<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en"><generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator><link href="https://ylan.segal-family.com/atom.xml" rel="self" type="application/atom+xml" /><link href="https://ylan.segal-family.com/" rel="alternate" type="text/html" hreflang="en" /><updated>2026-11-03T15:03:53-08:00</updated><id>https://ylan.segal-family.com/atom.xml</id><title type="html">on.code &amp;amp;&amp;amp; such</title><subtitle>Thoughts on Software Engineering</subtitle><author><name>Ylan Segal</name></author><entry><title type="html">The REPL: Issue 146 - October 2026</title><link href="https://ylan.segal-family.com/blog/2026/11/03/the-repl-issue-146-october-2026/" rel="alternate" type="text/html" title="The REPL: Issue 146 - October 2026" /><published>2026-11-03T14:50:53-08:00</published><updated>2026-11-03T14:50:53-08:00</updated><id>https://ylan.segal-family.com/blog/2026/11/03/the-repl-issue-146-october-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/11/03/the-repl-issue-146-october-2026/"><![CDATA[<h3 id="ai-promised-less-work-i-have-never-worked-more"><a href="https://oioannou.com/blog/ai-promised-less-work/">AI Promised Less Work. I Have Never Worked More</a></h3>

<p>Orestis Ioannou points out that he doesn’t have less work in the age of AI. Neither do I. The work is different, but I’m not sitting around drinking coffee. I never took the promise of less work seriously anyway. If anything, what I can do by myself has expanded.</p>

<h3 id="living-in-the-weimar-era-of-ruby"><a href="https://blog.julik.nl/2026/10/the-weimar-of-ruby">Living in the Weimar era of Ruby</a></h3>

<p>The title is clickbait, and the article is not what you imagine. I’m linking to it because it points to this exchange, which I found insightful.</p>

<p>Nate Berkopec:</p>

<blockquote>
  <p>I just think you need to have some epistemic humility
Most important development in computing in 20 years comes along, and the exact stack you’ve been working on that whole time happens to be the perfect stack for the next 20 years? OK.</p>
</blockquote>

<p>José Valim:</p>

<blockquote>
  <p>You are right but it also feels short-sighted? You have the most powerful tool you ever had, you have become a 1000x maker, and you can’t think of how to make your stack 10x better?</p>

  <p>Yes, Rust is fast, the type system catches bugs, but there is so much beyond it.</p>

  <ul>
    <li>Why not make Ruby 10x faster?</li>
    <li>Why not make Ruby (and therefore Rails) really great at concurrency?</li>
    <li>Why not add a banging type system to Ruby that can provide guarantees beyond Rust?</li>
    <li>Why not embed SQLite3 into Rails with a replication protocol so you can really compress the stack beyond a single server?</li>
    <li>How to preemptively fix most bugs from the most critical parts of Rails, such as database pools?</li>
  </ul>

  <p>Yes, who cares about ActiveSearch, but there are so many ways we can fundamentally improve our stacks that go beyond asking the agent to just crap out code. Phoenix is still the only framework that ships with presence, because it turns out it is deceptively a hard problem to solve correctly, but somehow all issues are solved?</p>

  <p>I understand some folks have no interest in doing any of the above, and that’s fine, but then at least pass the baton to the people who want to do exactly that.</p>

  <p>PS: I say all of this with love ❤️</p>
</blockquote>]]></content><author><name>Ylan Segal</name></author><category term="ruby" /><category term="machine learning" /><summary type="html"><![CDATA[AI Promised Less Work. I Have Never Worked More Orestis Ioannou points out that he doesn’t have less work in the age of AI. Neither do I. The work is different, but I’m not sitting around drinking coffee. I never took the promise of less work seriously anyway. If anything, what I can do by myself has expanded. Living in the Weimar era of Ruby The title is clickbait, and the article is not what you imagine. I’m linking to it because it points to this exchange, which I found insightful. Nate Berkopec: I just think you need to have some epistemic humility Most important development in computing in 20 years comes along, and the exact stack you’ve been working on that whole time happens to be the perfect stack for the next 20 years? OK. José Valim: You are right but it also feels short-sighted? You have the most powerful tool you ever had, you have become a 1000x maker, and you can’t think of how to make your stack 10x better? Yes, Rust is fast, the type system catches bugs, but there is so much beyond it. Why not make Ruby 10x faster? Why not make Ruby (and therefore Rails) really great at concurrency? Why not add a banging type system to Ruby that can provide guarantees beyond Rust? Why not embed SQLite3 into Rails with a replication protocol so you can really compress the stack beyond a single server? How to preemptively fix most bugs from the most critical parts of Rails, such as database pools? Yes, who cares about ActiveSearch, but there are so many ways we can fundamentally improve our stacks that go beyond asking the agent to just crap out code. Phoenix is still the only framework that ships with presence, because it turns out it is deceptively a hard problem to solve correctly, but somehow all issues are solved? I understand some folks have no interest in doing any of the above, and that’s fine, but then at least pass the baton to the people who want to do exactly that. PS: I say all of this with love ❤️]]></summary></entry><entry><title type="html">The REPL: Issue 145 - September 2026</title><link href="https://ylan.segal-family.com/blog/2026/10/06/the-repl-issue-145-september-2026/" rel="alternate" type="text/html" title="The REPL: Issue 145 - September 2026" /><published>2026-10-06T16:18:00-07:00</published><updated>2026-10-06T16:18:00-07:00</updated><id>https://ylan.segal-family.com/blog/2026/10/06/the-repl-issue-145-september-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/10/06/the-repl-issue-145-september-2026/"><![CDATA[<h3 id="kafgres-embedding-a-kafka-broker-into-postgres"><a href="https://rynr.dev/blog/kafgres/">Kafgres: Embedding a Kafka Broker into Postgres</a></h3>

