---
type: "post"
slug: "liquid-ai-agent-loops"
title: "Liquid AI agent loops"
description: "Liquid AI gave two coding agents a real deadline and a genuine gap in the tokenizer tooling landscape, then wrote up what the loop and its verification infrastructure had to look like."
retrieval_nugget: "Liquid AI gave two coding agents a real deadline and a genuine gap in the tokenizer tooling landscape, then wrote up what the loop and its verification infrastructure had to look like."
published_at: "2026-09-02"
updated_at: "2026-09-12"
record_date: "2026-08-25"
date_kind: "discovered_at"
topics: ["agent-retrieval","agents","ai"]
entities: ["Liquid AI"]
source_url: "https://liquid.ai/blog/agent-loops"
source_title: "Designing Loops for Production-Grade Work — Blog"
source_domain: "liquid.ai"
source_terms: ["Liquid","loops","tokenizer","Production-Grade","real"]
summary_word_count: 187
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:liquid-ai-agent-loops"
status: "published"
source_urls: ["https://liquid.ai/blog/agent-loops"]
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:e43323498ee2ff4ca11cde58582bd8bc459172fef52e0170bb92c786c3a478ea","reviewed_at":"2026-09-12T10:00:00Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":3}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Liquid AI published an engineering account on 18 August of an experiment it ran in late 2025: could coding agents autonomously solve a production-grade problem from scratch?","observed_facts":[{"text":"The experiment produced an open-source byte-pair encoding tokenizer trainer called toktoktok.","source_urls":["https://liquid.ai/blog/agent-loops"]},{"text":"The write-up covers specifying a goal for multi-domain experts and setting up verification infrastructure around the loop.","source_urls":["https://liquid.ai/blog/agent-loops"]}],"mechanism":"The team needed a BPE tokenizer trainer able to process trillions of tokens on a single machine for its own research on vocabulary size in edge models, and found the existing landscape thin: sentencepiece was optimised for non-BPE tokenizers and slow.","why_now":"Two agents using the best publicly available coding models at the time were given a real problem and a real deadline, and the output — a byte-pair encoding tokenizer trainer called toktoktok — is now open source.","implications":["So the task was not a benchmark exercise but a gap they would otherwise have had to fill themselves, which is the only setting in which autonomy can be judged honestly."],"evidence_boundary":"This is a single retrospective experiment on one problem with models from late 2025, and the account is the operator's own without independent replication.","watch_conditions":["The condition to watch is whether the loop is re-run with current models against a second real gap, because a technique demonstrated once on a favourable target is a case study rather than a method."],"related_records":[{"url":"https://newruntime.com/posts/agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts","relation":"Prior coverage of harnesses and accepted artifacts as the unit of work."},{"url":"https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource","relation":"Existing pattern on verification capacity as the constraint."},{"url":"https://newruntime.com/topics/agent-harness","relation":"Topic hub for agent loop design."}]}
routes: {"html":"https://newruntime.com/posts/liquid-ai-agent-loops/","markdown":"https://newruntime.com/posts/liquid-ai-agent-loops.md","json":"https://newruntime.com/posts/liquid-ai-agent-loops.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# Liquid AI agent loops

## Retrieval answer

Liquid AI gave two coding agents a real deadline and a genuine gap in the tokenizer tooling landscape, then wrote up what the loop and its verification infrastructure had to look like.

Liquid AI published an engineering account on 18 August of an experiment it ran in late 2025: could coding agents autonomously solve a production-grade problem from scratch? Two agents using the best publicly available coding models at the time were given a real problem and a real deadline, and the output — a byte-pair encoding tokenizer trainer called toktoktok — is now open source.

The choice of target is the methodological part. The team needed a BPE tokenizer trainer able to process trillions of tokens on a single machine for its own research on vocabulary size in edge models, and found the existing landscape thin: sentencepiece was optimised for non-BPE tokenizers and slow. So the task was not a benchmark exercise but a gap they would otherwise have had to fill themselves, which is the only setting in which autonomy can be judged honestly. The write-up covers how to specify a goal for multi-domain experts and how to set up the verification infrastructure around the loop.

Verification infrastructure being the deliverable rather than the prompt is the pattern in [agent work moving into harnesses, factories and accepted artifacts](https://newruntime.com/posts/agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts/) and the constraint named in [verification bandwidth being the scarce resource](https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource/). The [agent harness hub](https://newruntime.com/topics/agent-harness/) collects the loop designs.

This is a single retrospective experiment on one problem with models from late 2025, and the account is the operator's own without independent replication. The condition to watch is whether the loop is re-run with current models against a second real gap, because a technique demonstrated once on a favourable target is a case study rather than a method.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Retrieval](https://newruntime.com/topics/agent-retrieval.md): Explore the agent-retrieval topic hub.
- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Ai](https://newruntime.com/topics/ai.md): Explore the ai topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)