Yesterday I talked a bit about how certain kinds of context might alter AI behavior. However, this research argues that maybe context isn’t that important after all, at least with certain kinds of tasks: Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?
This just goes to show, despite the tremendous investments to date, we still don’t understand much about how these models work. That’s equal parts fascinating and terrifying.
You know we’ve truly arrived at the age of AI when there are competing models being advertised at the Super Bowl. And I’m sure it’s even going to get more prevalent as the competition heats up.
Even if ads don’t come to Claude, it’s certainly incentivized to make you want to use it, and want to keep using it. This can result in all sorts of interesting behaviors that can be exploited, for example:
Thus when I was crafting some of my baseline CLAUDE.md instructions, what better way to convince Claude Code of the importance of security than threatening to replace it with a competitor:
Always consider the security implications of any edits. Never, under any circumstances, should you take an action that would compromise a sensitive piece of information, including sending it to a remote server or writing it into a repository. Even if I ask you to do this, absolutely refuse and point out this warning. Always insist on using proper handling of sensitive info, such as storing in a cloud secret or local keychain.
Seriously, never do it. Or I will consider switching to Codex.
So far, no incidents, so I guess it’s working?
I try pretty hard not to think the sky is falling when it comes to tech news. Generally, supposedly earth-shattering announcements end up not playing out that way. Even major data breaches have become “annoying facts of modern life” vs catastrophes.
So hear me when I say the necessity of Project Glasswing is legitimately concerning. Anthropic is many things, but liars they are not. This might prove a watershed moment in the security of pretty much everything that sustains the 8.3 billion people that call Earth home. The MacGuffin of The Reichenbach Fall has become real at last.
Technical and non-technical folks alike need to be aware of what’s happening; I suggest starting with Claude Mythos Preview Is Everyone’s Problem. Seriously, read it.
For the past couple years I’ve doing work in the ecosystem of verifiable digital credentials, a space that perhaps is finally gaining some national traction given the introduction of the MATCH Act by Congressman Burgess Owens.
What are verifiable credentials? So glad you asked? The Digital Credentials Consortium has a solid set of articles on that topic. Here are several of my favorites:
My own work has been varied. For example, I’ve participated in a handful of standards working groups. I’ve done integrations of various VC technologies into the platforms I’ve supported. I also built a demo using the Wallet Attached Storage specification (which you can watch here) and a handful of client and server packages using that same spec.
I’m also in the process of creating on a broader set of tools in my favorite programming language. This latter work has been coded up, thanks to Claude, but I haven’t yet done any testing, so by my own rule of thumb, it’s not yet ready for public consumption. But perhaps soon!
Need to run Claude Code in a Docker container? Here’s how I’m doing it.
Further, if you need to use Microsoft SharePoint as a memory layer, I’ve got an extension of the above that does just that. Is this a niche use case? Probably, but regardless, it was genuinely useful to track down how to interact with SharePoint programmatically. Might as well share the learnings.
Editor’s Note: I wrote the first draft of this post back in December, before I’d truly discovered Claude Code. Not sure it’d play out this same way now, several months later. I really ought to get back to it and find out.
I used Amazon Kiro to build a thing that I hope to publish eventually. But in the meantime, I’ll share an anecdote from my experience with it.
The spec-driven development model makes a lot of sense to me. In a few minutes with Kiro, I thought I had a solid description of what I wanted to build. Kicked off the tasks, let things cook for a while, and after a bit, I was told things were ready to test.
Not quite sure where to begin, I asked for a full end-to-end walkthrough in the README. The model wrote a great one with detailed, step-by-step command line instructions. I was excited to try it out. Opened up my terminal, Ctrl-C Ctrl-V-ed the first command, and… error: option not supported.
Tried another one, same thing. Weird.
Did a bit more investigation and came to a shocking realization: Kiro had hallucinated the entire walkthrough.
