Source: https://mayphus.org/ai-workflow/ Title: How I work with AI Metadata: {"category":"note","date":"2026-07-16T02:42:23Z","id":"ai-workflow","kind":"article","language":"en","locale":"en","route":"/ai-workflow/","slug":"ai-workflow","tags":["ai","workflow","codex"],"type":"article"} # How I work with AI This is my July 2026 workflow. In October, I began a separate [iPad-first workflow experiment](https://mayphus.org/ipad-first-ai-workflow/); that later account describes its own limits and does not establish that every computer or app has been replaced. A separate [coordination experiment with dot and Muse](/ai-coordination-experiment/) records complementary access, a phone route that stopped, and the open question of learning from another assistant. In July, I used ChatGPT Desktop as the command center for thinking, investigating, creating, and operating systems. AI is the interface; the durable result stays in files, Git history, and Mayphus. ## What enters the system Most work can begin with voice. I can talk while walking or working on hardware. ChatGPT helps turn an unfinished thought into a question, a decision, a note, or a bounded task without waiting for a formal writing session. [Engineering in small moments](https://mayphus.org/notes/engineering-in-parenting-moments/) is one example. - For investigation, AI browses sources, compares evidence, and challenges assumptions before producing a useful conclusion. - For physical experiments, voice, photos, and YouTube livestreams capture what really happened. AI can help identify parts, record questions, and turn the session into an article or repair note. The [camera investigation workflow](https://mayphus.org/entries/chatgpt-camera-investigation-workflow/) shows this in practice. - For software and operations, Codex uses the repository, terminal, and tests to change the actual system. A livestream is therefore more than content. It is a record of real-world exploration that can later become searchable knowledge. ## The flow 1. Voice, video, and questions 2. ChatGPT Desktop 3. Investigation or task 4. Codex 5. Git 6. Mayphus ChatGPT is where I direct the work. The browser gathers evidence, Codex makes and verifies changes, Git records what actually exists, and Mayphus preserves and publishes the useful result. These tools matter, but none of them needs to be the center on its own. I wrote more about [ChatGPT Desktop as an AI superapp](https://mayphus.org/ai-workflow/#section-2026-07-12-ai-super-app). ## What remains my responsibility AI can choose local implementation details. I keep responsibility for purpose, public and private boundaries, acceptable complexity, and the definition of done. This keeps separate AI sessions moving in the same direction. ## My operating rules - One task should produce one clear, reviewable outcome. - Important claims need evidence; every change needs verification. - Repeated instructions become repository guidance or a reusable skill. - Each completed task should leave the system easier to understand. [Concrete before abstraction](https://mayphus.org/ai-workflow/#section-2026-07-14-concrete-before-abstraction) describes the same habit. The conversation is temporary. The artifacts are the source of truth. ## What I am learning next [From hands-on operations to AI-assisted systems engineering](https://mayphus.org/ai-assisted-systems-engineering/) connects my earlier delivery and repair experience to the abilities I am developing now, with proposed sandbox exercises and a learning checklist. ## Dated working notes These reflections record how the workflow developed in July 2026. They are preserved as historical notes, not current instructions or new permission to act. ### 2026-07-10 — restartable work [Infrastructure](https://mayphus.org/infrastructure/) The smallest useful project note records where the work stopped, the next physical action, and the evidence already captured. That is enough to make tomorrow easier. ### 2026-07-11 — proof before claims A personal site is more useful when it shows working artifacts and operating evidence instead of describing ambitions. Build the proof, then write the sentence. ### 2026-07-12 — ai super app ChatGPT desktop is starting to feel like the AI version of Emacs: one extensible environment with a terminal, browser, code, files, and conversations together. The important part is not having every tool in one window. It is that my operations and working state can become shared AI context, so I spend less time switching apps and explaining where the work stopped. ### 2026-07-12 — one public thread Short ideas should have one quiet public home. They do not each need a new page, project, or polished conclusion. A dated thread can show the thought while it is still moving. ### 2026-07-14 — concrete before abstraction Coding, architecture, and other high-level work can be important, but they can also abstract the problem until I lose sight of what actually needs to be solved. I want to begin with the concrete problem, make the smallest useful thing, and only add abstraction when the work asks for it. ### 2026-07-15 — faster codex workflow I’m starting to optimize my Codex workflow. Instead of making every turn follow a heavy default sequence of pull, change, test, and review—even when a small change takes only a moment—I’m moving toward a hierarchy of iteration, review, and ship. The depth of the workflow should match the size and risk of the change, so a small task does not automatically take two or three minutes. I’m also starting to use Git worktrees so independent work can run in parallel. ### 2026-07-17 — optimizing someone elses workflow When changing a system that someone else uses, I should preserve the behavior they recognize and ask before changing the interface. ### 2026-07-26 — website as working memory I do not want my website to be a separate job. It should become a durable public memory that participates in my AI work: when an idea or investigation becomes useful, I can ask the assistant to turn it into a concise note, keep private details out, and create a review before anything goes public. The website is not the work itself. It records the proof and lessons from real work.