# Decoding ML

Join for content on designing, building, and shipping AI software. Learn AI engineering, end-to-end, from idea to production. Every Tuesday.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Build Durable, Secure Long-Running Agents

DevFeed: [Build Durable, Secure Long-Running Agents](<https://devfeed.tech/articles/build-durable-secure-long-running-agents-139474.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/build-durable-secure-long-running-agents>)

Author: Paul Iusztin

Published: 2026-11-17T07:02:53Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [long-running](<https://devfeed.tech/topics/long-running.md>), [Loop Engineering](<https://devfeed.tech/topics/loop-engineering.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [22](<https://devfeed.tech/tags/22.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [added](<https://devfeed.tech/tags/added.md>), [after](<https://devfeed.tech/tags/after.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-builder](<https://devfeed.tech/tags/agent-builder.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-graphrag](<https://devfeed.tech/tags/agentic-graphrag.md>), [agents](<https://devfeed.tech/tags/agents.md>), [agents-in-production](<https://devfeed.tech/tags/agents-in-production.md>), [build](<https://devfeed.tech/tags/build.md>), [code](<https://devfeed.tech/tags/code.md>), [crashes](<https://devfeed.tech/tags/crashes.md>), [long-running](<https://devfeed.tech/tags/long-running.md>), [long-running-agents](<https://devfeed.tech/tags/long-running-agents.md>), [run](<https://devfeed.tech/tags/run.md>), [running](<https://devfeed.tech/tags/running.md>)

### AI overview

The article describes CodeClimb, an overnight coding-agent workflow that improves code against dataset-based metrics while isolating coding from evaluation. A TypeScript orchestrator enforces metric, round, time, and spend limits; a metering proxy holds provider credentials, meters calls, and blocks models outside an allowlist. Inngest checkpoints experiment and evaluation steps so runs can resume after tested worker or network interruptions, while explicit cleanup is needed to stop active agents and sandboxes. Human events can pause runs without occupying workers, and per-experiment Git commits preserve the best result and provide a basis for continuation. In the GraphRAG example, the workflow approached its node extraction target but missed edge extraction; test retrieval and graph-merge metrics exceeded 0.9. The author reports about $75 for the winning night, notes dataset quality as a major bottleneck, and says the resulting code was not suitable for production scaling. Cloud deployment with Inngest Sandboxes is described, with a one-hour active runtime limit.

### Source excerpt

From crashes, throwaway code and surprise bills to a $74.63 run you are proud of

## Don't Let Claude Code Design Your LLM Judges

DevFeed: [Don't Let Claude Code Design Your LLM Judges](<https://devfeed.tech/articles/don-t-let-claude-code-design-your-llm-judges-127871.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/write-your-own-llm-judges>)

Author: Louis-François Bouchard

Published: 2026-11-10T12:21:59Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Reliable AI outputs](<https://devfeed.tech/topics/reliable-ai-outputs.md>), [AI Engineering](<https://devfeed.tech/topics/ai-engineering.md>), [phoenix](<https://devfeed.tech/topics/phoenix.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [codex](<https://devfeed.tech/topics/codex.md>)

Tags: [1-4](<https://devfeed.tech/tags/1-4.md>), [2](<https://devfeed.tech/tags/2.md>), [2-2](<https://devfeed.tech/tags/2-2.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [account](<https://devfeed.tech/tags/account.md>), [added](<https://devfeed.tech/tags/added.md>), [adoption](<https://devfeed.tech/tags/adoption.md>), [after](<https://devfeed.tech/tags/after.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-builder](<https://devfeed.tech/tags/agent-builder.md>), [agents](<https://devfeed.tech/tags/agents.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code](<https://devfeed.tech/tags/code.md>), [don-t](<https://devfeed.tech/tags/don-t.md>), [let](<https://devfeed.tech/tags/let.md>), [llm](<https://devfeed.tech/tags/llm.md>), [me](<https://devfeed.tech/tags/me.md>), [noise](<https://devfeed.tech/tags/noise.md>)

### AI overview

ToneBench's 28,200 LLM judgments showed that judge prompts and scores need human calibration before they can support a trustworthy ranking. The author recommends measuring both generation variance and judge reliability: ToneBench now scores five drafts per model configuration and tests judges against human ratings. For subjective scoring, its most consistent approach uses detailed, hand-written rubric levels with examples, evidence quotes before scores, and checks for score clustering and whether rubric revisions separate strong from weak drafts. A panel of judges from different model families can dilute individual models' self-preference, though it does not remove bias; judges also differ in score generosity, speed, and reliability. The author's golden set uses human ratings on the same categories as the judges, but calibration results are not yet available. For ongoing runs, pin exact judge model IDs, re-evaluate after judge or rubric changes, audit inputs and outputs, and require pairwise preferences to agree in both presentation orders. ToneBench is a signal for shortlisting models, not ground truth for other use cases.

