{"slug": "weekly-dev-log-2026-w13", "title": "Weekly Dev Log 2026-W13", "summary": "A developer focused on iOS development with SwiftUI and tested a workflow using Codex to improve efficiency. They created structured design files in Figma, implemented a SwiftUI fretboard UI via Codex, and built a workflow with AGENTS.md files, Notion MCP, and Skills to make Codex act as a development partner. The developer also explored how AI agents can be better integrated into development processes.", "body_md": "##\n🗓️ This Week\n\n- This week, I mainly focused on\n**iOS development**📱.\n- I also tested a small workflow using Codex: organizing ideas, creating a design file, and implementing a SwiftUI prototype. Through that process, I realized that\n**Codex can be very helpful for making iOS development more efficient**💡.\n-\n**Until now, I had been using Codex in a rather ad hoc way**, sending instructions only when I wanted it to handle a specific task😂.\n- This week, I started thinking about\n**how to make that workflow more efficient, and how to make Codex act more like a development partner** that understands the project in the same way I do🧐.\n- As I explored this, I realized that\n**I needed to review my Codex settings, build better rule files for giving instructions to Codex, and rethink how I use Codex in my own workflow**🛠️.\n- Working through all of this slightly changed\n**the way I think about AI agents**🤖.\n\n##\n📱 iOS (SwiftUI)\n\n- I organized the UI structure for ToneDrill mini in Figma.\n- I created a structured design page and a small Mini Design System in Figma.\n- I summarized colors, typography, buttons, fretboard parts, spacing, and radius rules.\n- I updated the implementation note so Codex could understand the design direction.\n- I asked Codex to implement the SwiftUI fretboard UI based on the Figma design and implementation note.\n- I changed the instruction so app colors would be managed as Color Sets in\n`Assets.xcassets`\n\n.\n- I reviewed the SwiftUI code for ToneDrill implemented by Codex.\n- I asked Codex questions to better understand the implementation.\n- I checked how the fretboard UI is drawn and updated in SwiftUI.\n\n##\n🆙 Updating my workflow\n\n###\n🤖 AI agent workflow\n\n- I worked on setting up a more structured workflow for using Codex in my development projects.\n- I created\n**a global **`AGENTS.md`\n\nfile for Codex (`~/.codex/AGENTS.md`\n\n).\n- I created\n**a project-level **`AGENTS.md`\n\nfile at the root of the repository (`repo-root/AGENTS.md`\n\n).\n- I separated the responsibilities of the global\n`AGENTS.md`\n\nfile and the project-level `AGENTS.md`\n\nfile.\n- I created\n`docs/codex/TaskIndex.md`\n\nas a task management document. In it, I organized both immediate tasks and mid- to long-term tasks, and defined a consistent format for writing them down.\n- To reduce the need to repeatedly explain the project background, past work, and current tasks every time I move to a new chat, I decided to\n**build a workflow and create a mid- to long-term project memory in Notion** for the work and lessons I develop with Codex.\n- I created a Notion page called\n`AI Operating Manual`\n\nso **AI can summarize the project overview, what I learned, mistakes, improvements, fixes**, and any rules that should be added to the project-level `AGENTS.md`\n\n. I connected it through **Notion MCP** so Codex can update it.\n- I organized this entire process as a workflow.\n- At the start of a chat, Codex should read the\n`AI Operating Manual`\n\nNotion page, then read `AGENTS.md`\n\nand `TaskIndex.md`\n\n, so it can start working with the same understanding as me. I registered this as a Skill so I can call it with a single phrase.\n- At the end of a work session, Codex should summarize what was done and organize what should be saved as mid- to long-term project memory in the\n`AI Operating Manual`\n\nNotion page. I registered this as another Skill so I can call it with a single phrase.\n- Finally, I registered a separate Skill that can write the summarized content to the\n`AI Operating Manual`\n\nNotion page. I separated this from the summary Skill so that I can review the content first and decide what should actually be saved as mid- to long-term project memory before running the write operation.\n\n#\n💡 Key Takeaways\n\n##\n📱 SwiftUI Learning\n\n-\n**By asking Codex to implement the SwiftUI UI based on the Figma UI structure and the small design system page**, I was able to **get a UI that was very close to the design I created in Figma**.\n- I learned that\n**Codex sometimes adds many custom style definitions in the code**, even in areas where SwiftUI’s default styling system would be enough. I need to decide more carefully which parts should rely on **SwiftUI’s default styles** and which parts should be customized.\n- Through this implementation, I learned that I need to\n**separate the parts where I want to take advantage of SwiftUI’s built-in default styles from the parts where I want to apply my own custom design**.