<p>Interesting project: a Kafka broker built into Postgres as an extension. I’ve considered Kafka for several projects, but I’ve yet to use it.</p>

<h3 id="its-time-to-investigate-the-ai-labs"><a href="https://calnewport.com/its-time-to-investigate-the-ai-labs/">It’s Time to Investigate the AI Labs</a></h3>

<p>Cal Newport is wondering the same thing I am:</p>

<blockquote>
  <p>Why not pursue criminal liability for them running systems they knew would likely commit crimes?</p>
</blockquote>

<p>There are already laws on the books against unauthorized access to computer systems. I haven’t heard of any authorities investigating the AI labs for the hacks they’ve already made public. Why not? The labs directed their computers to do something on their behalf, and that something is a crime. Investigation seems like the logical next step. Did rubygems.org call the authorities after discovering their infrastructure was being abused? <a href="https://blog.rubygems.org/2026/09/11/update-may-spam-publishing-campaign.html">Their update</a> doesn’t say.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="postgres" /><category term="machine_learning" /><category term="security" /><summary type="html"><![CDATA[Kafgres: Embedding a Kafka Broker into Postgres Interesting project: a Kafka broker built into Postgres as an extension. I’ve considered Kafka for several projects, but I’ve yet to use it. It’s Time to Investigate the AI Labs Cal Newport is wondering the same thing I am: Why not pursue criminal liability for them running systems they knew would likely commit crimes? There are already laws on the books against unauthorized access to computer systems. I haven’t heard of any authorities investigating the AI labs for the hacks they’ve already made public. Why not? The labs directed their computers to do something on their behalf, and that something is a crime. Investigation seems like the logical next step. Did rubygems.org call the authorities after discovering their infrastructure was being abused? Their update doesn’t say.]]></summary></entry><entry><title type="html">The REPL: Issue 144 - August 2026</title><link href="https://ylan.segal-family.com/blog/2026/09/05/the-repl-issue-144-august-2026/" rel="alternate" type="text/html" title="The REPL: Issue 144 - August 2026" /><published>2026-09-05T12:19:08-07:00</published><updated>2026-09-05T12:19:08-07:00</updated><id>https://ylan.segal-family.com/blog/2026/09/05/the-repl-issue-144-august-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/09/05/the-repl-issue-144-august-2026/"><![CDATA[<p>Today’s REPL is different from most. I usually point to a few interesting articles. This time I found myself writing at length about a single one.</p>

<h3 id="accepting-a-messy-git-history"><a href="https://beza1e1.tuxen.de/git_two_users.html">Accepting a messy git history</a></h3>

<p>Andreas Zwinkau describes two kinds of git users: commit often, or rebase carefully. I don’t see it as a choice between the two. It’s more nuanced.</p>

<p>I do both. While developing, I commit whenever I reach a meaningful amount of work <sup id="fnref:1"><a href="#fn:1" class="footnote" rel="footnote" role="doc-noteref">1</a></sup>. Sometimes that’s failing tests that expose a bug, so the red-green cycle lives in the history. Sometimes it’s a class or module the feature needs but that isn’t hooked up to the rest of the system yet. It’s a thoughtful process, but it doesn’t always leave a pristine history.</p>

<p>Before I open a pull request, I go through that history. Sometimes I took a path and later backtracked. A file might exist only in this branch because I decided the naming needed work. None of that helps anyone later, so I rebase and squash to present a better story in the pull request.</p>

<p>Once I open a pull request, I don’t rebase, squash, or otherwise change the history. Not even squash merges. Once reviewers have commented, rewriting history creates more work for them. GitHub presents a “changes since your last review” button, which is extremely useful. A force-pushed rebase or squash destroys it, and the reviewer has to read the whole PR again. I care more about reducing friction with my teammates than about a pristine git history.</p>

<p>Is a pristine git history even worcth having?. Most arguments I’ve come across say a good history gives historical context about a code base. I agree, to a point. Commit messages that all say “fix code” help no one. But a squash-merged history <em>loses</em> the changes reviewers asked for. They live in the pull request, not in git. So does a history rewritten to look like the code arrived fully formed. I prefer the actual history to an abridged version.</p>

<p>As for stacked pull requests, I think a lot of folks miss the point. They <em>can</em> split a large change into smaller chunks for review, but in practice I use multiple PRs when I need to sequence deployments. For example, a PR that prepares the code for a column deletion, then a stacked PR with the migration that deletes it. Sometimes a conceptual unit of work has to ship in multiple PRs. That’s different from a commit history that tells a story but can be deployed all at once. As for the “read the commits individually” argument: I’ve never been on a team that does that. Maybe because GitHub doesn’t encourage it, but whatever the reason, I don’t see it as a practical way to review PRs.</p>

