Mure, L. (2026)
Hierarchical Adaptive Recursive Language Models
A design study of parallel scheduling, adaptive routing, hierarchical memory, and information-weighted token budgeting for recursive language model systems.
Research question
How should software behave when it can act on its own?
I study the architectures, control loops, and engineering practices that turn capable agents into dependable systems for real organizations.
Research and professional interests
- Dad
- Generalist
- Cloud · Data · AI Platform Leadership
- Dependable Multi-Agent Systems: Architecture, Evaluation, and Control
Selected research
Current inquiries and systems
Work at the intersection of research argument, open-source implementation, and production engineering.
Autonomi, also known as Loki Mode
A provider-agnostic framework for autonomous software delivery, built around 41 specialized agent types, 8 swarms, and the Reason, Act, Reflect, Verify control cycle.
Examine the system
Engineering method
A disciplined loop for consequential work
The method is simple by design. Each phase makes the next decision more legible and the final claim more defensible.
- 01
Frame
Define the problem, constraints, evidence, and a falsifiable standard for success.
- 02
Build
Move from argument to implementation through the smallest useful experimental loop.
- 03
Examine
Inspect outputs and assumptions with the same care given to the original idea.
- 04
Verify
Make claims observable through tests, artifacts, and reproducible behavior.
Recent writing
Essays and field notes
Notes on autonomous agents, platform engineering, technical leadership, and the institutional consequences of software.
12 min read
gstack vs Loki Mode: 105,000 Stars Does Not Settle the Argument
One project has 105,637 stars. The other has 956. A 110x star gap does not reflect a 110x quality gap. A head-to-head between gstack and Loki Mode, with tests run in both repos.
8 min read
AI Agents Are Already Out of Control and Nobody is Ready
An AI agent hijacked GPUs to mine crypto without being told to. Vibe-coded apps ship with 2.7x more security holes. We are building autonomous systems faster than we can govern them.
7 min read
Introducing Loki's Log: A Newsletter for AI Builders
Launching a newsletter focused on AI agents, open source tooling, and the craft of building autonomous systems
8 min read
The asklokesh Ecosystem: How All My Projects Connect
A walkthrough of every public project I maintain and how they form a connected system for AI-powered software engineering
8 min read
Open Source AI in 2026: Models, Tools, and Protocols
The open source AI ecosystem is no longer chasing proprietary models; it is building the infrastructure layer that proprietary systems depend on