<?xml version="1.0" encoding="UTF-8"?>
<rss  xmlns:atom="http://www.w3.org/2005/Atom" 
      xmlns:media="http://search.yahoo.com/mrss/" 
      xmlns:content="http://purl.org/rss/1.0/modules/content/" 
      xmlns:dc="http://purl.org/dc/elements/1.1/" 
      version="2.0">
<channel>
<title>Tech Blog</title>
<link>https://gw.ch3n2k.com/tech/en/</link>
<atom:link href="https://gw.ch3n2k.com/tech/en/index.xml" rel="self" type="application/rss+xml"/>
<description>Articles on software engineering, AI-assisted development, and team collaboration.</description>
<language>en</language>
<generator>quarto-1.10.18</generator>
<lastBuildDate>Wed, 18 Nov 2026 16:00:00 GMT</lastBuildDate>
<item>
  <title>What If a PR Comment Were a Mini PR?</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-11-19-mini-pr-comments/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-11-19-mini-pr-comments/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>Once a pull request is open, one person often keeps writing code while everyone else leaves comments. Someone spots a problem. Someone else knows how to fix it. Another reviewer wants to add a test. Turning all that feedback into code still falls to the PR author.</p>
<p>Reviewers with permission to push can update the PR branch directly. As soon as they do, the version under review changes for everyone. The author might already be editing the same code, and other reviewers might still be looking at the previous version.</p>
<p>What if a review comment that includes code were itself a mini PR? Reviewers could branch from the version they’re reviewing and attach a set of commits to their comment, along with an explanation. <strong>Each mini PR could run its own CI and accept other mini PRs targeting its branch.</strong></p>
<p>Suppose I add an <code>--output-dir</code> option to a command-line tool. While reviewing my PR, Alice notices that a failed write can leave behind a partially written file. She fixes the problem on her own branch, adds a regression test, and posts the change as a mini PR in a comment on the original PR.</p>
<p>Bob reviews Alice’s proposal and spots a missing case. If the destination file already exists, a failed write should leave it intact. He opens a mini PR against Alice’s proposal with a test for that case, and CI can check his change separately. If Alice accepts it, she can rerun CI on the combined changes. I can then merge Alice’s proposal into the original PR branch, bringing in the fix and both reviewers’ tests.</p>
<p>If you already use AI to review code, this is close to what you do today. When you spot an issue worth raising, ask AI to draft a fix and a regression test on your branch. The main extra step is checking the generated diff and test. Once you’re happy with the changes, post the commits with the explanation you would have written anyway. <strong>Posting the comment creates a mini PR and kicks off its own CI run.</strong></p>
<p>As the author of the original PR, you’d get a change you can try out as part of the feedback. You could keep working on your implementation while reviewers prepare their proposals, then merge the changes that fit the PR’s goal and test them together.</p>
<p>Within the same PR, the line between author and reviewer would blur, and the two could even trade roles. Working from the proposed changes and test results would help them refine an approach together and get from feedback to a verified change faster. The team could move implementation, tests, and documentation forward in parallel as part of the review. <strong>One person can open a PR and the team can finish it together.</strong></p>
<p>If GitHub or your code review tool supported this workflow, would you consider adopting it on your team?</p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Code Review</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-11-19-mini-pr-comments/</guid>
  <pubDate>Wed, 18 Nov 2026 16:00:00 GMT</pubDate>
</item>
<item>
  <title>Revisiting Hardware Efficiency in the AI Era</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-11-01-ai-hardware-efficiency/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-11-01-ai-hardware-efficiency/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>In the AI era, making good use of hardware may be a skill worth revisiting.</p>
<p>Decades ago, hardware resources were scarce, and programmers had to budget them carefully.</p>
<p>Developers on the Famicom and Game Boy built entire game worlds with very little graphics memory by reusing tiles and limiting their color palettes. The Doom engine relied heavily on fixed-point arithmetic and precomputed lookup tables to reduce work at runtime. By the PS2 era, developers had to coordinate computation on the vector units with data transfers to keep the hardware from sitting idle.</p>
