Builds

The Agent Stack #043 — Monday Build OpenAI’s new Lockdown Mode landed Saturday with zero fanfare and maximum impact. This isn’t another GPT wrapper announcement. It’s the first practical defence against prompt injection attacks that actually works in production. What Lockdown Mode Actually Does Prompt injection is the SQL injection of AI. Users slip malicious instructions into legitimate prompts to extract training data, bypass safety filters, or exfiltrate sensitive context. Standard defences like input sanitisation fail because LLMs are designed to follow instructions embedded anywhere in text. ...

The Agent Stack #040 — Monday Build NVIDIA just released Cosmos 3, and it’s the first open model that can reason about physics and plan actions in the real world. This isn’t another chatbot that hallucinates physics. It’s a foundation model trained on millions of videos that understands how objects move, collide, and interact. Why This Changes Everything for Builders Most AI agents live in pure text or image land. They can write code and analyse data, but ask them to help a robot stack boxes or navigate a room? Disaster. Cosmos 3 bridges that gap by understanding physical cause and effect. ...

The Agent Stack #037 — Monday Build Nvidia just dropped something that changes how we think about LLM inference. Their Nemotron-Labs diffusion language models generate text at what they’re calling “speed-of-light” performance. This isn’t marketing fluff. Traditional autoregressive models generate one token at a time. Diffusion models generate entire sequences in parallel. Think going from dial-up to fibre, but for text generation. Building with Diffusion Language Models The core insight is architectural. Instead of predicting the next token given all previous tokens, diffusion models start with noise and iteratively refine it into coherent text. This parallelisation is why they’re fast. ...

The Agent Stack #034 — Monday Build You don’t need a framework to build an AI agent. You need 300 lines of Python and a clear head about what ReAct actually does. This week’s most practical piece came from quantumentangled.dev - a walkthrough of ReAct’s core loop without the enterprise bloat. I’ve been saying this for months: most agent frameworks are solving problems you don’t have yet. The 300-Line Reality Check ReAct (Reasoning + Acting) boils down to three steps: ...

The Agent Stack #031 — Monday Build MachinaCheck just dropped something interesting. A multi-agent system for CNC manufacturability analysis running on AMD’s MI300X accelerators. This isn’t another ChatGPT wrapper—it’s proper industrial AI that could change how we validate designs before they hit the factory floor. The Architecture That Actually Works The MachinaCheck team built three specialised agents working in concert. First agent analyses CAD geometry for basic manufacturability. Second agent simulates toolpath generation and identifies potential issues. Third agent estimates costs and lead times based on real manufacturing constraints. ...

The Agent Stack #028 — Monday Build The Pentagon just gave us the biggest signal yet about where AI agents are heading. Last Friday, they signed deals with OpenAI, Google, Microsoft, AWS, Nvidia, and xAI to deploy AI on classified networks. Anthropic got left out over usage disputes. This isn’t just procurement news - it’s your blueprint for building enterprise-grade agents that handle sensitive data. Why This Matters for Builders Government contracts always preview enterprise requirements 18 months early. The Pentagon’s criteria tell us exactly what Fortune 500 CISOs will demand: ...

The Agent Stack #025 — Monday Build Distributed tracing just got its AI moment. Jaeger v2 dropped with OpenTelemetry at its core, specifically targeting the chaos of multi-agent workflows. The timing isn’t coincidental. Traditional observability breaks down when you have agents calling other agents calling APIs calling more agents. You end up with a mess of logs that tell you nothing about why your agent decided to delete the production database (yes, that actually happened this week). ...

The Agent Stack #022 — Monday Build NVIDIA just dropped Nemotron OCR v2, and it’s not the model that matters—it’s how they built it. They generated millions of synthetic text images to train a multilingual OCR system that beats commercial APIs. Here’s how to steal their playbook. The Synthetic Data Factory Traditional OCR training requires massive datasets of real documents. Expensive, slow, and you’re stuck with whatever languages and fonts exist in your training set. NVIDIA flipped this: generate infinite training data instead. ...

The Agent Stack #019 — Monday Build The OpenClaw drama last week wasn’t just about Anthropic flexing their pricing muscles. It highlighted the biggest pain point for AI agent builders: your brilliant agent becomes useless when it hits rate limits or gets temporarily banned. Here’s how to build agents that keep working when APIs fail. The Problem with Single-Provider Agents Most developers build agents that depend entirely on one LLM provider. When Claude goes down or your API key gets throttled, everything stops. OpenClaw’s creator learned this the hard way when Anthropic temporarily cut off access. ...

The Agent Stack #016 — Monday Build Most agent memory solutions want you to spin up vector databases and embedding services. MemWeave just dropped on GitHub with a different approach: Markdown files and SQLite. That’s it. The sachinsharma9780/memweave repo shows how to build persistent agent memory without any external dependencies. No Pinecone subscriptions. No ChromaDB containers. Just files your agent can read and search locally. Here’s the core architecture. Your agent writes memories as structured Markdown: ...