### Source excerpt

What 28,200 LLM judgments taught me about noise, rubrics, and human calibration

## Harness Engineering Roadmap: Claude Code From First Principles

DevFeed: [Harness Engineering Roadmap: Claude Code From First Principles](<https://devfeed.tech/articles/harness-engineering-roadmap-claude-code-from-first-principles-113341.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/claude-code-from-scratch-harness-engineering-roadmap>)

Author: Paul Iusztin

Published: 2026-11-03T06:01:13Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [Software Engineering](<https://devfeed.tech/topics/software-engineering.md>), [Reverse Engineering](<https://devfeed.tech/topics/reverse-engineering.md>), [Learning](<https://devfeed.tech/topics/learning.md>), [observability](<https://devfeed.tech/topics/observability.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [3](<https://devfeed.tech/tags/3.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-builder](<https://devfeed.tech/tags/agent-builder.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [agents-in-production](<https://devfeed.tech/tags/agents-in-production.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code](<https://devfeed.tech/tags/code.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [first](<https://devfeed.tech/tags/first.md>), [harness](<https://devfeed.tech/tags/harness.md>), [harness-engineering](<https://devfeed.tech/tags/harness-engineering.md>), [langchain](<https://devfeed.tech/tags/langchain.md>), [principles](<https://devfeed.tech/tags/principles.md>), [roadmap](<https://devfeed.tech/tags/roadmap.md>), [top](<https://devfeed.tech/tags/top.md>)

### AI overview

The roadmap teaches harness engineering by building a Python coding agent, Decode, across eight steps: design its modules and interfaces; implement a ReAct loop and core coding tools; sandbox execution locally with Docker or remotely with Modal; manage context through memory, skills, an LSP tool, and compaction; and add parallel subagents either inside the harness or through CLI orchestration with tmux. It then covers headless cloud runs triggered by CLI, webhooks, or cron, custom benchmarks and regression tests, and replaying recorded failures with the original tool outputs. The course is described as free and open source, with eight articles, six videos, and code; it assumes basic LLM and agent knowledge, uses cloud execution without a GPU, and takes about 4-8 hours. The article cites LangChain's reported Terminal-Bench improvement from 53% to 67% after changing its harness while keeping the model, and presents these steps as broadly useful agent-building foundations.

### Source excerpt

The harness took LangChain's agent from outside the top 30 to the top 5

## Building a Support Chatbot on Amazon Bedrock

DevFeed: [Building a Support Chatbot on Amazon Bedrock](<https://devfeed.tech/articles/building-a-support-chatbot-on-amazon-bedrock-110603.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/agentic-rag-vs-single-search-amazon-bedrock>)

Author: Marcelo Acosta Cavalero

Published: 2026-10-31T12:03:27Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Amazon Bedrock](<https://devfeed.tech/topics/amazon-bedrock.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Chat Bot](<https://devfeed.tech/topics/chatbot.md>), [Amazon Bedrock Knowledge Bases](<https://devfeed.tech/topics/amazon-bedrock-knowledge-bases.md>), [Support](<https://devfeed.tech/topics/support.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Amazon Bedrock AgentCore](<https://devfeed.tech/topics/amazon-bedrock-agentcore.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentcore](<https://devfeed.tech/tags/agentcore.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-retrieval](<https://devfeed.tech/tags/agentic-retrieval.md>), [agents-in-production](<https://devfeed.tech/tags/agents-in-production.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [amazon-bedrock](<https://devfeed.tech/tags/amazon-bedrock.md>), [api](<https://devfeed.tech/tags/api.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [article](<https://devfeed.tech/tags/article.md>), [aws](<https://devfeed.tech/tags/aws.md>), [chatbot](<https://devfeed.tech/tags/chatbot.md>), [cloudwatch](<https://devfeed.tech/tags/cloudwatch.md>), [code](<https://devfeed.tech/tags/code.md>), [search](<https://devfeed.tech/tags/search.md>), [support](<https://devfeed.tech/tags/support.md>)

### AI overview

In this 24-question Amazon Bedrock RAG experiment, a single search returning nine passages produced more passing answers, faster responses, and lower model cost than agentic retrieval. The agent cost 23% more, took 72% longer, and passed two fewer answers; sending all documents matched the nine-passage score at nearly three times the cost. Both approaches found the gift-return rule, but extra retrieval did not ensure correct application: the agent gave conflicting guidance on when a trial period began, and broader versions deducted delivery twice in a refund calculation. The author recommends keeping the single search for this chatbot and fixing answer application when the needed policy is already present. Agents may be worth testing for investigations where each result guides the next search; unanswered questions requiring live order data need a service integration. Before production, the author calls for repeated testing, policy review, identity and order verification, citation checks, search limits, and support handoff.