<p>CPU cycles, memory, storage capacity, and bandwidth all had to be budgeted.</p>
<p>Over time, processors got faster, memory and storage became more plentiful, and the cost per unit of hardware generally fell as developer time became more expensive. The industry put more emphasis on developer productivity. Teams accepted more compute, memory, storage, and communication overhead in exchange for faster development and lower maintenance costs.</p>
<p>Automatic memory management meant less manual work freeing memory, with garbage collection taking on some of the work at runtime. Electron lets teams build cross-platform desktop apps quickly with web technology, with the memory overhead of a browser runtime. An ORM reduces the need to write SQL by hand, and a simple call can hide extra queries, unneeded data, and many temporary objects. Microservices let teams develop and deploy independently, with costs from serialization, data copying, and network traffic.</p>
<p>Those trade-offs made economic sense. Using a little more CPU time or memory, or storing extra copies of data, often cost less than several weeks of engineering time. Over time, many of those costs became routine.</p>
<p>That calculation may be changing again.</p>
<p>AI can help developers read code, analyze performance data, refactor implementations, and write benchmarks. That makes it cheaper to try an optimization. Recent price increases for some hardware also make more efficient resource use more valuable.</p>
<p>Batching per-item database queries cuts down on round trips. Avoiding repeated JSON parsing and unnecessary copies saves CPU time and memory. More compact representations use less storage and bandwidth. Changing the data layout in hot paths can improve cache utilization and cut the time processors spend waiting for data.</p>
<p>In AI applications, choosing a smaller model that fits the task, quantizing it, or batching inference can also reduce compute and GPU memory requirements so the same hardware can serve more requests.</p>
<p>These optimizations have always had value. AI can make more of them worth the engineering time and let teams verify the gains sooner with real workloads.</p>
<p>Alongside implementing business logic, developers can spend more time on resource efficiency along the critical path. That means looking at how objects are allocated, how data is laid out, what causes cache misses, how parallel tasks are scheduled, and what it costs to store and move data.</p>
<p>Those low-level details that have faded behind layers of abstraction deserve our attention again.</p>
<p>As AI makes code easier to write, understanding how that code can use hardware efficiently will become an increasingly valuable engineering skill.</p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Performance Optimization</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-11-01-ai-hardware-efficiency/</guid>
  <pubDate>Sat, 31 Oct 2026 16:00:00 GMT</pubDate>
</item>
<item>
  <title>Rethinking Remote Development From Ericsson to Codex</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-09-21-remote-development-revisited/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-09-21-remote-development-revisited/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>By the time I listened to Syntax #1023, I’d already been using Codex for remote development for a while. The episode reminded me of developing remotely at Ericsson more than a decade ago.</p>
<p>Back then, the setup was considered extremely inefficient. Here I am using a similar setup again, and the familiarity made me want to revisit that judgment.</p>
<p>The basic idea behind remote development is simple. Put the development environment and its workloads on another machine, then connect to it. That machine could be a cloud server or a Mac mini at home.</p>
<p>The episode covered some practical benefits.</p>
<ul>
<li>Close the laptop and remote tasks keep running.</li>
<li>Move background processes to another machine to reduce load and save battery on the laptop.</li>
<li>Switch devices and pick up with the same tasks.</li>
</ul>
<p>For me, the interesting change is the rhythm of working with the machine.</p>
<p>When development depends on a steady stream of actions and immediate responses, latency and friction in a remote connection keep interrupting your train of thought. A tiny wait adds up when it happens hundreds of times.</p>
<p>Coding tools like Codex add another rhythm. Describe a task, let the tool work, and come back to check the result and adjust course. For tasks that can run independently for a while, it becomes more valuable to have a development environment that stays online and is ready whenever you reconnect.</p>
