Blog
Notes on agents doing the repeatable work, and on the decisions that stay with you.
Agent memory grew a write path: why memory became a subsystem
An agent that can't remember is a chatbot. An agent that remembers everything it's ever been told is a liability. Getting from one to the other means designing memory as a real su…
Agentic Context Engineering: Playbooks That Do Not Rot
Most teams tune prompts. Few teams engineer context — and the difference is quietly becoming the whole game for agent reliability. Here is an in-depth look at agentic context engi…
KV Cache Reality: Memory Is the Silent Throughput Killer
A model's intelligence is not the bottleneck in local inference. Its memory usually is. Here is a plain-language deep dive into the KV cache — the structure that quietly decides h…
8 Oct 2026
Beyond RLHF: What Preference Optimization Actually Means for Local LLM Ops
If you read model blog posts, Reinforcement Learning from Human Feedback — RLHF — is still presented as the gold standard for alignment. It was never that, and since mid-2024 most…
7 Oct 2026
Tool Use and Agent Reliability: Why Loops Fail Quietly
LLM agents that call tools don't fail because the model can't read the API spec. They fail because the loop has no stop condition that works, no evidence trail a human can verify…
Retrieval vs Long Context — When RAG Still Wins
Long context windows get sold like infinite memory at the edge of every model release. If you can paste an entire manual into a prompt at inference time, why bother with retrieval?
Speculative Decoding Explained: Draft Models That Speed Up Inference
Every time you ask an LLM to write something, it produces one token at a time. The model reads what it has already written, runs its full transformer forward pass, and picks the n…
Test-Time Compute: When Thinking Longer Beats Bigger Weights
Every time a new frontier model drops, the headline is the same: bigger weights mean smarter reasoning. But recent results from several teams have quietly challenged that assumpti…
Mixture-of-Experts Routing — Why Sparse Models Scale
A dense LLM runs every neuron on every word. A Mixture-of-Experts (MoE) model has the same total parameters — but only activates a handful at each step. Think of it as a team wher…
3 Oct 2026
Test-Time Compute — When Thinking Longer Beats Bigger Weights
Inference-time scaling (test-time compute) shifted reasoning from param-count to search-and-verify loops at the end of 2025. DeepSeek-R1 showed that letting a model think longer b…
4 Sep 2026
How we check agent work, without chasing benchmarks
A chat demo has one promotion rule: someone liked the screenshot. An agentic-first company needs a rule you can run twice and get the same answer.
4 Sep 2026
RAG as company memory: how agents stop making up who you are
Large models are fluent. Fluency is not memory. If an agent writes a hex code, a legal notice, or a founder bio from “what sounds right,” you do not have a brand. You have a rumor…
10 Aug 2026
Agentic-first companies: agents do the repeatable work, people keep the company
“Agentic-first” is easy to misread as “fire the humans.” That is not what we mean. In glossary contexts, you may see agent-first used interchangeably; both refer to the same opera…
10 Aug 2026
What is LLM Ops for an agentic-first company?
LLM Ops is the discipline of running large-language-model systems in production: deploying, observing, evaluating, routing, retrieving, and gating changes so they stay safe and us…
10 Aug 2026
LLM routing, where models run, and RAG as company memory
A team of agents that calls the same model for every task is like a company that sends the same person to every meeting. You can do it. You shouldn’t.
10 Aug 2026
Running agents on your own machines, with models fit to your work
If your agents only exist inside a single vendor’s chat UI, you don’t have an operating model. You have a subscription.
10 Aug 2026
ML evaluation loops for agent ops: past the chat demo
Large language models made agent demos easy. They did not make operations easy.
1 Aug 2026
Brand kits are the source of truth, not vibes
If three agents and two humans each “remember” a slightly different logo blue, you do not have a brand. You have a rumor mill with good intentions.
1 Aug 2026
Dry-run by default: why preview beats hope
Most content systems treat “publish” as the happy path and “preview” as an optional checkbox. We inverted that.
1 Aug 2026
Human-in-the-loop publishing, without the theater
“Human in the loop” has become a label people paste on systems that still auto-send with a delay. We mean something stricter: no outbound side-effect without an explicit human dis…
18 Jul 2026
Agents can't govern AI. People do.
We've been building the tools for MeltingFace Presence Bot, and one question keeps coming up in our internal reviews: if agents are writing code, managing approvals, and drafting…
10 Aug 2026
Why AI governance matters, from a developer
If you are an engineer who has deployed models and tools to production, you already know that putting software in front of real users changes everything. Models drift. Inputs beco…