<div class="footnotes" role="doc-endnotes">
  <ol>
    <li id="fn:1">
      <p>This has changed somewhat with the use of agents. <a href="#fnref:1" class="reversefootnote" role="doc-backlink">&#8617;</a></p>
    </li>
  </ol>
</div>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="git" /><summary type="html"><![CDATA[Today’s REPL is different from most. I usually point to a few interesting articles. This time I found myself writing at length about a single one. Accepting a messy git history Andreas Zwinkau describes two kinds of git users: commit often, or rebase carefully. I don’t see it as a choice between the two. It’s more nuanced. I do both. While developing, I commit whenever I reach a meaningful amount of work 1. Sometimes that’s failing tests that expose a bug, so the red-green cycle lives in the history. Sometimes it’s a class or module the feature needs but that isn’t hooked up to the rest of the system yet. It’s a thoughtful process, but it doesn’t always leave a pristine history. Before I open a pull request, I go through that history. Sometimes I took a path and later backtracked. A file might exist only in this branch because I decided the naming needed work. None of that helps anyone later, so I rebase and squash to present a better story in the pull request. Once I open a pull request, I don’t rebase, squash, or otherwise change the history. Not even squash merges. Once reviewers have commented, rewriting history creates more work for them. GitHub presents a “changes since your last review” button, which is extremely useful. A force-pushed rebase or squash destroys it, and the reviewer has to read the whole PR again. I care more about reducing friction with my teammates than about a pristine git history. Is a pristine git history even worcth having?. Most arguments I’ve come across say a good history gives historical context about a code base. I agree, to a point. Commit messages that all say “fix code” help no one. But a squash-merged history loses the changes reviewers asked for. They live in the pull request, not in git. So does a history rewritten to look like the code arrived fully formed. I prefer the actual history to an abridged version. As for stacked pull requests, I think a lot of folks miss the point. They can split a large change into smaller chunks for review, but in practice I use multiple PRs when I need to sequence deployments. For example, a PR that prepares the code for a column deletion, then a stacked PR with the migration that deletes it. Sometimes a conceptual unit of work has to ship in multiple PRs. That’s different from a commit history that tells a story but can be deployed all at once. As for the “read the commits individually” argument: I’ve never been on a team that does that. Maybe because GitHub doesn’t encourage it, but whatever the reason, I don’t see it as a practical way to review PRs. This has changed somewhat with the use of agents. &#8617;]]></summary></entry><entry><title type="html">The Backlog Is Infinite</title><link href="https://ylan.segal-family.com/blog/2026/08/22/the-backlog-is-infinite/" rel="alternate" type="text/html" title="The Backlog Is Infinite" /><published>2026-08-22T13:47:24-07:00</published><updated>2026-08-22T13:47:24-07:00</updated><id>https://ylan.segal-family.com/blog/2026/08/22/the-backlog-is-infinite</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/08/22/the-backlog-is-infinite/"><![CDATA[<p>A friend asked me if my company is going to fire half of my team. He doesn’t write software. A few weeks earlier he had installed <a href="https://claude.com/product/claude-code">Claude Code</a>, described an app he wanted, and got one. More or less. If he can do that knowing no programming at all, what does a company need a whole team of engineers for?</p>

<p>I get some version of this question regularly now, almost always from friends outside of software. I’m not worried. But not for the reason they expect.</p>

<!-- more -->

<p>The arithmetic is simple enough. Say we have a team of ten engineers, and AI makes each of them twice as productive. They now do the work of twenty. Fire five, keep the output you had before, save five salaries. That “twice as productive” is doing a lot of work, and I don’t take it as a given. Let’s assume it’s true anyway.</p>

<p>The reasoning makes sense if engineering is a cost center.</p>

<p>If software is something the company buys to keep the lights on – the internal tools, the reporting, the integration nobody outside the building will ever see – then it’s overhead. There’s a fixed amount of it you need, and every dollar you don’t spend on it is a dollar you keep. If each electrician can now keep twice as many lights on, you need half as many electricians. That isn’t a foolish manager talking. That’s just what a cost center is.</p>

<p>I have no doubt that some managers are running exactly this calculation right now, and that some of them are right.</p>

<p>Then there are the other companies, the ones that sell what engineering produces.</p>

<p>Same payroll, more output, more product to sell. You don’t lay off half of a factory line that’s running at a profit. You run it harder. A company that fires five of ten profitable engineers to save the salaries has voluntarily halved something that makes money.</p>

<p>The test isn’t quite “does the company sell software,” though. It’s much broader in practice: <em>does better software sell more of what we already sell?</em> A company shipping hardware with software inside passes – better software moves more boxes. So does the retailer whose app is how people actually buy things, and the bank where the deposits arrive through a phone. Stated that way, the number of engineering teams that are pure overhead is a lot smaller than my friends imagine.</p>

<p>What actually convinces me isn’t the economics. It’s what sprint planning looks like. So, to my friend’s chagrin, that’s what I explain.</p>

<p>My team has a backlog. For practical purposes, it is infinite. There is never a shortage of things to do: features customers want, bugs we know about, optimizations we keep meaning to get to, improvements we’ve been talking about for a year. Every sprint we sort by what would bring the most value, estimate how much we can actually finish, and draw a line.</p>

<p>If we overestimate, work spills over. It’s usually at the top of the next sprint, though not always – priorities move. If we underestimate, we have room left, and we pull in the next thing below the line. Notice what it doesn’t have: a state where we run out of work. I have never seen that list run dry.</p>

<p>AI did change real things. Projects that weren’t worth their cost before are worth it now, so there’s appetite for more ambitious work. More tasks fit in a sprint. We may be shipping fewer bugs, because our review process is more thorough than it used to be.</p>

<p>What didn’t change is the line. We still draw it every sprint, and there is still more below it than we can get to.</p>

<p>Which brings me back to the app my friend built.</p>

<p>He thinks he demonstrated that engineers are optional. What he actually demonstrated is that there was software worth building that nobody was building, because at the old price it wasn’t worth anyone’s time. Now it is, so he built it himself. Multiply that by everyone with an idea and no budget, and demand for software went up, not down.<sup id="fnref:1"><a href="#fn:1" class="footnote" rel="footnote" role="doc-noteref">1</a></sup></p>

<p>There’s an honest limit to this. It covers companies where software sells something, which is most of them, but not all. If you’re on the internal tools team at a company whose product has nothing to do with software, none of my reasoning protects you, and I don’t have a comforting version of this post for you.</p>

<p>For the rest of us, the backlog is infinite, the work above the line is worth more than it costs, and getting more of it for the same money is a good trade for everyone involved.</p>

