Ephemeral Agent Workbenches

Wed Jul 08 2026

When I work with agents, my first instinct is usually to “just prompt” through the task. That is fine for many coding-adjacent tasks, but not everything I do can be easily codified, and I bet the same thing happens to you!

I’ve explored how to make agents do subjective work better, and I think there is another underrated approach to using agents for subjective tasks, or tasks where the goals are fuzzy and you’d like to keep the final call: build local and ephemeral workbenches.

The idea is to have agents produce small pages/apps for the task at hand. For me, the clearest recent example is a grant review workbench I built. Instead of asking a model “which application is better?”, I got it to build me a workbench to help me better understand the task.

In the workbench, I could see and sort every application, add random tags or notes as I read through them, visualize them on a 2D map to spot duplicates or core applications, etc.

The useful part was the handoff. I could click a button to copy the state and ask my agent to do things like complete the labeling from my manual labels, recluster, add a new field to applications, suggest due diligence questions, etc. Intent did not have to fit in a prompt: it could also come through chat, images, or the selections, tags, and notes already in the workbench.

This gave both of us a high-bandwidth way to share state and made the feedback loop faster than chat. Chat alone is a bad interface for this kind of work because too much state stays implicit, or even hidden from you. With the workbench, I could inspect both the applications and the changes the agent wanted to make.

I have also applied this to other messy tasks: sorting and tagging invoices, closing duplicate issues, exploring random datasets, understanding a diff, and curating a reading/listening list. In all of them, I do not want the model making dumb decisions for me. I want a small shared space where I can make sense of the problem, inspect and change its state, and invite the agent in as an extra pair of hands.

Some things that have worked well for me:

• Start with one simple vanilla HTML file, and grow it when needed.
• Put a short note in the file with the goal and how you want to use it.
• Add save and restore features. Clipboard or URL-encoded data is usually enough.
• Add a similarity map when the task depends on judgment. Embeddings are still awesome.
• Keep it interactive (click, tag, compare, add notes, sort, …).

So, when facing a larger-than-usual task, I now ask for a small workbench before I ask for the answer or solution. It also helps me learn!