Willison expands condense-json with structural replacements for LLM workflows
Simon Willison released condense-json 1.1, expanding the tool's capability to handle non-string replacements and object-based merge operations. The update emerged from integrating the library into LLM, his command-line interface for large language models, revealing practical gaps in the initial design. For developers building LLM-adjacent tooling, this signals how real-world AI workflows expose missing abstractions in JSON handling, particularly when compressing token-heavy payloads for API calls. The iterative refinement reflects the maturing ecosystem around LLM interaction patterns.
Modelwire context
ExplainerThe real story isn't the feature additions themselves, but that they emerged from integrating condense-json into LLM's actual workflows. This reveals how theoretical tool design breaks under production constraints, forcing abstractions to evolve.
This connects directly to the CompressAgent research from early August, which quantified the reliability tradeoffs of prompt compression in tool-using agents. That work showed compression methods matter and failure modes are method-dependent. Willison's 1.1 release is the practical answer: when you compress token-heavy payloads for API calls, you need richer merge semantics and non-string replacements to preserve agent control logic without losing safety guarantees. The datasette-apps 0.2a0 release from the same week reinforces this pattern, treating agents as first-class operators whose infrastructure needs differ from human users. Together, these releases show the ecosystem moving beyond generic compression toward agent-aware tooling.
If the next LLM release (expected within weeks based on Willison's cadence) includes condense-json as a default compression strategy for system prompts, that confirms this is moving from optional utility to core infrastructure. If it doesn't, the tool remains niche despite the capability gains.
Coverage we drew on
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsSimon Willison · condense-json · LLM
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. Simon Willison originally reported this story as “condense-json 1.1”. The full content lives on simonwillison.net. If you’re a publisher and want a different summarization policy for your work, see our takedown page.