<p>So no, I am not worried. I hope I’m right.</p>

<div class="footnotes" role="doc-endnotes">
  <ol>
    <li id="fn:1">
      <p>Economists have a name for this shape: the <a href="https://en.wikipedia.org/wiki/Jevons_paradox">Jevons paradox</a>. In 1865 Jevons noticed that Watt’s far more efficient steam engine did not reduce Britain’s coal consumption. It increased it, because cheaper coal-power made coal worth burning in places it had never been worth burning before. I’m borrowing the shape of the argument, not claiming the result. Coal is a fuel and engineers are not, and whether the rebound in demand for software outruns the efficiency gain is a genuinely open question. <a href="#fnref:1" class="reversefootnote" role="doc-backlink">&#8617;</a></p>
    </li>
  </ol>
</div>]]></content><author><name>Ylan Segal</name></author><category term="machine_learning" /><category term="productivity" /><summary type="html"><![CDATA[A friend asked me if my company is going to fire half of my team. He doesn’t write software. A few weeks earlier he had installed Claude Code, described an app he wanted, and got one. More or less. If he can do that knowing no programming at all, what does a company need a whole team of engineers for? I get some version of this question regularly now, almost always from friends outside of software. I’m not worried. But not for the reason they expect.]]></summary></entry><entry><title type="html">The REPL: Issue 143 - July 2026</title><link href="https://ylan.segal-family.com/blog/2026/08/07/the-repl-issue-143-july-2026/" rel="alternate" type="text/html" title="The REPL: Issue 143 - July 2026" /><published>2026-08-07T16:28:03-07:00</published><updated>2026-08-07T16:28:03-07:00</updated><id>https://ylan.segal-family.com/blog/2026/08/07/the-repl-issue-143-july-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/08/07/the-repl-issue-143-july-2026/"><![CDATA[<h3 id="the-short-leash-ai-coding-method-for-beating-fable"><a href="https://blog.okturtles.org/2026/07/short-leash-ai-method/">The Short Leash AI Coding Method For Beating Fable</a></h3>

<p>The author describes staying in the loop as the AI writes code, steering constantly. One thing I’d add to their toolkit: ask the AI to produce diagrams. ERDs, state diagrams, sequence diagrams – whatever the task calls for. Just by drawing them, it often finds bugs in the design, before I even review them.</p>

<h3 id="dont-you-mean-extinct"><a href="https://fabiensanglard.net/extinct/index.html">DON’T YOU MEAN EXTINCT?</a></h3>

<p>If we are at the cusp of a paradigm change, it makes sense that we need to evolve. To adapt. Just like the stop-motion animators did. It turns out they had more to offer than stop-motion know-how. They also knew a great deal about dynamics, and about what movement looks natural.</p>

<h3 id="should-you-use-ai-for-a-task-heres-a-simple-way-to-decide"><a href="https://www.schneier.com/blog/archives/2026/07/should-you-use-ai-for-a-task-heres-a-simple-way-to-decide.html">Should You Use AI for a Task? Here’s a Simple Way to Decide</a></h3>

<p>Really useful heuristic: Is it work or the gym?</p>

<p>An example that comes to mind: I can ask AI to summarize the instruction manual for my Instant Pot. The manual was badly written in the first place, and all I need to know is which button to press. But asking AI to compress a novel into a paragraph? What’s the point? The enjoyment of a book is reading it, learning what happens at the book’s own pace.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="machine_learning" /><summary type="html"><![CDATA[The Short Leash AI Coding Method For Beating Fable The author describes staying in the loop as the AI writes code, steering constantly. One thing I’d add to their toolkit: ask the AI to produce diagrams. ERDs, state diagrams, sequence diagrams – whatever the task calls for. Just by drawing them, it often finds bugs in the design, before I even review them. DON’T YOU MEAN EXTINCT? If we are at the cusp of a paradigm change, it makes sense that we need to evolve. To adapt. Just like the stop-motion animators did. It turns out they had more to offer than stop-motion know-how. They also knew a great deal about dynamics, and about what movement looks natural. Should You Use AI for a Task? Here’s a Simple Way to Decide Really useful heuristic: Is it work or the gym? An example that comes to mind: I can ask AI to summarize the instruction manual for my Instant Pot. The manual was badly written in the first place, and all I need to know is which button to press. But asking AI to compress a novel into a paragraph? What’s the point? The enjoyment of a book is reading it, learning what happens at the book’s own pace.]]></summary></entry><entry><title type="html">The REPL: Issue 142 - June 2026</title><link href="https://ylan.segal-family.com/blog/2026/07/07/the-repl-issue-142-june-2026/" rel="alternate" type="text/html" title="The REPL: Issue 142 - June 2026" /><published>2026-07-07T03:36:39-07:00</published><updated>2026-07-07T03:36:39-07:00</updated><id>https://ylan.segal-family.com/blog/2026/07/07/the-repl-issue-142-june-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/07/07/the-repl-issue-142-june-2026/"><![CDATA[<h3 id="theyre-made-out-of-weights"><a href="https://maxleiter.com/blog/weights">They’re Made Out of Weights</a></h3>

<p>Written in the spirit of <a href="https://web.mit.edu/people/dpolicar/writing/prose/text/thinkingMeat.html">They’re Made Out of Meat</a>, a story I like a lot. A total ripoff, but it gets you thinking!</p>

<h3 id="how-building-an-html-first-site-doubled-our-users-overnight"><a href="https://mohkohn.co.uk/writing/html-first/">How building an HTML-first site doubled our users overnight</a></h3>

<p>It turns out that a simpler system consuming fewer resources is beneficial. Shocker.
I guess I am now part of the old guard that wants server-rendered HTML.
This blog has no JavaScript at all. <span class="emoji">🤷🏻‍♂️</span></p>

<h3 id="if-you-are-asking-for-human-attention-demonstrate-human-effort"><a href="https://tombedor.dev/human-attention-and-human-effort/">If You are Asking for Human Attention, Demonstrate Human Effort</a></h3>

