How to use Codex to build a day-trip planner
Christo Roberts corrected Codex's initial itinerary editor, then shipped Daytripper with deterministic activity scoring, saved plans and preference checks.
letsgochristo.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build [small planner] using [curated data] and explicit [budget, timing and selection constraints]. Show the central user loop before expanding features. Test every supported preference combination, exercise save and navigation flows, and report stale-data limitations before publication.
Reach people exploring AI agents
See placements →What happened
Daytripper deployed to Cloudflare Workers with curated activities and exhaustive preference-combination tests.
Reported by Christo Roberts on BlogHow the agent did it
- Initial brief produced wrong itinerary editor.
- Author clarified city-selection generator objective.
- Codex built preference scoring, local saves and sharing.
Where a person stepped in
Author corrected product definition and found broken profile/save affordances.
Christo Roberts learned Codex by building Daytripper, a small day-trip planner. The intended product let users choose a city, budget, time window, pace, interests and mood, then receive a matching day plan. The first prompt was underspecified: Codex produced a polished itinerary editor that missed that central generator loop.
Roberts inspected the result and corrected the product definition. The next implementation used curated activities and a deterministic TypeScript scoring engine, with stop swapping, extra activities, local saving and a shareable text version. It deliberately made no runtime LLM calls and needed no external activity API.
That bounded design made all 1,296 supported preference combinations testable. The reported suite checked chronology, city grounding, budget safety and distinct extras. It did not replace product inspection: Roberts later found a profile button that did nothing and a save flow without a useful return path. The repository's QA report recorded problems that were fixed before launch.
Daytripper was deployed to Cloudflare Workers with public source. Its limitations include a curated activity pool rather than fresh data.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Blog · Christo Roberts
Frequently asked questions
Why was the first version wrong?
The initial prompt led Codex to build an itinerary editor rather than the intended city-and-preference day-plan generator.
Does Daytripper call an LLM when planning?
No. The reported app uses curated activities and a deterministic scoring engine, with no runtime LLM or external API calls.
What did the exhaustive checks cover?
The source reports 1,296 supported combinations checked for chronology, city grounding, budget safety and distinct extras; product inspection still found interface defects.
More details
Remaining steps
- Inspected, critiqued, tested and refined before launch.
About this account
Codex implemented the clarified planner, scoring and saved-plan behavior; the author inspected product flows and corrected the initial wrong objective.
Primary first-person account opened and checked on 2026-10-14. No fresh data or runtime LLM; initial polished output missed product goal. Outcomes and test claims are attributed to the source and were not independently reproduced. Prompt is suggested by Agents In Use, not the author's quoted prompt.
Credit: Christo Roberts on Blog
Source published 1 Aug 2026
Source checked 14 Oct 2026
Tools and topics
Codex · Next.js · TypeScript · Cloudflare Workers · Node.js
Day trips · Product review · Deterministic planning
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.