Evaluation
When Agents Fail, Debug the Trajectory
Agent reliability improves when teams inspect the execution trace, locate the first critical failure, and fix the system around it.
Forward deployed engineering
Field Notes
Operating knowledge for building · operationalising · scaling agentic AI systems.
Forward-deployed notes from work that has to hold under real load: context and harness engineering, retrieval, evaluation, reliability, and the deployment discipline that takes agents to millions of people. Practical by default, fewer announcements, more reusable patterns.
dx/dt = −∇f
Agentic AI
Loop Engineering, Graph Engineering, and the Hill-Climb
The industry conversation is shifting from prompting agents to designing the loops and graphs that evaluate, retry, and improve them, including the outer hill-climb from production traces.
Agentic AI
The Agent Harness Is the Product
Why the next frontier in production AI agents is the runtime around the model: tools, sandboxes, memory, traces, and control.
Agentic AI
Memory Is Becoming First-Class Agent Infrastructure
Agent memory is moving beyond chat history into structured state, experiential traces, working context, filesystems, and trustworthy retrieval.
Security
Prompt Injection Is Becoming Social Engineering
As agents browse, retrieve, and act, prompt injection increasingly looks like social engineering against a bounded digital worker.
AI Scaling
System 1 vs System 2 in Modern AI: From AlphaGo to Large Language Models
How fast pattern matching and slower deliberative reasoning show up in modern AI systems, from AlphaGo to reasoning-oriented language models.
AI Scaling
Rethinking AI Efficiency: The Shift from Scaling Pretraining to Test-Time Compute
Why more capable systems increasingly depend on inference-time search, routing, and verification rather than pretraining scale alone.
AI Architecture
Marco-o1 and Large Reasoning Models for Open-Ended Tasks
A look at reasoning strategies that improve model accuracy when tasks require more than immediate pattern completion.
Information Retrieval
Agentic RAG: Retrieval That Plans and Adapts
How retrieval systems change when agents can plan, inspect, execute, and adapt around enterprise knowledge.
AI Safety
Understanding and Tackling Hallucinations in LLMs
The causes, impact, and mitigation patterns behind AI-generated fabrications in production language systems.
AI Scaling
Scaling Inference Compute for Long Context RAG Systems
Optimization techniques, hardware considerations, and architecture patterns for retrieval systems with long context windows.
Deep Learning
Softmax Dispersion: A Challenge for Robust Reasoning
Why softmax probabilities can be fragile for detecting out-of-distribution samples and unexpected reasoning failures.
Multimodal AI
Multimodal Language Models Across Text and Images
How multimodal language models combine text, images, and other signals into richer system behavior.
AI Security
Jailbreaking LLMs with Deceptive Delight: Exploring AI Vulnerabilities
A practical look at prompt-based model vulnerabilities and what they imply for evaluation and governance.
AI Architecture
Mixture of Experts vs Dense LLMs: Routing and Scale
The trade-offs between sparse expert routing and dense model architectures in large-scale language systems.
AI Evaluation
Leveraging LLMs for Scalable AI System Evaluation
How language models can be used to evaluate complex systems at scale while still preserving review discipline.
AI Accessibility
Democratizing Finetuning of LLMs on Your Data with PEFT
Methods that reduce the cost and infrastructure requirements of adapting models to specific domains.
AI Optimization
The Power of Quantization in AI: Efficiency Meets Performance
How lower-precision representations reduce model size and improve inference speed without giving up too much quality.
AI Safety
Detecting LLM Hallucinations with Semantic Entropy
A method for identifying unstable model answers by measuring meaning-level variation across generations.
AI Hardware
The Power of Parallel Processing with GPUs in AI
Why parallel computation matters for modern model training, inference, and real-time AI workloads.
Generative AI
What is Generative AI?
The foundations of generative AI and the applications reshaping software, media, and knowledge work.
Information Retrieval
Retrieval Augmented Generation for Grounded Answers
How retrieval-augmented generation improves model grounding and decision support across knowledge-heavy workflows.
AI Scaling
Pretraining AI Models Across Thousands of GPUs
The distributed systems concerns behind large-scale model training across vast GPU clusters.
Generative AI
Compound AI Systems: More Than Just a Big Brain
Why real-world AI systems combine models with tools, retrieval, execution environments, and feedback loops.