<p>It’s rude to ask other folks to read and parse what your <span class="emoji">🤖</span> spews out when you haven’t read it yourself, let alone <em>thought</em> it through. I’ve seen it happen. But most of the time, folks on my team edit, fix, and put in effort before passing along AI-generated output.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="machine_learning" /><category term="web" /><summary type="html"><![CDATA[They’re Made Out of Weights Written in the spirit of They’re Made Out of Meat, a story I like a lot. A total ripoff, but it gets you thinking! How building an HTML-first site doubled our users overnight It turns out that a simpler system consuming fewer resources is beneficial. Shocker. I guess I am now part of the old guard that wants server-rendered HTML. This blog has no JavaScript at all. 🤷🏻‍♂️ If You are Asking for Human Attention, Demonstrate Human Effort It’s rude to ask other folks to read and parse what your 🤖 spews out when you haven’t read it yourself, let alone thought it through. I’ve seen it happen. But most of the time, folks on my team edit, fix, and put in effort before passing along AI-generated output.]]></summary></entry><entry><title type="html">How I Review My Agents’ Code</title><link href="https://ylan.segal-family.com/blog/2026/06/16/how-i-review-my-agents-code/" rel="alternate" type="text/html" title="How I Review My Agents’ Code" /><published>2026-06-16T17:14:24-07:00</published><updated>2026-06-16T17:14:24-07:00</updated><id>https://ylan.segal-family.com/blog/2026/06/16/how-i-review-my-agents-code</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/06/16/how-i-review-my-agents-code/"><![CDATA[<p>Coding agents have changed how I work. I now give code-review-style feedback to agents many times a day.</p>

<p>In the before times, I used <code class="language-plaintext highlighter-rouge">git diff main...</code> to review my own changes locally before pushing to GitHub. As I noticed things, I wrote a todo list for myself to address. Rinse and repeat.</p>

<p>These days that feedback goes to an agent, not to my own todo list. I used to copy-paste file names and code snippets into the chat, but that gets old and verbose fast. Now I use <a href="https://revdiff.com/">revdiff</a>, a TUI built specifically for giving feedback to agents.</p>

<h2 id="giving-feedback-on-diffs">Giving Feedback on Diffs</h2>

<p>At its heart, <code class="language-plaintext highlighter-rouge">revdiff</code> opens a TUI that shows you changes in files. As you navigate, press <code class="language-plaintext highlighter-rouge">return</code> to write a comment on a line (or several). When you’re done, press <code class="language-plaintext highlighter-rouge">q</code> and the program prints your comments in the context of the lines they apply to. From their own example:</p>

<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>## handler.go (file-level)
consider splitting this file into smaller modules

## handler.go:43 (+)
use errors.Is() instead of direct comparison

## handler.go:43-67 (+)
refactor this hunk to reduce nesting

## store.go:18 (-)
don't remove this validation
</code></pre></div></div>

<p>I review code before every commit, to catch issues as early as possible. The default <code class="language-plaintext highlighter-rouge">revdiff</code> invocation shows unstaged changes. It fits my workflow exactly.</p>

<p>You can also pass <code class="language-plaintext highlighter-rouge">git</code> ranges: <code class="language-plaintext highlighter-rouge">revdiff main...</code>, <code class="language-plaintext highlighter-rouge">revdiff branch-1...branch-2</code>, and so on. That’s especially helpful for reviewing a whole branch.</p>

<h2 id="giving-feedback-on-plans">Giving Feedback on Plans</h2>

<p>I have the agent write specs and plans first. That alone produces much better code (I use <a href="https://github.com/obra/superpowers">superpowers</a>). I want to give feedback on those plans, but they aren’t diffs of anything, and most of the time I don’t even put them under source control. <code class="language-plaintext highlighter-rouge">revdiff</code> has me covered: <code class="language-plaintext highlighter-rouge">revdiff --only path/to/plan.md</code> opens the same feedback mechanism, with table-of-contents support for markdown. It beats what I did before: opening the file in my editor, adding comments prefixed with <code class="language-plaintext highlighter-rouge">---&gt;</code> (or similar), then telling the agent to go find them.</p>

<h2 id="no-skill">No Skill</h2>

<p><code class="language-plaintext highlighter-rouge">revdiff</code> ships with a skill that opens the feedback for you in an overlay window, supporting many different setups. I haven’t needed it. Because I have my agent set up to prompt me before committing any code, I get a yes/no question when it’s time to review a diff. I find it easy to switch to another terminal window, fire up <code class="language-plaintext highlighter-rouge">revdiff</code>, give feedback, and switch back. Then I accept the changes, or select no and paste the output. The skill would require me to stop the agent just to invoke it.</p>

<h2 id="copy-to-clipboard-on-exit">Copy to Clipboard on Exit</h2>

<p>Shuttling output between windows by hand added friction, so I automated copying the revdiff output to my clipboard every time. My <a href="https://fishshell.com/">fish shell</a> configuration:</p>

<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>function revdiff --description "Run revdiff and copy annotations to clipboard"
    set -l tmp (mktemp /tmp/revdiff.XXXXXX)
    command revdiff --output=$tmp $argv
    and cat $tmp
    and pbcopy &lt; $tmp
    rm -f $tmp
end
</code></pre></div></div>

<p>I don’t worry about overriding my clipboard, because I use a <a href="https://www.raycast.com">Raycast</a>’s clipboard manager.</p>

<h2 id="config-tweaks">Config Tweaks</h2>

<p>Lastly, I’ve tweaked the configuration:</p>

<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>$ cat ~/.config/revdiff/config
theme = nord
cross-file-hunks = true
compact = true
</code></pre></div></div>

<p>I can’t say enough good things about the <a href="https://www.nordtheme.com">Nord theme</a> in dark mode.</p>