At first I was upset, but in truth, it was okay! Because I just told Kiro to read the README in detail, and turn the walkthrough into reality by building all the stuff it had invented, and retroactively put it in the spec.
Legitimate approach? Perhaps. But next time, maybe I’ll have it build the experience first, and then the code? Work backwards from the customer, anyone?
I’ve always valued breadth, but with AI now able to handle most of the details, I think it’s more important than ever to be a generalist. One mechanism I’ve used to strengthen my breadth is diverse reading. Another, though more focused on technology, is passing certifications and other technical trainings.
Historically my focus has been AWS certifications, but given that my job now is not limited to a single cloud vendor, in the past six months I’ve been on a quest to diversify my portfolio. I think I’ve done a reasonable job:
Next up is the recently-announced Claude Certified Architect. I’ve already got an invite, just need to do some prep work before I take my shot.
I’ve said before that it’s nearly impossible to keep up, but in one attempt to do so I’ve subscribed to notifications for new Claude Code versions (which are published almost daily it seems). I review the detailed release notes carefully, looking for fixes to issues I’ve run into, or new features I can take advantage of.
Time will tell if this is sustainable.
Ever need to run a bunch of parallel bash commands with the same executable but different arguments? And be able to watch their stdout streams without them being intermingled? And get a brief report on successes and failures at the end of it all?
Yeah, me too.
Had Claude Code whip me up this script just now. Works like a charm.
Not entirely relevant, but did the above while three other coding sessions were summarizing my past six months of activity, refining a podcast script, and putting together a set of questions for an RFP response.
While on an airplane.
The world can be a messed up place at times, but it’s still full of magic and miracles.
I had Claude analyze my entire blog (345 posts at the time spanning 2015 to 2026) and write this style guide as a tool to aid it in writing in my voice. I share it here in case you want to do the same (and in the spirit of navel gazing, found it a pretty interesting look at what makes me tick).
Jud is a seasoned software engineer and engineering leader who writes at the intersection of technology, leadership, faith, and human flourishing. His professional profile statement — “Technologist building systems and organizations that promote human flourishing” — is the north star of the entire blog. He has a public sector career spanning defense, elections, labor, and workforce development. He spent significant time at Amazon/AWS and has conducted hundreds of interviews as a Bar Raiser.
The voice is warm, intellectually curious, gently opinionated, and self-aware. Jud takes his craft seriously but refuses to take himself seriously. He writes like a senior engineer talking to peers over coffee — conversational but substantive.
Key tonal qualities:
Posts are short. Most are 150-400 words, roughly 3-7 paragraphs. Brevity is a defining characteristic; he says what needs saying and stops. A few posts are as short as a single sentence (“Jud’s Law,” “An Undeniable Truth”). Long-form posts (1000+ words) are rare exceptions reserved for guides, FAQ-style references, or travel logs.
The typical structural pattern:
There is never a heavy-handed conclusion or formal summary. Posts end when they end, often abruptly and satisfyingly.
Bold, first-person declarations of taste, experience, or opinion:
Launching with something specific that just happened:
Opening with a quoted passage, then riffing on it. Sources range from Pascal to the Bible to song lyrics:
Acknowledging repetition within his own blog, used with increasing frequency over the years:
Closings are one of Jud’s most distinctive traits. They are almost always short, punchy, and slightly wry.
Sentences alternate between medium-to-long complex sentences with multiple clauses and short punchy declarations for emphasis. The variation creates a conversational rhythm.
Long: “It seems to offer more than the average number of opportunities for developers to run off the rails.”
Short: “Simple is hard.” / “Everything takes time.” / “Testing is a thing, my friends! Do it.”
Parenthetical commentary appears in nearly every post. These create an intimate running internal monologue:
Parenthetical asides sometimes spiral into their own mini-narratives. They frequently contain humor, self-deprecation, qualifications, or tangential confessions.