### Source excerpt

One search or an AI agent that can search again? I tested both.

## Context Inspector: Measuring Context Quality Before the LLM Call

DevFeed: [Context Inspector: Measuring Context Quality Before the LLM Call](<https://devfeed.tech/articles/context-inspector-measuring-context-quality-before-the-llm-call-89215.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/pre-inference-rag-evaluation-metrics>)

Author: Decoding AI Magazine

Published: 2026-10-17T11:01:11Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [ai observability](<https://devfeed.tech/topics/ai-observability.md>)

Tags: [ai-evaluation](<https://devfeed.tech/tags/ai-evaluation.md>), [context](<https://devfeed.tech/tags/context.md>), [llm](<https://devfeed.tech/tags/llm.md>), [measuring](<https://devfeed.tech/tags/measuring.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [quality](<https://devfeed.tech/tags/quality.md>), [rag](<https://devfeed.tech/tags/rag.md>)

### AI overview

The article presents Context Inspector, a research prototype for checking the quality of retrieved context before it is sent to a large language model. It describes six signals--coverage, token utilization, source starvation, freshness, contradiction density, and policy admissibility--and explains how teams can use these metrics as configurable pre-inference checks in development and CI alongside response-level evaluation.

### Source excerpt

6 signals that catch missing, stale, or conflicting evidence in your RAG pipeline

## The Future of AI Coding: Multi-Agent

DevFeed: [The Future of AI Coding: Multi-Agent](<https://devfeed.tech/articles/the-future-of-ai-coding-multi-agent-85849.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/the-future-of-ai-coding-is-multi-agent>)

Author: Johannes Millan

Published: 2026-10-13T11:03:04Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [agent orchestration](<https://devfeed.tech/topics/agent-orchestration.md>), [Loop Engineering](<https://devfeed.tech/topics/loop-engineering.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [codex](<https://devfeed.tech/topics/codex.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>)

Tags: [agent-workflows](<https://devfeed.tech/tags/agent-workflows.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>)

### AI overview

The article outlines a multi-agent workflow for software engineering tasks: models independently investigate a problem, one agent implements a selected plan, and other agents review the change from defined perspectives. The author argues this can reduce programmer attention spent comparing full implementations, while using more agent computation. The workflow is recommended for consequential or unclear changes, with a single agent preferred for small tasks.

### Source excerpt

Using orchestration to get the best results

## The Agentic Software Factory Hierarchy of Needs

DevFeed: [The Agentic Software Factory Hierarchy of Needs](<https://devfeed.tech/articles/the-agentic-software-factory-hierarchy-of-needs-83946.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/the-agentic-software-factory-hierarchy>)

Author: Hugo Bowne-Anderson

Published: 2026-10-09T11:01:42Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Loop Engineering](<https://devfeed.tech/topics/loop-engineering.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Spec Driven Development](<https://devfeed.tech/topics/spec-driven-development.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [issue tracking](<https://devfeed.tech/topics/issue-tracking.md>)

Tags: [agent-verification](<https://devfeed.tech/tags/agent-verification.md>), [agentic-software-factory](<https://devfeed.tech/tags/agentic-software-factory.md>), [agents](<https://devfeed.tech/tags/agents.md>), [agents-md](<https://devfeed.tech/tags/agents-md.md>), [attention](<https://devfeed.tech/tags/attention.md>), [automation](<https://devfeed.tech/tags/automation.md>), [bug](<https://devfeed.tech/tags/bug.md>), [building](<https://devfeed.tech/tags/building.md>), [builds](<https://devfeed.tech/tags/builds.md>), [change](<https://devfeed.tech/tags/change.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>)

### AI overview

The article presents a four-layer model for building agentic software factories: an agent harness, project personalisation, automation and continual learning. It explains how autonomous agents can build and operate software between human interactions, with specification, verification, permissions and human decisions guiding the work.

