Intel / article

Introducing Laguna S 2 1

Source: poolside.ai
Author: Poolside
Date: Published Jul 21, 2026

Context

Poolside, which builds coding AI models, says it pointed its Laguna S 2.1 model at the software it uses to train, evaluate and run its models, then let the model work in an automated loop. Poolside first added timing and memory measurements so the model could see where time and memory went. The rules were one approach at a time, a measurement after every change, and keeping only changes that measurably helped, with each result fed back to the model (case study 2).

Poolside reports that over several hours this made the software 5.2% faster with about 70% lower memory allocation. Fixes included replacing repeated string joining with buffers, caching repeated computations and sizing lists exactly in advance. Once speed gains were hard to measure, the model shifted to memory, which was measured more reliably. This is one company-described example, and Poolside says its benchmarks are not a full production test. It checked the final run with Go's race detector and go vet, and tested the final artifact.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source poolside.ai