Emily’s Brain Case Study 4 — AI Code Generation

Learning to Direct AI Code Generation

How a practice exercise in AI report generation became a repeatable workflow for executive-style analytics.

The assignment I gave myself

I set myself a practice exercise: take an AI report-generation flow end to end — from instructions to a polished, publishable artifact — and document the process as a learning guide. I used Codex for the generation.

The problem I was solving

Knowing how to prompt is only half the skill. The other half is directing: giving the AI a clear assignment, checking the output against the requirements, and shaping the result into something stakeholder-ready. I wanted to practice that full loop on a realistic analytics report.

The prompt and instructions I provided

I asked for an analytics report on video game releases covering 2016–2026, styled with a blue-and-orange corporate theme: executive KPI cards, a release-volume trend chart, a genre-mix breakdown, concise takeaways, and a sample data table.

How I supplied the data and report requirements

The dataset is synthetic demonstration data — modeled video game releases, not real business reporting and not connected to any employer system. My requirements were: four headline KPIs, a year-by-year trend visualization, a genre share breakdown, short executive takeaways, and a representative sample table.

How the AI generated and formatted the report

The flow returned a single self-contained HTML file: four KPI cards (total releases, peak year, top genre, average rating), an SVG line chart of release volume by year, horizontal bars for genre mix, three insight cards (platform leaders, quality pattern, market takeaway), and an eleven-row sample release table — all in the requested blue-and-orange theme.

The final output

The report below is the generated artifact, embedded in full and published as-is except for one change: I removed the vendor wordmark and kept the color palette. There is no proprietary content, customer data, or internal-system detail in this case study.

What worked, what I learned, what I would improve

What worked: a single generation pass produced a complete, polished, self-contained artifact — layout, charts, and copy in one file.

What I learned: the review pass matters as much as the prompt. Checking branding, data labels, and the fine print before publishing is part of directing, not an afterthought.

What I would improve: record the exact prompt next time, so the workflow is reproducible word for word.

Assumptions

• All video-game data in this report is synthetic demonstration data. It is not real business reporting and describes no real market.
• The blue-and-orange visual theme is inspired by publicly visible brand patterns. This case study has no affiliation with or endorsement by Fiserv, and it contains no confidential, proprietary, or customer information.
• I did not record the exact prompt text. The prompt description above summarizes my instructions; it is not a verbatim quotation.

Sources & resources

• View the full generated report — the complete self-contained HTML artifact, produced with Codex’s report-generation flow on September 19, 2026. All data shown is synthetic demonstration data.
• This case study is part of Emily’s Brain, my AI portfolio of real problems solved with AI and documented step by step.