<p>Long builds and background jobs already existed more than a decade ago, and today’s development still involves plenty of immediate interaction. The latency and friction of remote development still matter in the AI era.</p>
<p><strong>How efficient an architecture is depends on how you actually work.</strong></p>
<p>The episode made me look at my remote setup across the whole workflow. From assigning a task through checking, correcting, and accepting the result, how much time does it save, and what does it cost to maintain?</p>
<p>That earlier experience is worth remembering, and today’s setup is worth evaluating on its own terms.</p>
<p>Do you remember what early remote development was like? Have AI coding tools changed how you think about it?</p>
<p><a href="https://syntax.fm/1023">Listen to Syntax #1023</a></p>
<p>#RemoteDevelopment #Codex #DeveloperExperience</p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Remote Development</category>
  <category>Developer Experience</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-09-21-remote-development-revisited/</guid>
  <pubDate>Sun, 20 Sep 2026 16:00:00 GMT</pubDate>
</item>
<item>
  <title>How Do You Keep Your Coding Skills Sharp When AI Writes the Code?</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-09-19-ai-coding-skills-atrophy/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-09-19-ai-coding-skills-atrophy/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>A developer asked the Syntax hosts a question. AI was generating 80–90% of the code in his project, and he was finding it hard to write even an HTTP request on his own. Should he go back and practice the basics?</p>
<p>That question captures both the productivity gains of AI-assisted coding and the anxiety about losing skills that can come with them.</p>
<p>Scott Tolinski and Wes Bos raised several points in Syntax #1031 that got me thinking.</p>
<p><strong>Scott’s point was that skills need regular practice.</strong></p>
<p>Those skills can fade when we stop practicing them ourselves and stop working through how an implementation fits together. Scott kept coming back to the idea of being an active participant.</p>
<p>Even when AI writes the code, we still need to take part in the reasoning, understand how the solution came together, and judge the result.</p>
<p><strong>Wes distinguished two kinds of skill loss that are easy to confuse.</strong></p>
<p>One is forgetting syntax details, such as the order of the arguments to a <code>reduce</code> callback.</p>
<p>The other is losing track of how a program works. How does a request move through the application? Where are APIs called? How do the modules work together? Where might performance problems arise?</p>
<p>Tools can help us look up syntax. Understanding how the program works is central to solving problems.</p>
<p>Wes is particularly concerned that AI makes it easy to move a project forward while skipping the code. Over time, we may lose the habit of asking why it works the way it does. Accepting the first result can take the place of researching, weighing options, and learning along the way.</p>
<p><strong>The way we divide our attention can make the problem worse.</strong></p>
<p>Scott said that while waiting for AI to respond, he easily switches to something else. At one point he had 13 projects open at once. They all seemed to be moving forward, and it was getting harder for him to go deep on any one of them.</p>
<p>Wes also finds it harder to get into the flow of writing code. He still feels that focus and excitement when researching a problem, designing an architecture, choosing technologies, or using AI to explore unfamiliar territory.</p>
<p>My takeaway is that growth shows up both in what I deliver and in how well I understand it.</p>
<p>When working with AI-generated code, I can ask myself three questions.</p>
<ul>
<li>Can I explain why it works this way?</li>
<li>If it breaks, do I know where to start looking?</li>
<li>Can I explain the trade-offs and how I’d verify the approach?</li>
</ul>
<p><strong>AI can generate code. Understanding it and exercising judgment still take practice.</strong></p>
<p>When you code with AI, which parts do you deliberately do yourself or work through on your own?</p>
<p><a href="https://syntax.fm/1031">Listen to Syntax #1031</a></p>
<p>#AICoding #SoftwareEngineering #DeveloperGrowth</p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Developer Growth</category>
  <category>Developer Experience</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-09-19-ai-coding-skills-atrophy/</guid>
  <pubDate>Fri, 18 Sep 2026 16:00:00 GMT</pubDate>
</item>
<item>
  <title>Code Review in the AI Era Starts With Intent</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-09-18-ai-code-review-intent/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-09-18-ai-code-review-intent/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>The faster we can generate code, the earlier we should discuss why we want to make a change the way we do.</p>
