Category
AI without the hype: models, agents and use cases that work in production.
Goose is an open-source AI agent created at Block that runs on your own machine as a desktop app, CLI and API. It reads and writes files, runs commands and tests, and works with more than fifteen model providers and MCP extensions. It is free, licensed under Apache 2.0, and you pay only for model usage.
OpenHands, formerly OpenDevin, is an open-source platform that solves programming tasks end to end: it takes a request, opens a sandbox container and edits files, runs commands and browses the web until it is done. It runs with Docker on your own machine and works with the model you choose.
The Codex CLI is OpenAI's coding agent that works inside your terminal: you describe a task, it reads your repository, proposes the changes and runs commands inside a sandbox you control. It is open source, installs with npm and works with your ChatGPT account or with an API key.
Roo Code is an open-source VS Code extension, born as a fork of Cline, that turns the editor into a team of agents with specialised modes (Code, Architect, Debug, Ask and Orchestrator). The project was archived in May 2026 at version 3.54.0, but its community continuation ZooCode keeps the same features alive.
Cline is a VS Code extension that turns your editor into an autonomous coding agent: it reads your project, plans changes in Plan mode and carries them out in Act mode, showing every edit as a diff you approve. It is open source and works with your own API key or with local models.
A multi-agent system splits a task across several specialised agents coordinated by a design pattern. The three most common are orchestrator-workers, where a lead agent delegates to parallel subagents; hierarchical, with teams of teams; and network, where any agent hands control to another through a handoff.
ReAct and plan-and-execute are the two control patterns for an AI agent. ReAct decides one step at a time, reasoning and acting in a loop; plan-and-execute draws up a full plan first and then executes it step by step. The former adapts better to surprises; the latter uses fewer calls and plans long tasks with more order.
Context engineering is the craft of deciding what information enters a model's window at each step of an agent. Beyond prompt engineering, it manages the whole set of tokens: instructions, tools, memory and history. Its goal is the smallest possible set of high-signal tokens that still completes the task.
Human-in-the-loop is the pattern that keeps a person inside an AI agent's decision loop: the agent stops at an approval point before an irreversible action, waits for your confirmation and resumes with its state intact. Frameworks such as LangGraph and OpenAI's Agents SDK implement it with interruptions and tool approval.
Memory is what lets an AI agent remember beyond a single conversation. Its working memory is the context window, ephemeral and limited; its long-term memory stores facts, experiences and procedures in an external store, almost always a vector database, and retrieves them when they are needed to keep acting coherently.
The reflection pattern makes an agent critique its own output and rewrite it before accepting it. One model generates, a second step evaluates and flags mistakes, and a third revises, in a loop of one or two rounds. It improves quality on tasks with clear criteria, but each cycle adds model calls, tokens and latency.
Planning lets an AI agent solve long tasks: instead of improvising step by step, it first breaks the goal into an ordered list of subtasks and then runs them. The planner-executor pattern separates thinking from acting, cuts the number of model calls and lets the agent replan when a step fails midway through the job.
The ReAct pattern (Reason + Act) organizes an agent as a repeating three-step loop: reason about what to do, take an action with a tool, and observe the result. Introduced by Yao and colleagues in 2022, it interleaves reasoning and acting so the model can plan, consult external sources, and fix its own mistakes as it goes.
An AI agent is a program that uses a language model as its brain to decide for itself which steps to take toward a goal: it reasons, calls external tools, observes the result and repeats that loop until it is done. Unlike a chatbot, it does not just answer; it acts.
Docker Agent is a Docker CLI plugin for building and running AI agents from a declarative YAML file, no code required: you define one or more agents, give them MCP tools, run them with docker agent run, and share them through any OCI registry, on OpenAI, Anthropic, Gemini or other providers.
Open GSD (Git. Ship. Done.) is an open-source, MIT-licensed toolkit for steering coding agents without losing context: it splits work into five phases (discuss, plan, execute, verify and ship) and delegates the heavy lifting to subagents that each start with a clean context. Its core is the gsd-core engine and the gsd-pi terminal agent.
A vector embedding is a list of real numbers that represents the semantic meaning of a piece of text, an image, or any other data. Two sentences with the same meaning produce vectors that are close together; two unrelated ones produce vectors that are far apart. Semantic search, RAG, and recommendation systems are all built on this principle.
Embeddings, HNSW indexing, reranking, evaluation, context window, latency under load. Full stack with code and measurable SLOs.
On 2 August 2026 high-risk, transparency and Commission enforcement powers kick in. A per-system checklist with downloadable template.
The OTel GenAI spec stabilizes attributes for LLMs, tools, and agents. Practical Python implementation with Anthropic + Grafana Tempo.
Recipe for oMLX on a Mac M5 Max with 128 GB: install, TurboQuant at 3.5-bit, the Qwen 3.6 35B-A3B model stack, Claude Code wiring and our own benchmarks. Revised for oMLX 0.6.4, with what changed since 0.3.8.
Three frameworks, three mental models. When to use each and why — with a real orchestration case.
This guide shows how to build a production-ready agent with the Anthropic SDK in Python: the tool-use loop with the Messages API, streaming with backpressure via a bounded queue, prompt caching with cache_control, your own MCP server registered with the Claude Agent SDK, OTel GenAI traces, and a non-root Docker container ready for production.
Three production coding agents, five real tasks measured by time, tokens, and PR quality. No marketing — the actual measurement.
After eighteen months of multi-vendor adoption, MCP is the de facto standard for connecting models to tools. The complete guide: architecture, servers, policies, authentication, composition, and the antipatterns we’ve already seen in production.