\n- I also noticed that\n**Codex tends to implement SwiftUI previews using older syntax**. This helped me understand the difference between older preview code and the newer style I learned in the SwiftUI tutorials.\n- I also learned that if I want colors to be managed in\n`Assets.xcassets`\n\n, I need to explicitly tell Codex to do that.\n- I learned that I should add rules for Codex so that the same kind of issue does not happen in the next implementation.\n- I learned how SwiftUI draws horizontal and vertical lines using\n`.frame(width:height:)`\n\nand `.position(x:y:)`\n\n.\n- I learned that\n`.position(x:y:)`\n\nplaces the center of a view.\n- I learned that a parent view’s\n`@State`\n\nchange causes SwiftUI to recalculate the parent view’s body.\n- I learned that child views are updated when new values are passed from the parent view.\n- I learned that\n`@ViewBuilder`\n\ndoes not trigger UI updates by itself.\n-\n`@ViewBuilder`\n\nis used to write conditional view logic, such as showing or hiding a view depending on a condition.\n\n##\n🆙 Updating my workflow Learning\n\n###\n🤖 AI agent workflow\n\n####\n🐛 How My Workflow Changed\n\n- Previously, I had to explain in each prompt who I am, how I prefer to work, and what kind of reasoning I expect from Codex. For example, I often had to explain my background and say things like, “Because I think this way, please do not do that.” By organizing this information in advance,\n**I no longer need to repeat those explanations every time, and the direction of Codex’s responses is now much closer to what I expect**.\n- I can now\n**start talking with Codex about the task I want to work on right away, without wasting time aligning the process**. Before, when I felt that Codex was moving in a slightly different direction, I had to explain the project background, current progress, and what I wanted to do next in order to bring Codex back to the same understanding.\n- By formatting the workflow,\n**I no longer feel unsure about what information I should give Codex when working on any project**. I created a system that lets Codex reach a consistent level of understanding with a minimal prompt. Because I also defined a clear way to operate Codex in order to protect that system, I feel that I can work with it more confidently and without hesitation.\n\n####\n📝 What I Learned From This\n\n- I felt that\n**in order to use an AI agent effectively**, I need to **clearly tell it who I am and what kinds of actions or workflows I prefer**.\n- At the same time, I realized that\n**this kind of workflow becomes possible when I clearly define what I will delegate to the AI agent and what I will handle myself**. In my workflow, the AI agent should not immediately start working on a task. Instead, it should first propose a plan, and then I, as the human user, review and approve it before the task is executed.\n- This feels\n**similar to the relationship between a manager and a team member**. In other words, I felt that **using an AI agent effectively depends on how well the human can act as a capable manager**. I also realized that **the skills used to manage work in real-world teams can be useful when working with AI agents**.\n-\n**For regular applications that are used simply as tools**, **the important skills are things like understanding the features** and knowing how to use the application well.\n-\n**I found it interesting that the skills needed for AI agents are different** from the skills needed for ordinary tools. This difference feels like one of the unique characteristics of AI agents.\n\n#\n🚀 Next Week\n\n- Continue developing the top menu screen for ToneDrill.\n- Organize the UI adjustment points for the portfolio site implemented by Codex in Notion, then start making small UI refinements.\n- Continue working on the AI Security Learning Path.\n\n#\n🌈 Goals for This Year\n\n##\n📱 iOS (SwiftUI)\n\n- Build a solid foundation in SwiftUI and create at least one iOS app.\n\n##\n🌐 Web Development\n\n- Continue posting learning logs on Dev.to and eventually turn them into a portfolio site using React Router v7.\n\n##\n🔐 Security (TryHackMe)\n\n- Continue learning cybersecurity on TryHackMe.", "url": "https://wpnews.pro/news/weekly-dev-log-2026-w13", "canonical_source": "https://dev.to/umitomo-lab/weekly-dev-log-2026-w13-51o0", "published_at": "2026-07-10 21:08:52+00:00", "updated_at": "2026-07-10 21:43:29.375286+00:00", "lang": "en", "topics": ["developer-tools", "artificial-intelligence", "ai-agents", "ai-tools", "large-language-models"], "entities": ["Codex", "SwiftUI", "Figma", "Notion", "ToneDrill", "Notion MCP"], "alternates": {"html": "https://wpnews.pro/news/weekly-dev-log-2026-w13", "markdown": "https://wpnews.pro/news/weekly-dev-log-2026-w13.md", "text": "https://wpnews.pro/news/weekly-dev-log-2026-w13.txt", "jsonld": "https://wpnews.pro/news/weekly-dev-log-2026-w13.jsonld"}}