<p>In an interview with The Pragmatic Engineer, Codex co-creator Tibo Sottiaux described code review as a way to check correctness, share knowledge, and agree on intent. AI is changing how teams carry out those responsibilities.</p>
<p>Sottiaux said OpenAI had added AI-powered security checks to its PR process and would block a merge if the checks flagged a security issue. Models can also follow dependencies to find discrepancies between documentation and implementation. Those discrepancies can take a person a long time to track down.</p>
<p>He expects more correctness and security checks to become automated. Teams still need to work through a few questions.</p>
<ul>
<li>What problem are we trying to solve?</li>
<li>Is this change worth making?</li>
<li>What should the system guarantee, and what trade-offs can we accept?</li>
</ul>
<p>Ideally, those discussions happen before anyone writes the code.</p>
<p>PRs have often been the last checkpoint for requirements and design. The code is already written by the time people start asking whether the goal makes sense. With AI speeding up implementation, teams that wait until code review to agree on intent may find themselves cycling through rework faster too.</p>
<p>Engineering teams can strengthen automated review and get clear about requirements, interface contracts, and acceptance criteria earlier.</p>
<p>As code review evolves, the way teams collaborate needs to evolve with it.</p>
<p>#CodeReview #SoftwareEngineering #AI #EngineeringLeadership</p>
<p><a href="https://newsletter.pragmaticengineer.com/p/building-codex-with-tibo-sottiaux">Building Codex with Tibo Sottiaux on The Pragmatic Engineer</a></p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Code Review</category>
  <category>Team Collaboration</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-09-18-ai-code-review-intent/</guid>
  <pubDate>Thu, 17 Sep 2026 16:00:00 GMT</pubDate>
</item>
<item>
  <title>AI Makes Coding Faster. Is the Work Any Easier?</title>
  <link>https://gw.ch3n2k.com/tech/en/posts/2026-09-18-ai-coding-fatigue/</link>
  <description><![CDATA[ 




<p><a href="../../../../tech/en/index.html">Tech Blog</a> · <a href="../../../../tech/posts/2026-09-18-ai-coding-fatigue/index.html" lang="zh-Hans" hreflang="zh-Hans" rel="alternate">简体中文</a></p>
<p>In episode 1039 of Syntax, a listener said he and his colleagues were producing more code with AI. His work increasingly felt like back-to-back PR reviews, and he found it exhausting.</p>
<p>It reminded me that generating code is one part of delivering a change.</p>
<p>When you implement a feature step by step, you build an understanding of the requirements, the data flow, and the boundaries of the change. When you receive AI-generated code, you may need to build that understanding afterward.</p>
<p>For example, an export feature might come together quickly. Someone still has to check that it exports the right data, handles failures correctly, and follows existing interface contracts.</p>
<p>Between answering questions, checking progress, and switching among several agents, the time saved during implementation can turn into more concentrated review and decision-making.</p>
<p>Teams face the same problem. If changes come in for review faster than the team can get through them, the backlog keeps growing.</p>
<p>The measure I care most about is how much time the team spends implementing, understanding, reviewing, and reworking a change from the initial request through delivery.</p>
<p>A few adjustments are worth trying.</p>
<ul>
<li>Give each change a clear purpose and a way to verify it. This makes it easier to review unfamiliar code in manageable pieces.</li>
<li>Agree up front on how to handle routine decisions. That leaves more attention for important trade-offs and cuts down on repeated confirmations.</li>
<li>Limit work in progress to what the team can review. Leave enough room to finish work that’s already underway.</li>
</ul>
<p>For a small, familiar change, implementing it directly can be efficient. How much AI contributes can vary by task.</p>
<p>What changes have made AI-assisted coding easier on your team?</p>
<p>The discussion starts at 26:19 in <a href="https://syntax.fm/1039">Syntax #1039</a>.</p>
<p>#AICoding #SoftwareEngineering #DeveloperExperience</p>



 ]]></description>
  <category>Software Engineering</category>
  <category>AI-Assisted Development</category>
  <category>Code Review</category>
  <category>Developer Experience</category>
  <guid>https://gw.ch3n2k.com/tech/en/posts/2026-09-18-ai-coding-fatigue/</guid>
  <pubDate>Thu, 17 Sep 2026 16:00:00 GMT</pubDate>
</item>
</channel>
</rss>