The idea of UI generated on the fly instead of pre-built reached production in 2025. After a year of real-world use, the balance is more nuanced than the initial enthusiasm suggested.
Direct Preference Optimization (DPO) and its variants, IPO, KTO, and SimPO, have displaced RLHF as the preferred alignment method for language models: they drop the separate reward model, cut training cost, and are easier to reproduce. RLHF still has an edge only for frontier models with very large budgets.
Skills package reusable capabilities; subagents isolate bounded-task execution. Together they form the most effective pattern for composing complex agents in 2026.
Synthetic data has moved from a precarious substitute for real data to a central component of modern model training: the most reliable pattern expands a real core of 500 examples with thousands of synthetic paraphrases, provided you validate diversity, correctness, and distribution, and keep at least 30% real data to avoid model collapse.
The Model Context Protocol, proposed by Anthropic in late 2024 and adopted through 2025-2026 by Anthropic, OpenAI, Google, and the open-source community, already has proven operational patterns: separating generic servers from custom ones, explicit per-tool policies, credentials kept outside the model, prefixed composition, and contract tests. This is the state of the art in 2026.
Using an LLM to judge another LLM became widespread in 2024 and remains, in 2026, the only scalable way to evaluate qualitative quality in LLM systems. It is reliable when judge-human correlation exceeds 0.7 on 30 cases and gets recalibrated quarterly; below that threshold, do not trust the number.
Claude Sonnet 4.6 is the default model for most 2026 production workloads: it covers 80% of traffic with quality indistinguishable from Opus 4.7 in blind tests, at roughly 60% of Opus per-token price. Opus is still needed for complex reasoning and agentic coding on large codebases.
Hybrid RAG in 2026 combines dense and lexical search fused with RRF, cross-encoder reranking over the top-50 candidates, structure-aware chunking, and continuous evaluation with Ragas or TruLens. It is the pattern that survives in serious production systems three years after the initial embeddings boom.
Opus 4.7 launched as Anthropic's most capable model, with emphasis on long-horizon agentic work. After two months of intensive use, these are the practical changes versus Opus 4.6.
The first invoice for a production agent usually runs double or triple the estimate. This article walks through five real levers, in priority order, caching, routing, context control, batching, and telemetry, to cut cost without touching perceived quality.
While OpenAI and Anthropic dominate headlines with rounds worth hundreds of millions, a growing group of niche AI startups generates one to ten million dollars in revenue with teams of two to ten people. They share five patterns: narrow vertical focus, 70-80% margins, community distribution, iteration cycles in days, and AI as an internal lever.
AI agents fail in production, and what matters is how you respond in the first twenty minutes. This runbook covers severity classification, isolating before investigating, purging contaminated memory, communicating without inventing facts, and turning every incident into a regression test before closing it as done.
LLM red teaming has gone from an esoteric activity to a mandatory practice. With the OWASP Agentic Top 10 and the CSA Agentic AI Red Teaming Guide converging on shared vocabulary, this is the operational playbook any team deploying agents needs to have.
Reliable agents come from measurement, not from better models or prompts. A production agent evaluation setup starts with a golden dataset of 30 to 200 cases, split roughly 60 percent normal usage, 30 percent edge cases and 10 percent adversarial, and it never uses the same model as both worker and judge.
Prompt engineering has moved from viral tricks to a discipline with reproducible patterns: few-shot, chain-of-thought, and structured output with function calling. Teams treating prompts like code (versioned, tested, and monitored) get consistently better results than those who improvise.
An Agent OS is a runtime layer built to run AI agents rather than ordinary applications, and after six months of production deployments the trade-off is clear. A dedicated agent stack starts slower but stays stable; Kubernetes with orchestration bolted on top moves faster early, then hits observability and policy limits. It pays off from five active agents.
Ollama 0.5 or newer runs Llama 3.3 70B and Mistral Large 2 locally on Ubuntu 24.04: Q4_K_M quantization lets a single NVIDIA GPU with 24 GB of VRAM, an RTX 4090 for example, handle the full model. This guide installs the drivers, sets up Open WebUI, and exposes the service behind Traefik with TLS.
A year after GraphRAG left the lab, one statistic holds: it works where corporate information has dense relational structure, fails where there are only loose documents. Patterns, ingestion costs, and architectural decisions that have survived a year of real deployment.
The Model Context Protocol has gone from proposal to de facto standard for connecting editors with external tools. This practical guide walks through standing up a local MCP server, wiring it into VS Code or your client of choice, and understanding exactly what you are exposing.
After two years of pilots and a year of agents in production, governance has moved from an aspirational committee to an operational control. What audits ask for, what broke in 2025, and which guardrails absorb most incidents.
By late 2025, 57.3 percent of organizations had agents in production, up from 51 percent a year earlier, according to LangChain's survey of more than 1,300 professionals. Three failure modes dominate the postmortems: degenerative reasoning loops, hallucinated data in RAG systems, and silent misalignment between the request and the interpretation.
Twenty months after the initial announcement, Model Context Protocol went from curiosity to de-facto standard among agent clients and servers. What is available, which servers are worth it, which problems remain open, and how it compares to earlier protocol maps.
The European AI Act took effect on 1 August 2024 with a staggered calendar, and its Annex III high-risk rules no longer land in August 2026. The Digital Omnibus, closed by the Parliament and the Council, moves that deadline 17 months to 2 December 2027. Prohibitions since February 2025 and general-purpose AI duties since August 2025 still apply.
