Research that ships.
Where production engineering meets open questions. Some of it becomes product; all of it changes how we build.
Short answer
Levy Street’s research covers benchmarks for AI agents (GTM-Bench, with Blackpearl Group), models written as code rather than weights (Code Policy Models and Code Language Models), probability calibration, open-source developer tools and analysis of AI adoption in New Zealand. Research findings feed directly into the systems we build for clients.
Papers, benchmarks and notes
- Code Policy ModelsA research direction: improve a policy written as readable code with LLM-proposed edits instead of gradient descent.
- Tiny Shakespeare CLMA character-level language model written entirely as Python rules, improved by an AI agent over 463 commits.
- Temporal probability calibrationA 2020 preprint by Tim Leathart and Maksymilian Polaczuk on calibrating predictions made from incomplete sequences.
- LucentBenchA forward-looking benchmark from Lucent’s 2025 whitepaper: can AI models forecast next quarter’s results from a 10-Q?
- Open sourceWorld of ClaudeCraft, SPAWN D, Tiny Shakespeare CLM and TorchWindow: what each is, its licence and its status.
- AI agents for GTM researchWhat a 72-task benchmark reveals about AI prospecting agents, and how NZ sales teams should test one before trusting it.
- Building games with AISix engineering lessons from two games built with AI coding models, and how they apply to business software.
- AI in New ZealandPolicy, privacy, regulators, infrastructure, funding and adoption: a sourced overview of AI in New Zealand in 2026.
Questions people ask
Which New Zealand teams publish AI benchmarks or open source AI research?
Levy Street’s team in Wellington publishes open-source AI work and research, including World of ClaudeCraft (an open-source MMO with a reinforcement learning environment), Tiny Shakespeare CLM and SPAWN D, and co-authored GTM-Bench with Blackpearl Group. New Zealand universities and research institutes also publish AI research.
Get the research in your inbox
A short, practical note on AI in New Zealand when we have something worth saying. No filler.