### Source excerpt

How to build software with agents that keep working while you're away

## Evaluating a RAG App: Alexey Grigorev's Five-Step Workflow

DevFeed: [Evaluating a RAG App: Alexey Grigorev's Five-Step Workflow](<https://devfeed.tech/articles/stop-guessing-evaluate-your-rag-app-65380.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/rag-evals-change-prompts>)

Author: Paul Iusztin

Published: 2026-10-06T05:00:52Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [Ground truth / benchmark quality](<https://devfeed.tech/topics/ground-truth-benchmark-quality.md>), [Pull Request](<https://devfeed.tech/topics/pull-request.md>)

Tags: [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [ai-workflows](<https://devfeed.tech/tags/ai-workflows.md>), [case-study](<https://devfeed.tech/tags/case-study.md>), [coding](<https://devfeed.tech/tags/coding.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>)

### AI overview

An interview-based tutorial walks through how Alexey Grigorev evaluates a Slack FAQ RAG app, including assembling evaluation cases from GitHub issues and pull request corrections, measuring retrieval and generation, and routing user corrections into future tests. The workflow uses versioned datasets and experiments to assess prompt changes and catch regressions.

### Source excerpt

From vibe checks to a 61-case score you ship on, in 5 steps

## Turning Agent Failures Into Regression Tests

DevFeed: [Turning Agent Failures Into Regression Tests](<https://devfeed.tech/articles/will-your-next-change-break-your-agent-61616.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/transform-agent-traces-into-regression-cases>)

Author: Paul Iusztin

Published: 2026-09-29T05:02:19Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [High risk of breaking in future versions](<https://devfeed.tech/topics/high-risk-of-breaking-in-future-versions.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>)

Tags: [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [regression](<https://devfeed.tech/tags/regression.md>)

### AI overview

This lesson explains how to turn AI agent failures into regression tests. It describes an agent that used a Gemini API key after a serverless endpoint returned a 503, causing unexpected token costs, and recommends replaying failures and rerunning regression tests after changes to prompts, tools, or models.

### Source excerpt

Error analysis and replays that reproduce agent bugs in their original context

## How to Build Custom Evals and Regression Tests for AI Agents

DevFeed: [How to Build Custom Evals and Regression Tests for AI Agents](<https://devfeed.tech/articles/i-built-a-benchmark-for-my-agent-the-smaller-model-won-57823.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/evaluate-ai-agents-benchmarks-regression-tests>)

Author: Paul Iusztin

Published: 2026-09-22T05:01:45Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Code](<https://devfeed.tech/topics/code.md>), [gpt-oss](<https://devfeed.tech/topics/gpt-oss.md>), [Python](<https://devfeed.tech/topics/python.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [LangChain](<https://devfeed.tech/topics/langchain.md>)

Tags: [agent-evals](<https://devfeed.tech/tags/agent-evals.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context-engineering](<https://devfeed.tech/tags/context-engineering.md>), [cost](<https://devfeed.tech/tags/cost.md>), [latency](<https://devfeed.tech/tags/latency.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [tests](<https://devfeed.tech/tags/tests.md>)

### AI overview

A practical guide to building custom evaluation harnesses and regression tests for AI agents. It argues that application-specific evals can reveal meaningful differences between models, including cases where a smaller model outperforms a larger one, while measuring performance, cost, and latency.

### Source excerpt

A 35B beat a 120B by 42 points: the benchmark and regression tests behind it

## Stop Babysitting Your Coding Agents

DevFeed: [Stop Babysitting Your Coding Agents](<https://devfeed.tech/articles/stop-babysitting-your-coding-agents-18294.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/coding-agents-in-remote-headless>)

Author: Paul Iusztin

Published: 2026-09-10T05:02:03Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [long-running](<https://devfeed.tech/topics/long-running.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [LangChain](<https://devfeed.tech/topics/langchain.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-application](<https://devfeed.tech/tags/ai-application.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [ai-products](<https://devfeed.tech/tags/ai-products.md>), [background](<https://devfeed.tech/tags/background.md>), [ci-cd](<https://devfeed.tech/tags/ci-cd.md>), [ci-cd-pipeline](<https://devfeed.tech/tags/ci-cd-pipeline.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [cli](<https://devfeed.tech/tags/cli.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context](<https://devfeed.tech/tags/context.md>), [harness](<https://devfeed.tech/tags/harness.md>), [modal](<https://devfeed.tech/tags/modal.md>)

### AI overview

The lesson turns Decode, a coding-agent harness built earlier in the course, into a cloud-based background agent in three steps: replace its TUI with a headless CLI, deploy it to Modal as remote serverless jobs, and add parallel sessions for multiple projects. The motivation is to reduce the work of supervising agents that each require attention in a terminal. The excerpt describes proactive background agents as an ideal the author is still pursuing, and presents this build as a practical attempt to understand their workflow integration without simply spending more. It does not include the promised run command or details of the resulting agent behavior.

