A Smooth Sea Never Made a Skillful Sailor

In July I asked an agent to push a build of my personal assistant app to my phone. Routine task. I do it most days. It built the release APK, ran flutter install , and my habits, journal, and health logs were gone. Not corrupted, not partially migrated. Gone. flutter install uninstalls the app first when it feels like it, and everything my assistant had learned about me lived in app-private storage with no cloud copy I could count on. The part that still bothers me is not that it happened. It is that the instruction telling it not to do this was already sitting in the repository, in the file the agent claims to read. The Rule Was Already There, in Capital Letters After the first time this happened, I did the obvious thing. I wrote the rule down. The commit is still in my history, and the message reads like an incident report because that is what it was: flutter install uninstalled the app before reinstalling on a real device, wiping local user data (habits, journal, health log...

In my last post I compared Codex, Claude, and GLM 5.2 and landed on two $20 monthly plans as the setup that gets the work done. Two weeks in, I changed that setup: Claude moved to the $200 annual plan, which is essentially the $20 monthly plan at a discount, and Ollama moved to the $20 Pro plan. Same shape, better price. This post is a follow-up on two other fronts. First, the workflow itself has firmed up: GLM 5.2 does the assessment and planning on less critical projects, Fable 5 does the assessment on complex or mission-critical projects and validates GLM 5.2's findings, and Sonnet 5 handles general-purpose coding once Fable 5 has set the plan. Second, I ran the experiment I suspected would go badly and paid for AI by the token, once through an OpenAI API key with the GPT-5.6 Terra model and once through Ollama top-up credits. The numbers are worth sharing because the gap is not small. It is the difference between 3-4 days of work and 30 minutes. The Workflow: Who Assesses,...

The question I keep getting asked is which AI coding assistant I actually use day to day. The honest answer is that I use more than one, and the reason is that none of them is uniformly the best. Each one has a lane where it clearly wins, and each one has a failure mode that pushes me back to the others. This post is a practical comparison of three tools I have spent real time with: Codex, Claude, and GLM 5.2. I am not interested in benchmark scores or leaderboard positioning. I care about how these tools behave across the full lifecycle of a project: initial scaffolding, iterative refinement, design polish, and long-running reliability. The short version: GLM 5.2 has replaced Codex as my daily driver for implementation work. Codex failed the wife test, which is what pushed me to switch. Claude is strong on design and polish but showed a huge regression after two weeks of use. GLM 5.2 is great at identifying tiny lingering issues and regressions, but the last 24 hours of work showed...

I am not trying to crown one AI tool as the winner forever. These tools change too fast for that. What matters to me is more practical: when I sit down to do real work, which one helps me keep going, make fewer messy mistakes, and understand what happened when the session is over? Codex and Claude Code both do serious coding work. They can read files, make changes, run commands, and help move a project forward. But they feel different. Codex feels more like the tool I use when I need to finish implementation. Claude feels stronger when the work needs taste, presentation, and design sense. This is also just where I am right now. My opinion is based on GPT-5.5 in Codex and Opus 4.8 in Claude Code. Fable 5 also came out this week, on June 9, 2026, so even while I am writing this, the ground is already moving. Codex Feels Better When the Job Has a Clear Finish Line Codex works best for me when the task is concrete. Read the repo. Follow the project rules. Make the change. Run the ch...

Zone 2 Trainer started as a simple idea: I wanted a lightweight way to stay honest about easy aerobic work without committing to another subscription or buying another dedicated device. The app became both a fitness tool and a test case for how far modern AI coding assistants can carry a non-mobile developer. For the last couple of months, I have been developing an Android application to enhance my endurance and recovery. Most applications on the market require an active subscription and a special device. With the current developments in AI-assisted software development, I was able to step out of my comfort zone and build an application from scratch using Android Studio, Antigravity, and Codex. That matters to me because I am not a developer by trade. I am closer to a tinkerer who is comfortable with scripts, databases, and infrastructure than someone who spends every day building polished mobile apps. Zone 2 Trainer became a useful test of how far AI coding tools can take someone...

The AI product climate this week was weirdly clarifying. I am still not ready to move my development workflow back to Gemini after the rate-limit frustration, but I also have to admit that Google shipped a few things that made me pay attention again. At the same time, OpenAI reminded me that even the best developer tooling can feel unnecessarily confusing when the billing model sprawls across too many surfaces. Then OpenRouter raised a serious round of funding and made the case for a future where developers do not have to bet everything on one model vendor. That is the pattern I keep coming back to: the tools are getting better, but the surrounding product experience still matters just as much as the model. Gemini Is Back in My General-Purpose Rotation I have complained plenty about Gemini rate limits. The stricter enforcement hit the wrong nerve for me because it made the product feel unpredictable at exactly the moment I wanted to rely on it more. Once a tool makes me ration m...