LLMs: a practical accelerator for how I learn, design, and ship software.

I first discovered GPT-2 in January 2021. At the time, it felt like a strange preview of a different way to learn: ask better questions, compare explanations, and use the feedback loop to understand software faster. Over time, that shifted from curiosity into a real part of my workflow. LLMs now help me increase output without skipping the engineering judgment that makes the work durable.

Plans

Break vague product asks into scoped steps, risks, and implementation order.

Tests

Identify useful coverage before and after feature work, especially around behavior changes.

Features

Move from architecture to code while keeping the surrounding system and user experience in view.

Implementation partner

Codex and Claude Code

I use Codex and Claude Code to turn product goals into implementation plans, targeted tests, and working feature changes. It is especially useful when I need to reason across a codebase, preserve existing patterns, and keep professional and personal projects moving with more structure. The value is not just speed; it is having another system help pressure-test assumptions.

Wireframes and decks

Claude Design

For early product direction, I have used Claude Design to create wireframes that become the basis for starting points in the UI. I have also it to create slide decks for stakeholders, translating rough product ideas into something concrete enough to discuss.

Asset systems

OpenAI, Gemini, Recraft

Recently, we used OpenAI, Gemini, and Recraft to build an asset generation system for a national print vendor. The system combines model output, creative constraints, and production needs so generated assets can move through a real print workflow.

Safeguarding Output

In the asset generation system, we utilize OpenAI's omni moderation for multi-provider model output. That layer matters because professional AI systems need more than generation. They need guardrails, review paths, and predictable behavior when creative output becomes part of a customer's workflow.