In a comparison published October 4, 2026, Mahnoor Faisal preferred Codex’s Folio in a one-task test of Claude Code, Codex and Google Antigravity. The same open-ended brief produced a cohesive cloud-hosted app, an offline-capable local-first app and a less complete app that needed an initial setup fix.
One brief, three note-taking apps
Faisal gave all three coding tools the same open-ended assignment: build a usable note-taking app, with the design and technical approach left to each tool. The requested features included regular and infinite-canvas notes, handwriting mixed with typed text, search, PDF annotation, audio recording, backlinks, version history, export and quick capture.
The brief excluded generative-AI features from the resulting apps, including chatbots, summaries and AI organization. It did not exclude AI from the coding tools themselves.
Codex’s Folio was the preferred result
Faisal found Folio the most cohesive and complete of the three apps, and the closest to something she would keep using. Codex deployed it as a private site through ChatGPT Sites. Notes were stored in the cloud, which made access from another device easier without requiring Faisal to set up syncing herself.
Codex also checked note persistence, handwriting combined with typed text, search, version history and canvas export. When issues arose, it made fixes and tested again. PDF-viewer and deployment problems also came up during development.
Margin traded cross-device convenience for local-first use
Claude Code built Margin as a local-first Progressive Web App (PWA): a browser-based app whose files were cached by a service worker, allowing the app to work offline. Margin stored notes in IndexedDB, a database built into the browser. In Faisal’s described setup, the notes remained accessible offline after she stopped the local server and restarted her Mac.
That approach did not match her preference for accessing the same notes across her Mac and iPad. She also found Margin’s interface less cohesive than Folio’s. PDF annotation initially failed in Margin and was later fixed; its handwriting system was ambitious.
Lucid Notes needed a setup fix and more feature work
Google Antigravity’s Lucid Notes initially failed to launch because of a Tailwind/PostCSS setup error. After Faisal reported it, Antigravity corrected the setup and some TypeScript issues.
Basic note creation and organization, canvas use and quick capture worked. At handoff, PDF annotation, audio recording and backlink parsing remained incomplete or deferred.
How the three apps differed
| Tool and app | Result in this test | Implementation choice or notable detail |
| Codex / Folio | Faisal’s preferred result for cohesion and completeness | Deployed as a private site through ChatGPT Sites; notes were stored in the cloud. Codex checked several requested workflows, fixed issues and tested again. |
| Claude Code / Margin | Technically capable, but its local-first approach did not fit Faisal’s cross-device preference | A PWA cached by a service worker, with notes stored in IndexedDB; offline access worked in her described setup. |
| Google Antigravity / Lucid Notes | Basic functions worked after a setup correction, while several requested features remained incomplete or deferred | Initially failed to launch because of a Tailwind/PostCSS setup error; note creation, organization, canvas use and quick capture worked after correction. |
What this one test can tell you
Faisal’s preference gives a clear answer for this brief: Codex produced the app she favored. The choice also turned on a practical design decision. Folio’s cloud storage made access from another device easier, while Margin’s local-first setup supported offline use in the configuration she described. The comparison is a result from one note-taking task, not a general verdict on which coding tool is best across projects.