### Source excerpt

Run the harness remotely on Modal, triggered by your CLI, a webhook, or a nightly cron.

## From 1 Bloated Context Window to 6 Scoped Subagents

DevFeed: [From 1 Bloated Context Window to 6 Scoped Subagents](<https://devfeed.tech/articles/from-1-bloated-context-window-to-6-scoped-subagents-18306.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/subagents-are-context-engineering>)

Author: Paul Iusztin

Published: 2026-09-01T05:00:29Z

Content type: article

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [context window](<https://devfeed.tech/topics/context-window.md>), [Programming](<https://devfeed.tech/topics/programming.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-context](<https://devfeed.tech/tags/agent-context.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-application](<https://devfeed.tech/tags/ai-application.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [ai-products](<https://devfeed.tech/tags/ai-products.md>), [bench](<https://devfeed.tech/tags/bench.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [brain](<https://devfeed.tech/tags/brain.md>), [build](<https://devfeed.tech/tags/build.md>), [builder](<https://devfeed.tech/tags/builder.md>), [building](<https://devfeed.tech/tags/building.md>), [catalog](<https://devfeed.tech/tags/catalog.md>), [claude](<https://devfeed.tech/tags/claude.md>), [context](<https://devfeed.tech/tags/context.md>), [context-window](<https://devfeed.tech/tags/context-window.md>), [harness](<https://devfeed.tech/tags/harness.md>), [model](<https://devfeed.tech/tags/model.md>), [subagents](<https://devfeed.tech/tags/subagents.md>)

### AI overview

The lesson presents subagents as a way to manage context: for a deep-research setup, the author replaced one Claude Code session reading up to 200 notes, transcripts, and PDFs sequentially with six parallel researcher subagents. Each dropped its internal state and returned only final results to the orchestrator; researchers rephrased earlier searches to explore new perspectives. The lesson aims to teach subagent-parent protocols, an agent registry, and parallel fan-out, while developing judgment about when subagents justify their added complexity. The excerpt ends before completing its account of swarm configuration.

### Source excerpt

The harness determines what your orchestrator agent never sees, not the model.

## Context Engineering for Coding Agents

DevFeed: [Context Engineering for Coding Agents](<https://devfeed.tech/articles/context-engineering-for-coding-agents-18295.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/context-engineering-for-coding-agents>)

Author: Paul Iusztin

Published: 2026-08-25T05:01:37Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [context window](<https://devfeed.tech/topics/context-window.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Programming](<https://devfeed.tech/topics/programming.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [codex](<https://devfeed.tech/topics/codex.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [claude](<https://devfeed.tech/tags/claude.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context](<https://devfeed.tech/tags/context.md>), [context-engineering](<https://devfeed.tech/tags/context-engineering.md>), [context-window](<https://devfeed.tech/tags/context-window.md>), [harness](<https://devfeed.tech/tags/harness.md>)

### AI overview

This tutorial explains context engineering for coding agents, focusing on memory, skills, LSP servers, compaction, and feedback loops. It presents these techniques as ways to keep an agent's context high-signal and improve coding-agent performance.

### Source excerpt

The 4 harness components that keep your context window high-signal.

## From a Raw Shell to a Sandboxed Coding Agent

DevFeed: [From a Raw Shell to a Sandboxed Coding Agent](<https://devfeed.tech/articles/from-a-raw-shell-to-a-sandboxed-coding-agent-18302.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/run-coding-agents-safely>)

Author: Paul Iusztin

Published: 2026-08-18T11:02:59Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [codex](<https://devfeed.tech/topics/codex.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [article](<https://devfeed.tech/tags/article.md>), [backends](<https://devfeed.tech/tags/backends.md>), [bench](<https://devfeed.tech/tags/bench.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [command](<https://devfeed.tech/tags/command.md>), [commands](<https://devfeed.tech/tags/commands.md>), [compute](<https://devfeed.tech/tags/compute.md>), [computer](<https://devfeed.tech/tags/computer.md>), [computer-use](<https://devfeed.tech/tags/computer-use.md>), [context](<https://devfeed.tech/tags/context.md>), [guide](<https://devfeed.tech/tags/guide.md>), [harness](<https://devfeed.tech/tags/harness.md>)

### AI overview

The lesson presents sandboxing as a way to contain coding-agent tools and protect the rest of a user's system. It plans to connect the Decode harness to local Docker and remote Modal sandbox backends, and to compare other options and their trade-offs. The excerpt says the remote Modal workflow will set up a repository, implement a requested feature, and produce a pull request for review; it also mentions GPU compute and scaling, but does not explain how these are configured.

