The Design Philosophy of LensBench — Built for AI-Native Workflows

LensBench is a sequential lens design workbench. The biggest difference from other optical software is not an added chat box: starting from the data model, it is designed for people and AI changing a design together.

This is how we want you to work: you tell the AI "change the thickness of the last surface to a paraxial focus solve and see how much the total length changed". The AI first does a dry run and lays out the diff and the before/after focal length, back focal length, F-number and total length; it writes only when you nod; every number it reports carries an evidence id that traces back to the specific computation; and any step can be undone, with every change kept on the timeline.

Behind this are a few principles, each written up in its own post:

1. A design is just text: the AI can read it and write it, and changes are naturally reviewable patches; everything speaks Zemax
2. One command table: the desktop app, the command line and the AI use the same set of commands, so the AI has no hidden channel
3. See the diff first, then change the design: the confirm button always stays in your hands, and changes are undoable and traceable
4. Every number has a source: evidence ids and independently recomputed acceptance, never trusting "the optimizer says it got better"
5. What an AI-native workflow looks like: a real conversation, and three ready-made templates for diagnose-and-optimize, Zemax reconciliation and report material

What it can do now ​

Design, analysis and optimization of sequential imaging systems, plus file interchange and numerical reconciliation with Zemax. Multi-configuration and tolerancing are already usable; non-sequential, stray light and physical optics propagation are not done at this stage. The desktop app ships as an installer only; see Install. For the item-by-item scope, see Capabilities and scope.

What to do in the community ​

• Share your AI workflow: which model you used, how you prompted it, where the AI went wrong; see Share your AI workflow
• Reconcile: post numbers that don't match Zemax to the forum's "Zemax numerical reconciliation"
• Report problems: post to "Bug reports"; ones with design files go through confidential feedback, visible only to you and the development team. The forum bot reproduces and follows up automatically
• Make suggestions: post to "Suggestions and feedback", where everyone can vote