<h2 id="conclusion">Conclusion</h2>

<p><code class="language-plaintext highlighter-rouge">revdiff</code> has been an immediate productivity boost. I use it many times a day, and it has made interacting with agents much easier.</p>]]></content><author><name>Ylan Segal</name></author><category term="machine_learning" /><category term="productivity" /><summary type="html"><![CDATA[Coding agents have changed how I work. I now give code-review-style feedback to agents many times a day. In the before times, I used git diff main... to review my own changes locally before pushing to GitHub. As I noticed things, I wrote a todo list for myself to address. Rinse and repeat. These days that feedback goes to an agent, not to my own todo list. I used to copy-paste file names and code snippets into the chat, but that gets old and verbose fast. Now I use revdiff, a TUI built specifically for giving feedback to agents. Giving Feedback on Diffs At its heart, revdiff opens a TUI that shows you changes in files. As you navigate, press return to write a comment on a line (or several). When you’re done, press q and the program prints your comments in the context of the lines they apply to. From their own example: ## handler.go (file-level) consider splitting this file into smaller modules ## handler.go:43 (+) use errors.Is() instead of direct comparison ## handler.go:43-67 (+) refactor this hunk to reduce nesting ## store.go:18 (-) don't remove this validation I review code before every commit, to catch issues as early as possible. The default revdiff invocation shows unstaged changes. It fits my workflow exactly. You can also pass git ranges: revdiff main..., revdiff branch-1...branch-2, and so on. That’s especially helpful for reviewing a whole branch. Giving Feedback on Plans I have the agent write specs and plans first. That alone produces much better code (I use superpowers). I want to give feedback on those plans, but they aren’t diffs of anything, and most of the time I don’t even put them under source control. revdiff has me covered: revdiff --only path/to/plan.md opens the same feedback mechanism, with table-of-contents support for markdown. It beats what I did before: opening the file in my editor, adding comments prefixed with ---&gt; (or similar), then telling the agent to go find them. No Skill revdiff ships with a skill that opens the feedback for you in an overlay window, supporting many different setups. I haven’t needed it. Because I have my agent set up to prompt me before committing any code, I get a yes/no question when it’s time to review a diff. I find it easy to switch to another terminal window, fire up revdiff, give feedback, and switch back. Then I accept the changes, or select no and paste the output. The skill would require me to stop the agent just to invoke it. Copy to Clipboard on Exit Shuttling output between windows by hand added friction, so I automated copying the revdiff output to my clipboard every time. My fish shell configuration: function revdiff --description "Run revdiff and copy annotations to clipboard" set -l tmp (mktemp /tmp/revdiff.XXXXXX) command revdiff --output=$tmp $argv and cat $tmp and pbcopy &lt; $tmp rm -f $tmp end I don’t worry about overriding my clipboard, because I use a Raycast’s clipboard manager. Config Tweaks Lastly, I’ve tweaked the configuration: $ cat ~/.config/revdiff/config theme = nord cross-file-hunks = true compact = true I can’t say enough good things about the Nord theme in dark mode. Conclusion revdiff has been an immediate productivity boost. I use it many times a day, and it has made interacting with agents much easier.]]></summary></entry><entry><title type="html">The REPL: Issue 141 - May 2026</title><link href="https://ylan.segal-family.com/blog/2026/06/01/the-repl-issue-141-may-2026/" rel="alternate" type="text/html" title="The REPL: Issue 141 - May 2026" /><published>2026-06-01T19:15:11-07:00</published><updated>2026-06-01T19:15:11-07:00</updated><id>https://ylan.segal-family.com/blog/2026/06/01/the-repl-issue-141-may-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/06/01/the-repl-issue-141-may-2026/"><![CDATA[<h3 id="nates-dotfiles"><a href="https://github.com/nateberkopec/dotfiles/blob/main/files/home/.claude/skills/dev-env-setup/SKILL.md">Nate’s Dotfiles</a></h3>

<p>A couple of nuggets I picked up:</p>

<ul>
  <li><code class="language-plaintext highlighter-rouge">mise.local.toml</code> lets me add personal tool configuration to any project – even one whose <code class="language-plaintext highlighter-rouge">mise.toml</code> I don’t control, or that has none at all.</li>
  <li><code class="language-plaintext highlighter-rouge">.git/info/exclude</code> works like <code class="language-plaintext highlighter-rouge">.gitignore</code>, but for my eyes only. <code class="language-plaintext highlighter-rouge">.gitignore</code> is itself tracked by git; <code class="language-plaintext highlighter-rouge">.git/info/exclude</code> ignores files without recording the exclusion in the repository.</li>
</ul>

<h3 id="three-inverse-laws-of-ai"><a href="https://susam.net/inverse-laws-of-robotics.html">Three Inverse Laws of AI</a></h3>

<p>In a nod to Isaac Asimov’s laws <em>for</em> robots, the author proposes three laws for the humans <em>using</em> them – incomplete by their own admission, but worth chewing on:</p>

<ul>
  <li>Non-Anthropomorphism</li>
  <li>Non-Deference</li>
  <li>Non-Abdication of Responsibility</li>
</ul>

<p>It all comes down to one habit: keep exercising critical thinking.</p>

<h3 id="appearing-productive-in-the-workplace--no-ones-happy"><a href="https://nooneshappy.com/article/appearing-productive-in-the-workplace/">Appearing Productive in The Workplace — No One’s Happy</a></h3>

<blockquote>
  <p>What discipline looks like, in this environment, is almost embarrassingly old-fashioned and may seem obvious to most of you until you try to avoid it. Use the tool where you can verify precisely what it produces. Never ask a model for confirmation; the tool agrees with everyone, and an agreement that costs the agreer nothing is worth nothing.</p>
</blockquote>

<p>I’ve been using this example with my non-tech friends: ask the 🤖 for a contract to sell your car in California and it will happily produce one, brimming with confidence that it’s sound. I’m not a lawyer, so I can’t judge it. I can skim it for anything obviously crazy, <em>but</em> I don’t know what’s missing, what it fails to protect me from, or whether every clause is even legal.</p>

