Coding with AI
Coding with AI › Sample Project Walkthrough

Worked Example

How this kind of project comes together

A recommended end-to-end path for creating a digital resource with AI and human collaborators—from discovery through deployment and maintenance. This is a sample sequence you can adapt to a website, app, store, portfolio, internal tool, marketing experience, lesson, resource, or other digital project.

Open this project's Style & Component GuideOpen Prompt Library
How to use this page: Terms marked * open their plain-English resource definition. Items marked open the project-specific Style & Component Guide. The sequence below is organized as a recommended workflow.
Cross-cutting requirements: Accessibility*, privacy/security, and maintainability are considered throughout the workflow—not only during final verification. Requirements are identified during discovery, reflected in standards and architecture, checked during implementation, and verified before release.
1

Stage 1

Understand the need and the people

Start with a needs analysis* rather than with code. Identify who the experience is for, what they need to accomplish, where and how they will use it, accessibility needs, device conditions, organizational constraints, and what success means.

This sample: A searchable reference for people across experience levels who want to work more effectively with AI and human collaborators on digital projects. It must work for newcomers without limiting advanced users.
Example direction to AI
Conduct a needs analysis before designing or building. Separate known facts, assumptions, unknowns, and AI inferences. Ask only the questions that could materially change the solution. Do not build yet.
2

Stage 2

Turn the need into a project frame

Create a project/build brief* that defines purpose, audience, project type, major tasks, scope, non-goals, constraints, success measures, and the desired first deliverable. Add acceptance criteria* so “done” has a concrete meaning.

This sample: A browser-based resource library with guided navigation, global search, ASK, reusable prompt copy buttons, and a structured source of truth for resources.
Example direction to AI
Using the approved needs analysis, create a concise project brief. Identify unresolved decisions and risks. Recommend the architecture and first representative deliverable; do not build the full system yet.
3

Stage 3

Establish the visual, content, and interaction rules

Define the design system*, content style guide*, typography, colors, spacing, interaction states, accessibility expectations, and what is Locked, Controlled, Flexible, or Approval Required. Record approved names in a project style & component guide*.

This sample: Typography, color tokens, cards and navigation, and copy-prompt controls are documented before adding large amounts of content.
Example direction to AI
Before scaling, define typography, colors, spacing, interaction states, accessibility requirements, voice/tone, and reusable component rules. Classify each rule as Locked, Controlled, Flexible, or Approval Required.

See the related project style rule

4

Stage 4

Design the information architecture and source of truth

Organize the experience so people can reach information quickly. Decide the hierarchy, section indexes, cross-links, search behavior, and a durable source of truth*. Keep content separate from presentation where practical.

This sample: The home page routes to Start Here, Design & Interface, Work with AI, Code & Integrate, Files & Projects, Review & Fix, Ship & Maintain, AI Words in Plain English, Prompt Library, and All Resources. The resource data lives in resources.json and can appear through multiple navigation paths without duplicating the source content.
Example direction to AI
Recommend an information architecture that lets a first-time user reach useful information within about three choices. Keep one source of truth for resource content and surface it through multiple relevant paths.
5

Stage 5

Name reusable blocks and components before multiplying them

Create approved blocks/components*, a component library*, and stable names*. A card*, for example, is a contained information unit; a block/section can contain several cards.

This sample: navigation cards, resource cards, prompt blocks, and the floating ASK control are repeatable patterns rather than individually redesigned objects.
Example direction to AI
Identify repeated layout patterns and turn them into reusable named components before creating many instances. Use approved components unless a new pattern is explicitly authorized.

See the related project style rule

6

Stage 6

Prototype one representative experience

Use prototype first, scale second*. If structure is uncertain, a wireframe* may help; if a working version is cheap to change, build one representative implementation and stress-test it before scaling.

This sample: Validate one section page, one resource card, one prompt-copy interaction, search behavior, and ASK behavior across phone/tablet/desktop before treating the pattern as approved.
Example direction to AI
Build one representative implementation with realistic content and edge cases. Test the pattern before generating the remaining pages or cards. Do not scale until the baseline is approved.
7

Stage 7

Review independently, then synthesize

Run specialist reviews and first-time-user simulations. A multi-agent review panel* can include User Experience, User Interface/visual design, accessibility, content, technical quality, and security perspectives. Synthetic users* are useful for finding likely friction, but their results are hypotheses—not real human usability evidence.

This sample: Review navigation, terminology, click/tap affordances, mobile behavior, search, copy buttons, ASK, accessibility, and likely novice confusion before wider rollout.
Example direction to AI
Have specialist reviewers evaluate independently before synthesis. Record issue, severity, confidence, and evidence. Then run diverse synthetic first-time-user task walkthroughs. Treat simulated-user findings as hypotheses, not human-participant evidence.
8

Stage 8

Implement the approved changes in the working project

Once the design is approved, tell AI to implement/apply the changes* in the working project. This is different from deploying*. Set the authority boundary clearly so AI does not commit, push, publish, or deploy unless authorized.

This sample: HTML, Cascading Style Sheets (CSS), JavaScript, structured resource data, and server-side ASK behavior are updated in the working project first.
Example direction to AI
Implement the approved changes in the current project files. Run the appropriate checks and report what changed. Do not commit, push, publish, or deploy without approval.
9

Stage 9

Verify behavior, accessibility, responsiveness, and security

Do not confuse generated with verified. Use generated → implemented → tested → verified → approved → deployed* as separate states, and define a Definition of Done*.

This sample: Check links, search, copy buttons, ASK, phone/tablet/desktop layouts, keyboard/focus behavior, reduced motion, and server-side secrets. If an AI service is used, an API key* is stored securely—not in browser code, screenshots, or documentation. An environment variable* is a broad configuration concept; for this Cloudflare example, the API key is stored as a Cloudflare Secret*, not as an ordinary plaintext variable.
Example direction to AI
Verify the implementation against the acceptance criteria and Definition of Done. Check supported viewports, keyboard behavior, accessibility, links, error states, security boundaries, secrets, and regressions. Report evidence, not just “done.”
10

Stage 10

Deploy only after approval

After verification and approval, deploy* the approved build to its target environment. Confirm production configuration, domains, protected secret bindings, logs, and rollback options. A credential is not safe merely because code is called “server-side”; verify how the value is stored, who can access it, and that it cannot reach browser/client code.

This sample: The static resource pages and assets are served with a server-side worker endpoint for ASK. Production secrets use protected server-side secret bindings and are never sent to browser/client code.
Example direction to AI
Prepare the approved build for production. Before deployment, confirm production configuration, secrets, domain/HTTPS behavior, monitoring, and rollback. Deploy only after those checks pass.
11

Stage 11

Maintain a recoverable project

Keep the project portable and reversible. Use version control, a checkpoint*, handoff notes, and clear records of decisions. Update the shared style/component guide when a new approved pattern is introduced.

This sample: Future additions should reuse the documented typography, tokens, components, prompt patterns, information architecture, and security rules rather than reinventing them.
Example direction to AI
Before a substantial change, create a checkpoint and identify the rollback path. After approval, update the handoff and style/component guide if the project vocabulary or approved patterns changed.

See the related project style rule

Core pattern: Discover → Define → Standardize → Structure → Prototype → Review → Implement → Verify → Approve → Deploy → Maintain.