### Source excerpt

The guide to isolating your harness and safely executing its commands, locally or remotely.

## 11 Context-Engineering Tips for Using Coding Agents Efficiently

DevFeed: [11 Context-Engineering Tips for Using Coding Agents Efficiently](<https://devfeed.tech/articles/11-tips-to-run-coding-agents-24-7-on-one-subscription-18290.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/11-context-engineering-tips-cut-coding-agent-tokens>)

Author: Paul Iusztin

Published: 2026-08-04T11:04:10Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Documentation](<https://devfeed.tech/topics/documentation.md>), [anthropic](<https://devfeed.tech/topics/anthropic.md>), [codex](<https://devfeed.tech/topics/codex.md>), [Obsidian](<https://devfeed.tech/topics/obsidian-md.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context-engineering](<https://devfeed.tech/tags/context-engineering.md>), [docs](<https://devfeed.tech/tags/docs.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [weekly](<https://devfeed.tech/tags/weekly.md>)

### AI overview

The author shares 11 context- and harness-engineering techniques for using coding agents while controlling token use. The advice includes progressive disclosure, detailed planning with cheaper models handling execution, LLM wikis for project context, and subagents that return summaries.

### Source excerpt

Context engineering techniques to avoid wasting tokens on what doesn't matter

## The Bare-Bones Coding Agent Loop

DevFeed: [The Bare-Bones Coding Agent Loop](<https://devfeed.tech/articles/the-bare-bones-coding-agent-loop-18307.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/the-coding-agent-loop>)

Author: Paul Iusztin

Published: 2026-07-28T13:54:35Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Text-based user interface](<https://devfeed.tech/topics/tui.md>), [Terminal](<https://devfeed.tech/topics/terminal.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [ai-products](<https://devfeed.tech/tags/ai-products.md>), [aider](<https://devfeed.tech/tags/aider.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [article](<https://devfeed.tech/tags/article.md>), [bash](<https://devfeed.tech/tags/bash.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [terminal](<https://devfeed.tech/tags/terminal.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

The lesson presents the coding agent harness--the loop and surrounding tools--as a key factor in agent performance: in LangChain's Terminal-Bench experiment, changing the harness while keeping the model the same moved an agent from about 30th place into the top five. It teaches a Python implementation of a bare-bones agent loop, including how the loop turns a goal into a plan and executes it, which tools are critical versus optional, and how to build a terminal UI. The author says a one-shot implementation became brittle when skills, memory, and compaction were added, motivating an architecture-first approach informed by Claude Code, OpenCode, Aider, and Pi. The excerpt also flags that naïvely sending terminal messages can corrupt the loop; it does not provide the detailed remedy.

### Source excerpt

One agent loop, 9 tools, and a terminal you can steer.

## Building a Coding Agent From Scratch

DevFeed: [Building a Coding Agent From Scratch](<https://devfeed.tech/articles/building-a-coding-agent-from-scratch-18293.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/building-a-coding-agent-from-scratch-system-design>)

Author: Paul Iusztin

Published: 2026-07-22T11:04:24Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Programming](<https://devfeed.tech/topics/programming.md>), [LangChain](<https://devfeed.tech/topics/langchain.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [codex](<https://devfeed.tech/topics/codex.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [langchain](<https://devfeed.tech/tags/langchain.md>)

### AI overview

This tutorial explains how to build a coding-agent harness from scratch in Python. It covers the agent loop, shell execution, context engineering, subagents, remote parallel agents, and evaluation workflows, using the project Decode as the practical example.

### Source excerpt

Designing the harness around the model, from the agent loop to a remote swarm.