<h3 id="we-replaced-redis-with-mysql-for-inventory-reservationsand-it-scaled-2026"><a href="https://shopify.engineering/scaling-inventory-reservations">We replaced Redis with MySQL for inventory reservations—and it scaled (2026)</a></h3>

<p>A clear account of moving the locking mechanism from Redis into the database to guarantee atomicity, with <code class="language-plaintext highlighter-rouge">FOR UPDATE SKIP LOCKED</code> put to great effect. The idea new to me was the “replenishing” pool: a set of reservable items large enough to absorb concurrent locks, yet small enough to stay practical.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="machine_learning" /><category term="databases" /><category term="unix" /><summary type="html"><![CDATA[Nate’s Dotfiles A couple of nuggets I picked up: mise.local.toml lets me add personal tool configuration to any project – even one whose mise.toml I don’t control, or that has none at all. .git/info/exclude works like .gitignore, but for my eyes only. .gitignore is itself tracked by git; .git/info/exclude ignores files without recording the exclusion in the repository. Three Inverse Laws of AI In a nod to Isaac Asimov’s laws for robots, the author proposes three laws for the humans using them – incomplete by their own admission, but worth chewing on: Non-Anthropomorphism Non-Deference Non-Abdication of Responsibility It all comes down to one habit: keep exercising critical thinking. Appearing Productive in The Workplace — No One’s Happy What discipline looks like, in this environment, is almost embarrassingly old-fashioned and may seem obvious to most of you until you try to avoid it. Use the tool where you can verify precisely what it produces. Never ask a model for confirmation; the tool agrees with everyone, and an agreement that costs the agreer nothing is worth nothing. I’ve been using this example with my non-tech friends: ask the 🤖 for a contract to sell your car in California and it will happily produce one, brimming with confidence that it’s sound. I’m not a lawyer, so I can’t judge it. I can skim it for anything obviously crazy, but I don’t know what’s missing, what it fails to protect me from, or whether every clause is even legal. We replaced Redis with MySQL for inventory reservations—and it scaled (2026) A clear account of moving the locking mechanism from Redis into the database to guarantee atomicity, with FOR UPDATE SKIP LOCKED put to great effect. The idea new to me was the “replenishing” pool: a set of reservable items large enough to absorb concurrent locks, yet small enough to stay practical.]]></summary></entry><entry><title type="html">The REPL: Issue 140 - April 2026</title><link href="https://ylan.segal-family.com/blog/2026/05/01/the-repl-issue-140-april-2026/" rel="alternate" type="text/html" title="The REPL: Issue 140 - April 2026" /><published>2026-05-01T09:30:55-07:00</published><updated>2026-05-01T09:30:55-07:00</updated><id>https://ylan.segal-family.com/blog/2026/05/01/the-repl-issue-140-april-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/05/01/the-repl-issue-140-april-2026/"><![CDATA[<h1 id="keeping-a-postgres-queue-healthy--planetscale"><a href="https://planetscale.com/blog/keeping-a-postgres-queue-healthy">Keeping a Postgres queue healthy — PlanetScale</a></h1>

<p>Simeon Griggs explains the challenges and dynamics of queue workloads in Postgres. A job queue in Postgres is desirable because of transactionality, and because it spares you from maintaining a separate system. At a certain scale, though, it outpaces the database’s ability to clean up after itself. I haven’t evaluated PlanetScale’s solution to the problem.</p>

<h3 id="highlights-from-git-254"><a href="https://github.blog/open-source/git/highlights-from-git-2-54/">Highlights from Git 2.54</a></h3>

<p><code class="language-plaintext highlighter-rouge">git</code> is my most used command. I’m glad to see the CLI still improving. <code class="language-plaintext highlighter-rouge">git history</code> looks genuinely useful for splitting or rewording commits. The new hook configuration also offers a much cleaner way to declare hooks – clearly needed, judging by the many popular hook-management systems out there.</p>

<h3 id="do-i-belong-in-tech-anymore"><a href="https://ky.fyi/posts/ai-burnout">Do I belong in tech anymore?</a></h3>

<p>Ky Decker writes a personal account of disillusionment with the tech sector. In the “The psychic toll of AI” section, Ky describes a thoughtless use of AI in his workplace, where critical thinking has gone out the window and AI is shoved everywhere. A harrowing account. I’m glad I’m not in that situation. In a short time, AI has changed how I plan features, explore alternatives, and code. Yet the technology has limitations and needs a human in the loop to check its output.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="postgres" /><category term="machine_learning" /><category term="git" /><summary type="html"><![CDATA[Keeping a Postgres queue healthy — PlanetScale Simeon Griggs explains the challenges and dynamics of queue workloads in Postgres. A job queue in Postgres is desirable because of transactionality, and because it spares you from maintaining a separate system. At a certain scale, though, it outpaces the database’s ability to clean up after itself. I haven’t evaluated PlanetScale’s solution to the problem. Highlights from Git 2.54 git is my most used command. I’m glad to see the CLI still improving. git history looks genuinely useful for splitting or rewording commits. The new hook configuration also offers a much cleaner way to declare hooks – clearly needed, judging by the many popular hook-management systems out there. Do I belong in tech anymore? Ky Decker writes a personal account of disillusionment with the tech sector. In the “The psychic toll of AI” section, Ky describes a thoughtless use of AI in his workplace, where critical thinking has gone out the window and AI is shoved everywhere. A harrowing account. I’m glad I’m not in that situation. In a short time, AI has changed how I plan features, explore alternatives, and code. Yet the technology has limitations and needs a human in the loop to check its output.]]></summary></entry><entry><title type="html">The REPL: Issue 139 - March 2026</title><link href="https://ylan.segal-family.com/blog/2026/04/12/the-repl-issue-139-march-2026/" rel="alternate" type="text/html" title="The REPL: Issue 139 - March 2026" /><published>2026-04-12T14:05:13-07:00</published><updated>2026-04-12T14:05:13-07:00</updated><id>https://ylan.segal-family.com/blog/2026/04/12/the-repl-issue-139-march-2026</id><content type="html" xml:base="https://ylan.segal-family.com/blog/2026/04/12/the-repl-issue-139-march-2026/"><![CDATA[<h3 id="llms-can-be-absolutely-exhausting"><a href="https://tomjohnell.com/llms-can-be-absolutely-exhausting/">LLMs can be absolutely exhausting</a></h3>

