TikZ Editor Is a WYSIWYG LaTeX Figure Tool Built Almost Entirely by Codex

A Show HN post hit the front page today featuring TikZ Editor, an open-source WYSIWYG editor for TikZ figures in LaTeX. The creator, Dominik Peters, described the project as "built almost entirely by Codex."

The reception was enthusiastic. The technical details are worth examining for anyone interested in what agent-built software actually looks like when it ships.

Last updated: June 23, 2026

What TikZ Editor Does#

TikZ is a widely-used LaTeX package for drawing figures in academic papers. It uses commands like \draw[->] (0,0) -- (1,2); to draw lines, shapes, and text. Academics typically code up figures by hand, tweaking coordinates and recompiling until things look right.

TikZ Editor solves this with a dual-pane approach:

Live source editing - you can edit the TikZ code directly
Visual manipulation - you can drag and resize elements on a rendered preview
Bidirectional sync - changes in either pane update the other

The key technical insight is that when you drag an element, the editor only modifies the coordinate numbers in the source code. It preserves line breaks, indentation, and everything else. The source never turns into unreadable machine-generated soup.

This is the kind of careful design that makes tools feel professional rather than gimmicky.

Why It Took an Agent to Build#

The creator's explanation in the Show HN is direct:

This approach essentially required reimplementing a large fraction of TikZ, which is the kind of task that no human would ever want to do.

TikZ is a complex package with loops, macros, and extensive syntax. Parsing it accurately enough to track source locations for every drawn object is not technically impossible - it is just tedious. The kind of tedious that stops projects before they start.

The argument is not that agents are smarter than humans at this task. The argument is that agents have a higher tolerance for grunt work. If a task requires implementing hundreds of edge cases with no intellectual novelty, an agent can do it while a human would give up.

Peters describes several "side quests" that came out of the project:

Converters from SVG, PowerPoint, and Ipe to TikZ
A re-implementation of the LaTeX hyphenation and line-breaking algorithm for multi-line nodes
A color picker using LaTeX's red!20!black color mixing notation

Each of these is a small, self-contained problem that an agent can solve without needing high-level architectural judgment.

What HN Is Saying#

The Hacker News thread is overwhelmingly positive, which is notable for a Show HN. A few representative comments:

Wow, this is really, really great. Congratulations on an excellent offering and piece of tech!

All STEM students and researchers from the world thank you.

The killer feature for me is not drawing TikZ visually, but being able to touch old TikZ without turning the source into generated-looking soup.

That last point is worth emphasizing. The editor respects existing code. It does not reformat or regenerate. It edits in place. This is what makes it usable for real academic workflows where papers are co-authored and version controlled.

Some commenters asked for related features:

Support for CeTZ (the Typst equivalent of TikZ)
Support for pgfplots (the TikZ extension for charts and plots)

These are reasonable requests that fit the same technical pattern: implement enough of the target syntax to enable visual editing without destroying source formatting.

The Case Study for Agent-Built Software#

TikZ Editor is interesting as an existence proof. It shows:

1. Agents can handle tedious implementation work. Reimplementing TikZ parsing is exactly the kind of task that agents excel at. The specification exists (TikZ has documentation). The success criteria are clear (does the rendered output match?). The work is repetitive.

2. Quality depends on the guiding vision. The editor is not a random pile of generated code. It has a coherent design - bidirectional editing, source preservation, converter ecosystem. That design came from a human. The agent implemented it.

3. Open source matters for trust. Agent-built software raises natural questions about code quality and maintainability. Publishing the source lets the community verify and contribute. It also makes the project sustainable - others can fix bugs and add features.

4. The "impossibly tedious" category is real. There are projects that never get built because the ratio of grunt work to interesting work is too high. Agents shift that ratio. Some previously impossible projects become possible.

The Limits#

The Show HN post does not describe any significant debugging struggles or agent failures. That is unusual. Most agent-built projects involve substantial human intervention - reviewing generated code, fixing subtle bugs, rejecting bad approaches.

It is possible that TikZ parsing happens to be an unusually good fit for agent work: clear specs, testable output, minimal ambiguity. It is also possible that the creator is understating the human effort involved.

Either way, the result is a functional, useful tool. The question of how much human effort was really required is interesting but secondary.

Practical Takeaways#

For developers considering agent-assisted builds:

Look for tedious-but-specified tasks. If the work is boring and the success criteria are clear, an agent can probably help. If the work requires judgment about trade-offs or user experience, you will need more human involvement.

Design before generating. TikZ Editor works because someone thought carefully about bidirectional editing and source preservation before writing any code. The agent implemented a design. It did not invent one.

Publish the source. Agent-built software benefits from transparency. If the code is good, publishing it builds trust. If the code has issues, the community can help fix them.

Match the tool to the task. The creator used Codex, which has particular strengths in code generation and transformation. Different agents have different strengths. Pick the one that fits your task.

My Take#

TikZ Editor is the kind of project that makes agent-assisted development feel real rather than hypothetical. It is not a demo or a proof of concept. It is a production tool that solves a genuine problem for a specific audience.

The fact that it was built "almost entirely by Codex" is noteworthy, but the more important point is that it works. Users do not care whether the code was written by a human or an agent. They care whether the tool does what they need.

This is the right frame for thinking about agent-built software. The question is not "who wrote the code?" The question is "does it work, can it be maintained, and does it solve a real problem?"

For TikZ Editor, the answer appears to be yes on all three counts.

FAQ#

What is TikZ Editor?#

TikZ Editor is an open-source WYSIWYG editor for TikZ, the LaTeX package used to draw figures in academic papers. It lets you edit TikZ code visually while preserving the original source formatting.

How was TikZ Editor built?#

The creator, Dominik Peters, states it was "built almost entirely by Codex." The agent handled tedious implementation work like parsing TikZ syntax and tracking source locations.

Does the editor reformat TikZ code?#

No. A key design choice is that visual edits only modify coordinate numbers in the source code. Line breaks, indentation, and other formatting are preserved.

Is TikZ Editor open source?#

Yes. The project is available as both a web app and a desktop application, with source code publicly available.

What other converters does it include?#

The project includes converters from SVG, PowerPoint, and Ipe to TikZ format, plus a re-implementation of LaTeX hyphenation for multi-line nodes.

Continue Reading#

How to Use GLM 5.2 and Other Custom Model Providers in Codex
Codex Now Encrypts Multi-Agent Prompts, Breaking Local Auditability
Codex Logging Bug Can Write Terabytes to Your SSD
OpenMontage Shows the Real Future of AI Video: Agents, Not Editors
Codex - plans, install routes, and alternatives on one page
OpenAI Codex course - the CLI, cloud tasks, and IDE extension, hands-on

Sources#