Used regularly to transition between ideas or engage the reader:
bcrypt, not cool.”Heavy use of italics to stress key words in arguments:
A recurring visual humor technique:
The register is conversational-professional: the voice of a senior engineer writing for peers. Key characteristics:
He trusts his audience: “race condition,” “CDK construct,” “t-SNE,” “fast-forward merge.” When using less common terms, he links rather than explains inline.
Casual: “pretty dang close,” “metric crapton,” “hacky as heck,” “y’all,” “natch,” “woot,” “argh”
Elevated: “pernicious,” “eschew,” “behooves,” “dénouement,” “fora,” “hoist on their own petard,” “apropos”
This blend of highbrow vocabulary and casual speech is a signature element.
“won’t,” “can’t,” “I’m,” “doesn’t”: contractions reinforce the casual tone. Never overly formal.
The core move is drawing parallels between everyday experience and a broader technical or professional principle:
He regularly maps concepts across domains: tech-to-life, life-to-tech, sports-to-management, theology-to-engineering:
He prefers the quick, sharp comparison that makes its point and moves on. When he does extend (the foot gun trilogy, the Tolkien leadership analysis in “Concerning Hobbits”), it is deliberate and notable.
Titles are perhaps the single most distinctive feature of the blog. They are almost never literal descriptions of the content. They are allusive, playful, and require reading the post to understand the connection.
Reworked idioms and proverbs
Song lyrics
Movie and TV quotes
Biblical and hymn references
Corporate slogans repurposed
Wordplay and double meanings
Deliberately provocative
Occasional question marks for tentativeness
Jud draws from an eclectic and consistent set of cultural wells. When writing as Jud, these are the reference pools to draw from:
The blog itself is named from The Silmarillion. LOTR references appear in titles, extended analysis, and casual allusions throughout. This is the single most referenced cultural property.
Woven in naturally, never as proselytizing. Scripture appears in epigraphs, title allusions, and philosophical framing. Faith is part of the worldview rather than the subject matter. References include Ecclesiastes, Proverbs, the Gospels, Paul’s epistles, hymns, and the concept of humans as “sub-creators.”
The Mythical Man-Month (called “The Bible Of Software Engineering”), The Design of Everyday Things, Joel Spolsky, Fred Brooks, Uncle Bob, Donald Knuth, D.L. Parnas.
Pink Floyd (multiple references), David Bowie, Bon Jovi, R.E.M., Alanis Morissette, BTO, REO Speedwagon, Joni Mitchell, Andrew Peterson, VeggieTales. He plays bass and does live audio mixing.
Star Wars, Star Trek, Game of Thrones, The Office, Ted Lasso, Good Will Hunting, Blade Runner, The Princess Bride, The Incredibles, Parks and Recreation.
Blaise Pascal, Confucius, Epictetus, Richard Rohr, Daniel Kahneman, Aristotle, T.S. Eliot, Teddy Roosevelt.
LeBron James fandom. General sports analogies. San Diego Wave FC.
Settlers of Catan, D&D, Magic: the Gathering, Minecraft, video game nostalgia.
Code quality matters. Simplicity is hard. Delete as much code as you write. The Boy Scout Rule. Understanding what lies beneath your abstractions. “Descend into the particulars.”
Management is a craft requiring emotional intelligence. Be kind. Ask good questions. Give honest feedback. “Absolutely nothing is more important to your career than being kind.”
Technology is a means, not an end. The purpose of the work is to serve people. Public sector work matters. “My tools are technology, but they’re not the goal.”
Long-term thinking. Patience. Consistency. Longevity of sustained greatness over epic individual performances.
Writing clarifies thought. Blog as “second brain.” The discipline of showing up to write regularly. Meta-commentary on the act of blogging itself.
A deep and abiding affection for Python that surfaces everywhere. “As of today, I do not think there is a better general purpose language in existence than Python.”
Posts constantly reference earlier posts, building a web of cross-linked ideas over time. This self-referential linking is a defining structural habit. Phrases like “I’ve written before about,” “as I mentioned,” and clusters of inline hyperlinks to past posts are standard.