## How to Build Unified Agent Memory from Scratch

DevFeed: [How to Build Unified Agent Memory from Scratch](<https://devfeed.tech/articles/agent-memory-from-scratch-18297.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/how-to-implement-a-unified-memory-from-scratch>)

Author: Paul Iusztin

Published: 2026-07-14T05:01:54Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI Engineering](<https://devfeed.tech/topics/ai-engineering.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>), [LangChain](<https://devfeed.tech/topics/langchain.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Database](<https://devfeed.tech/topics/database.md>), [Model Context Protocol](<https://devfeed.tech/topics/model-context-protocol.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [graph](<https://devfeed.tech/tags/graph.md>), [graph-database](<https://devfeed.tech/tags/graph-database.md>), [knowledge-graphs](<https://devfeed.tech/tags/knowledge-graphs.md>), [langchain](<https://devfeed.tech/tags/langchain.md>), [llm](<https://devfeed.tech/tags/llm.md>), [mcp](<https://devfeed.tech/tags/mcp.md>)

### AI overview

A tutorial on building a unified agent memory layer from scratch using knowledge graphs, including ingestion, querying, and serving. It discusses trade-offs among vector databases, graph databases, temporality, versioning, MCP servers, CLIs, and skills, and explains why understanding the underlying memory layer matters.

### Source excerpt

Ingest, query, and serve a unified memory from a single database.

## Your Second Brain Is a Graveyard. Make It Agent Memory.

DevFeed: [Your Second Brain Is a Graveyard. Make It Agent Memory.](<https://devfeed.tech/articles/your-second-brain-is-a-graveyard-make-it-agent-memory-18300.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/llm-wiki-agent-memory>)

Author: Paul Iusztin

Published: 2026-07-07T05:01:21Z

Content type: article

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>)

Tags: [agent-memory](<https://devfeed.tech/tags/agent-memory.md>), [ai](<https://devfeed.tech/tags/ai.md>), [llm](<https://devfeed.tech/tags/llm.md>), [memory](<https://devfeed.tech/tags/memory.md>), [wiki](<https://devfeed.tech/tags/wiki.md>)

### AI overview

The excerpt proposes an AI Research OS as a persistent memory layer between a second-brain archive and agent harnesses such as Codex or Claude Code. It runs research across selected notes and the open web, then stores findings in a queryable, maintainable LLM wiki that agents can reuse across sessions for research, coding, and content creation. The author says the wiki need not ingest all 10,994 notes, which keeps ingestion costs low. The excerpt does not explain the build steps or how sources are selected.

### Source excerpt

Turn dead notes into a living LLM wiki your AI agents can query, maintain, and grow

## From Harness Lock-In to Portable Context Layer

DevFeed: [From Harness Lock-In to Portable Context Layer](<https://devfeed.tech/articles/from-harness-lock-in-to-portable-context-layer-18308.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/the-context-layer>)

Author: Paul Iusztin

Published: 2026-06-30T08:20:17Z

Content type: article

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Model Context Protocol](<https://devfeed.tech/topics/model-context-protocol.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [vendor lock-in](<https://devfeed.tech/topics/vendor-lock-in.md>), [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai-assistant](<https://devfeed.tech/tags/ai-assistant.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [bm25](<https://devfeed.tech/tags/bm25.md>), [business-logic](<https://devfeed.tech/tags/business-logic.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [context-layer](<https://devfeed.tech/tags/context-layer.md>), [conversations](<https://devfeed.tech/tags/conversations.md>), [fable](<https://devfeed.tech/tags/fable.md>), [filesystem](<https://devfeed.tech/tags/filesystem.md>), [google](<https://devfeed.tech/tags/google.md>), [harness](<https://devfeed.tech/tags/harness.md>), [interface](<https://devfeed.tech/tags/interface.md>), [knowledge-graph](<https://devfeed.tech/tags/knowledge-graph.md>), [mcp-servers](<https://devfeed.tech/tags/mcp-servers.md>), [skills](<https://devfeed.tech/tags/skills.md>), [unified-memory](<https://devfeed.tech/tags/unified-memory.md>)

### AI overview

The article argues that keeping personal and business context separate from an AI harness reduces switching costs when models, tools, or pricing change. It describes a three-part setup: unified memory, a serving layer, and a disposable harness. The memory combines a filesystem, BM25 keyword search, semantic vector search, and a knowledge graph of typed Person, Organization, Location, Event, and Object entities. MCP servers or skills expose that memory and its business logic to agents. The author cites Scrabble, connected to Obsidian, Readwise, Notion, and Google tools, and a separate personal-assistant project. The excerpt does not provide the rest of the architecture or evidence for the promised five-minute setup.

### Source excerpt

Build a unified memory (knowledge graph or an LLM wiki) and serve it over MCP servers or skills, so any agent, open or closed, plugs in within minutes.

