Skip to content

Pelican Rides a Bicycle

Below are two SVG solutions to the Simon Willison's benchmark: "Generate an SVG of a pelican riding a bicycle."

Claude Sonnet 3.5:

claude-3-5-sonnet-20240620.svg

Claude-Code:

pelican-bicycle.svg

What Changed?!

As agentic gen/ai exploded in 2025, it felt like Willison's benchmark became a solvable problem if we used these new agentic powers: skills, sub-agents, tool-use and more. Using Claude-Code, I built a pipeline that now solves this problem.

The Github repo github.com/0x6a77/pelican-rides-a-bicycle let's you replicate this work and see an example of Claude-Code skills with tool-calling.

How It Works

The Claude-Code skills pipeline orchestrates specialized capabilities:

  • Agentic/AI Orchestrates the pipeline
  • Image Generation: Flux diffusion model generates a bitmap of the prompt
  • Vector Tracing: autotrace converts the bitmap to SVG paths

The agent's job is to know which tool to use when. This is fundamentally different from asking a transformer to generate SVG vectors from scratch.

2026, The Year of Agentic Pipelines?

Some might argue this doesn't solve Willison's benchmark - it solves the problem his benchmark reveals. It shows transformers can't do one-shot compositional spatial reasoning. My pipeline shows that agentic gen/ai doesn't need to because it can orchestrate specialized tools instead. The pelican benchmark is valuable for tracking raw model capabilities, but to build systems that work today, we need orchestration.1

The right way forward with gen/ai is to embrace agentic gen/ai in 2026 in deeper and more interesting ways. Stay tuned for an expansion of this idea.

  1. (Stephen Wolfram's work on computational irreducibility suggests the pelican task may require a level of spatial reasoning that transformers can't do, so tool orchestration isn't just pragmatic, but potentially necessary. I plan to write about this in a future post.) ↩