I’d always enjoyed mentoring, but suddenly I was able to <em>do something</em> when I felt a person was ready to reach that next level — whether that meant being given a project to lead or a new skill to master, a new title, or a raise. I’d always felt that my decisions would be better than other peoples’; now I had to put up or shut up and actually deliver on them. As much as I’d been dreading that responsibility for years, when I was able to lead the team to its first successful delivery entirely under my watch I felt an incredible sense of accomplishment. Suddenly I wasn’t just delivering my own “stuff”, I was bringing about the success of <em>an entire team</em>.</p><p>I also realized that my manager was completely right — I’d been doing a lot of the “people management” side of things already, just not thinking of it that way. I knew what motivated my peers and what type of work they were good at or liked or disliked; now that they reported to me, it meant that I could help steer things towards or away from what was really needed there. As I put it now, even though my manager had to drag me kicking and screaming onto the leadership track because he saw my potential, I’ll always thank him for making me take the plunge.</p><p>Since then I’ve come to love the leadership track. Some of my most rewarding work experiences have been helping people find their way to success and watching them grow. It can still be “exciting” (in several senses of the word) to put my name on a decision that can have a significant real-world effect on not only myself but my entire organization or company, but watching those decisions pay off brings a real feeling of impact that the IC track just doesn’t compare with. It took me a couple of years to really realize it, but I can make a much more powerful impact by improving a hundred peoples’ efficiency, skills, and productivity by 10% than by working as a high-powered IC even if I could somehow be a dozen times more productive than the person sitting next to me.</p><p>I’ve also recognized that this is <em>not</em> a career for everyone. Some people are far, far more effective as ICs than as leaders due to their skill set or personality. Others might be interested in leadership, but don’t want to take on that responsibility (and risk!) of having to deliver large projects across multiple squads and potential points of failure. Still others don’t want to deal with the tougher parts around prioritization, “translating” between different business functions, or the ever-present risk of HR problems lurking behind the scenes. These are all entirely valid reasons. Lastly, simply by nature of the job, there aren’t as many leadership positions as there are IC positions; there’s just less <em>opportunity</em> from that perspective.</p><p>Story time over! What attracts you to one side or the other (IC vs leadership track)? How have you found those lines blurred or crossover positions? What successes have you found on your own terms?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eefbb9e773dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Work and hobbies: why I’m writing a PRD (but don’t code) in my off time]]></title>
            <link>https://medium.com/@bryan_lee_gregory/work-and-hobbies-why-im-writing-a-prd-but-don-t-code-in-my-off-time-6493a6443a81?source=rss-552f85099e8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/6493a6443a81</guid>
            <category><![CDATA[skills-development]]></category>
            <category><![CDATA[documentation]]></category>
            <category><![CDATA[work-life-balance]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Bryan Lee Gregory]]></dc:creator>
            <pubDate>Wed, 06 Aug 2025 21:05:07 GMT</pubDate>
            <atom:updated>2025-08-06T21:05:07.880Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A stick figure between icons for “work” and “hobbies”" src="https://cdn-images-1.medium.com/max/1024/1*u7qHmHCeVlFE0Ow9Jb-S6A.png" /></figure><p>This might be a cardinal sin to admit, but I don’t do a lot of coding in my off time. I’ll dabble at it here and there where I find interest, but in general I try to keep my “work related” skills focused on work and my “hobby related” skills to the other part of that nebulous “work life balance”.</p><p>Part of this is very practical. I learned early on in my career that if I try to practice the same skills constantly in both work and non-work time I run a high risk of boredom. It’s easy to “code too much” or “mentor too much” or so forth and take the magic — or enjoyment, if you’d prefer a more mundane description — out of all of it. I’m a better, more productive person when I can really focus and engage with a set of skills within a well-defined work/life boundary.</p><p>At the same time, I pride myself (delude myself?) on having a strong generalist set of skills. I lead engineers, but I think there’s a powerful advantage in being able to “translate” between different business functions — and that means understanding business, finances, product, analytics, and all the other functions that make a modern software company tick.</p><p>In one of my hobbies, the group supports a complex piece of software that has grown and evolved somewhat organically over time from a group of volunteers (sounds a bit familiar to open-source contributors, doesn’t it?). Along the way, it’s become a behemoth that nobody really understands the full functionality of. Whenever the group talks about updating it or (gasp) rebuilding it with more modern framework, architecture, and language choices, the reality is that there’s currently no way to fully describe what we’d actually need from a replacement. Estimates thus vary wildly from “this is a weekend’s worth of work” to “this will take two years and thousands of dollars we don’t have”.</p><p>That’s why I’ve found myself writing a PRD (Product Requirements Document, for those unfamiliar with the term; basically a document that describes what we want a product — in this case the software application at the heart of this hobby — to do). There’s a few reasons I put myself in this position:</p><ul><li>Although it’s “career adjacent” in that I regularly weigh in and give feedback on such documents, I haven’t actively <em>authored</em> a PRD in a long time. That frees up the “don’t overdo a work task in your free time!” concern I note above.</li><li>It’s interesting! I love learning deeply about new things, and along the journey of discovery that’s taking place I get to think about improvements and how things could be better organized (and add comments to that point).</li><li>It’s valuable! This document will go a long ways towards helping shape the path of how the hobby runs moving forwards, and it’ll serve as a reference document that’s been sorely needed for some time.</li><li>It’s a useful skill! I really enjoy writing (though usually in the form of genre fiction), and this helps polish up some skills that really can be valuable in the workplace. <strong>I’m a firm believer of “don’t critique what you can’t do yourself”, and if I’m going to give solid feedback to <em>other peoples’ </em>PRDs, I better be willing to write one myself and put it out for feedback.</strong></li></ul><p>Right now I’m about 20 pages into the document, and perhaps a quarter of the way through the functionality. Still a long ways to go. But at the end of the day it’ll make for a better foundation for my hobby group as well as help me brush up on some skills I like to keep sharp, so it seems like a win-win all around.</p><p>How do you separate your work and non-work time in regards to skills? Are you someone who can code all day at work, then go home and do it again and love it?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6493a6443a81" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>