<p>The author points out that some of the frustrating things about working with LLMs are derived from <em>how</em> you work with the LLM. The author is not wrong, but those same things he identifies as sources of diminishing returns are true <strong>without</strong> LLMs too. Fatigue has an impact on output, as do lack of thinking ahead of time about the correct end-state and slow feedback loops. There is nothing new under the sun.</p>

<h3 id="what-is-agentic-engineering---agentic-engineering-patterns"><a href="https://simonwillison.net/guides/agentic-engineering-patterns/what-is-agentic-engineering/">What is agentic engineering? - Agentic Engineering Patterns</a></h3>

<p>Agentic Engineering is a new term for me. I like it better than vibe coding, for the same reason that I always describe myself as a software engineer and not a coder. Engineering is designing and building systems. Not just knowing the correct syntax to talk to a computer.</p>

<blockquote>
  <p>Writing code has never been the sole activity of a software engineer. The craft has always been figuring out what code to write.</p>
</blockquote>

<p>Exactly. Agents have made coding easier, but where I’ve felt they really increase my productivity is in the long planning sessions, effortless design documents, generating as many diagrams as I want in seconds. This lets me really dig into a design and hone it.</p>

<h3 id="warranty-void-if-regenerated"><a href="https://nearzero.software/p/warranty-void-if-regenerated">Warranty Void If Regenerated</a></h3>

<p>I liked this short story a lot. At first glance a story about farming and AI seems weird. Farming is “close to the earth” and seems not very tech-oriented. Upon closer inspection, farming is an industry that has changed dramatically because of technology. Modern farming makes it possible for 5% of the population to produce all our food. It used to take &gt;95%.</p>

<p>This story is imagining yet more changes to farming.</p>

<h3 id="reports-of-codes-death-are-greatly-exaggerated"><a href="https://stevekrouse.com/precision">Reports of code’s death are greatly exaggerated</a></h3>

<p>AI can already write a novel today. You can do it right now. Will that be a great novel? Probably not. And no one seems to be saying that it will lead to novelists being completely out of the job. In fact, I think AI is a mediocre writer, as writers go. But most people are not good at writing. Having an AI proofread or write what they want to say actually improves their communication.</p>

<p>Is it the same with code? Maybe there is something to that analogy. Today, AI can create great scripts that you can use locally. Even apps for personal use that will truly solve problems for individuals, even small businesses. Does that mean no one will need to know anything about software and vibe code Netflix’s streaming infrastructure? I’m not holding my breath.</p>

<p>I have a couple of friends that have small businesses and are vibe coding some internal apps. I am truly worried that they will get into trouble by publishing credentials, leaving open infrastructure, etc. And the thing is, they know so little about the risks and how the internet works, that it’s hard to explain to them what they are doing. “What if I ask 🤖 to review for security?”. Sure. It will help, but they probably don’t know enough to prompt it correctly.</p>]]></content><author><name>Ylan Segal</name></author><category term="the repl" /><category term="machine_learning" /><summary type="html"><![CDATA[LLMs can be absolutely exhausting The author points out that some of the frustrating things about working with LLMs are derived from how you work with the LLM. The author is not wrong, but those same things he identifies as sources of diminishing returns are true without LLMs too. Fatigue has an impact on output, as do lack of thinking ahead of time about the correct end-state and slow feedback loops. There is nothing new under the sun. What is agentic engineering? - Agentic Engineering Patterns Agentic Engineering is a new term for me. I like it better than vibe coding, for the same reason that I always describe myself as a software engineer and not a coder. Engineering is designing and building systems. Not just knowing the correct syntax to talk to a computer. Writing code has never been the sole activity of a software engineer. The craft has always been figuring out what code to write. Exactly. Agents have made coding easier, but where I’ve felt they really increase my productivity is in the long planning sessions, effortless design documents, generating as many diagrams as I want in seconds. This lets me really dig into a design and hone it. Warranty Void If Regenerated I liked this short story a lot. At first glance a story about farming and AI seems weird. Farming is “close to the earth” and seems not very tech-oriented. Upon closer inspection, farming is an industry that has changed dramatically because of technology. Modern farming makes it possible for 5% of the population to produce all our food. It used to take &gt;95%. This story is imagining yet more changes to farming. Reports of code’s death are greatly exaggerated AI can already write a novel today. You can do it right now. Will that be a great novel? Probably not. And no one seems to be saying that it will lead to novelists being completely out of the job. In fact, I think AI is a mediocre writer, as writers go. But most people are not good at writing. Having an AI proofread or write what they want to say actually improves their communication. Is it the same with code? Maybe there is something to that analogy. Today, AI can create great scripts that you can use locally. Even apps for personal use that will truly solve problems for individuals, even small businesses. Does that mean no one will need to know anything about software and vibe code Netflix’s streaming infrastructure? I’m not holding my breath. I have a couple of friends that have small businesses and are vibe coding some internal apps. I am truly worried that they will get into trouble by publishing credentials, leaving open infrastructure, etc. And the thing is, they know so little about the risks and how the internet works, that it’s hard to explain to them what they are doing. “What if I ask 🤖 to review for security?”. Sure. It will help, but they probably don’t know enough to prompt it correctly.]]></summary></entry></feed>