Blog

Every mngr plugin, explained

Fine-grained HTTP permissions for AI agents

How AI code review can make correct code worse

Blueprint: One-shot bigger coding tasks

There Will Be a Scientific Theory of Deep Learning

Keystone-Eval: Our agentic benchmark for self-configuring agents

How to usefully run 1,000 agents in parallel

Bouncer: Heal your feed

How Offload Works: Inside the Rust CLI that sped up our tests by 6x

A case study in testing with 100+ of Claude agents in parallel

Usefully run hundreds of Claudes in parallel with mngr

Introducing Keystone: Building self-configuring agents

Offload: We made agents run our test suite 6x faster

Vet: Prevent coding agents from making mistakes

Latchkey: Use plain old REST instead of MCP

LLM-based Evolution as a Universal Optimizer

Beating ARC-AGI-2 with Code Evolution

A more radical Imbue

Choices and Knives

Digital Freedom Depends on Access Rights: Matt Boulos in Lawfare

How we made sandboxed coding agents 10x faster to start

A Healthy Ecosystem for AI Agents: Introducing the AFI

Closed vs. open AI ecosystems: Kanjun on Decoder

Sculptor: the missing UI for parallel coding agents

How to build robust software systems with AI: Josh at the AI Engineer World's Fair 2025

Empowering humans in the age of AI

Sculptor: Catch and fix issues as you code

One-person enterprise: Kanjun at the World Economic Forum Annual Meeting

The future of AI is human-centered: Kanjun on the Pioneers of AI podcast

Glenn McDonald: On Dungeons & Dragons, data alchemy, and radically open software

Keeping humans in the loop: Josh at the Democratize Intelligence Summit

Human agency vs. “agentic” AI: Kanjun on the NonZero podcast

A more empowering view of general agents: Kanjun on NYSE TV

Transforming AI agents into personal collaborators: Kanjun on the NVIDIA AI Podcast

The ethical and practical future of AI: Matt Boulos at the MIT Club of Northern California

The AI paradox: Kanjun at the 2024 Progress Conference

Rylan Schaeffer, Stanford: On investigating emergent abilities and challenging dominant research ideas

Agency in the age of AI: Kanjun on the Minus One podcast

Reinventing AI agents: Kanjun on the Gradient Dissent podcast

Ari Morcos, DatologyAI: On leveraging data to democratize model training

From bare metal to a 70B model: infrastructure set-up and scripts

Training >70B LLMs on 10,000 H100 clusters: Josh on the Latent Space podcast

Open-sourcing CARBS: how we used our hyperparameter optimizer to scale up to a 70B-parameter language model

Training a 70B model from scratch: open-source tools, evaluation datasets, and learnings

Sanitized open-source datasets for natural language and code understanding: how we evaluated our 70B model

Percy Liang, Stanford: On the paradigm shift and societal effects of foundation models

'We need legal protections from AI’s risks': Matt Boulos in The Hill

Seth Lazar, Australian National University: On legitimate power, moral nuance, and the political philosophy of AI

Tri Dao, Stanford: On FlashAttention and sparsity, quantization, and efficient inference

Developing agents that can reason and code: Kanjun and Josh on the No Priors podcast

The need for trust: Analyzing the individual comments to the NTIA’s AI RFC

Imbue Raises Fresh Capital, Amazon's Alexa Fund Joins Investor Lineup

Why AI agents don’t work (yet): Kanjun on the Latent Space podcast

Imbue raises $200M to build AI systems that can reason and code

Our technical hiring process

Bryden Fogelman: On coding agents and finding meaningful work at Imbue

In-person culture and habits

Where are the good policy ideas? Analyzing the NTIA's AI RFC

What's next in AI?: Kanjun and Reid Hoffman at Config 2023

On the Stepwise Nature of Self-Supervised Learning

Jamie Simon, UC Berkeley: On theoretical principles for how neural networks learn and generalize

Scaling Laws For Every Hyperparameter Via Cost-Aware HPO

Our submission to the NTIA's AI accountability RFC

Towards guardrails, not guidelines: a policy framework for powerful AI systems

Update: The state of LLM ethical decision-making

Bill Thompson, UC Berkeley: On how cultural evolution shapes knowledge acquisition

Ben Eysenbach, CMU: On designing simpler and more principled RL algorithms

Jim Fan, NVIDIA: On foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant

Sergey Levine, UC Berkeley: On the bottlenecks to generalization, why simulation is doomed to succeed, and picking good research problems

The state of LLM ethical decision-making

Noam Brown, FAIR: On achieving human-level performance in poker and Diplomacy, and the power of spending compute at inference time

Sugandha Sharma, MIT: On biologically inspired neural architectures, how memories can be implemented, and control theory

Nicklas Hansen, UCSD: On long-horizon planning and why algorithms don't drive research progress

Jack Parker-Holder, DeepMind: On open-endedness, evolving agents and environments, online adaptation, and offline learning

Celeste Kidd, UC Berkeley: On attention and curiosity, how we form beliefs, and where certainty comes from

Understanding self-supervised and contrastive learning with "Bootstrap Your Own Latent" (BYOL)