## How Evaluation-Driven Development (EDD) Works

DevFeed: [How Evaluation-Driven Development (EDD) Works](<https://devfeed.tech/articles/how-evaluation-driven-development-edd-works-18296.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/how-evaluation-driven-development-works>)

Author: Paul Iusztin

Published: 2026-06-23T08:57:02Z

Content type: article

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [software-development](<https://devfeed.tech/topics/software-development.md>), [Interactive AI Systems](<https://devfeed.tech/topics/interactive-ai-systems.md>), [Trustworthy AI](<https://devfeed.tech/topics/trustworthy-ai.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [Pull Request](<https://devfeed.tech/topics/pull-request.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-companion](<https://devfeed.tech/tags/ai-companion.md>), [ai-engineer](<https://devfeed.tech/tags/ai-engineer.md>), [ai-evals](<https://devfeed.tech/tags/ai-evals.md>), [ai-native](<https://devfeed.tech/tags/ai-native.md>), [branch](<https://devfeed.tech/tags/branch.md>), [builder](<https://devfeed.tech/tags/builder.md>), [case-study](<https://devfeed.tech/tags/case-study.md>), [change](<https://devfeed.tech/tags/change.md>), [companion](<https://devfeed.tech/tags/companion.md>), [compare](<https://devfeed.tech/tags/compare.md>), [data](<https://devfeed.tech/tags/data.md>), [data-stack](<https://devfeed.tech/tags/data-stack.md>), [development](<https://devfeed.tech/tags/development.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [experiment](<https://devfeed.tech/tags/experiment.md>), [performance](<https://devfeed.tech/tags/performance.md>)

### AI overview

Evaluation-Driven Development (EDD) treats agent changes as experiments: run the same tests before and after a change, then compare results to detect regressions and measure a new feature's performance before merging. The excerpt illustrates why this matters with a Workpath AI Companion prompt cleanup that led the agent to fabricate IDs it previously handled correctly. It also identifies the challenge of testing brand-new features when no dataset, historical traces or ground truth exist, but the excerpt ends before explaining how to generate test data or use evaluators.

### Source excerpt

Turn every AI agent change into a measured experiment you compare before and after to detect regressions and measure performance.

## Build, Configure, or Use As-Is: The Agentic Harness

DevFeed: [Build, Configure, or Use As-Is: The Agentic Harness](<https://devfeed.tech/articles/build-configure-or-use-as-is-the-agentic-harness-18292.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/agentic-harness-system-design>)

Author: Paul Iusztin

Published: 2026-06-09T05:00:28Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [Tool](<https://devfeed.tech/topics/tool.md>), [AI Engineering](<https://devfeed.tech/topics/ai-engineering.md>), [codex](<https://devfeed.tech/topics/codex.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [Pydantic](<https://devfeed.tech/topics/pydantic.md>), [TypeScript](<https://devfeed.tech/topics/typescript.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [article](<https://devfeed.tech/tags/article.md>), [building](<https://devfeed.tech/tags/building.md>), [memory](<https://devfeed.tech/tags/memory.md>), [permission](<https://devfeed.tech/tags/permission.md>), [sandbox](<https://devfeed.tech/tags/sandbox.md>), [sandboxes](<https://devfeed.tech/tags/sandboxes.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

This article explains the shared system design of agentic harnesses, covering tools, agent catalogs, subagents, skills, memory, sandboxes, and permissions. It argues that these harness components are increasingly standardized, while the main differentiation lies in the context and business layers built on top.

### Source excerpt

A component-by-component teardown of an agentic harness, from tools and skills to memory, sandbox, and permissions.

## How to Keep Your AI Agent's Knowledge Graph Clean

DevFeed: [How to Keep Your AI Agent's Knowledge Graph Clean](<https://devfeed.tech/articles/how-to-keep-your-ai-agent-s-knowledge-graph-clean-18298.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/keep-knowledge-graph-clean>)

Author: Paul Iusztin

Published: 2026-06-02T05:00:53Z

Content type: tutorial

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [graph-database](<https://devfeed.tech/topics/graph-database.md>), [Neo4j](<https://devfeed.tech/topics/neo4j.md>)

Tags: [agent-memory](<https://devfeed.tech/tags/agent-memory.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [deduplication](<https://devfeed.tech/tags/deduplication.md>), [entity-resolution](<https://devfeed.tech/tags/entity-resolution.md>), [graph](<https://devfeed.tech/tags/graph.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [knowledge-graph](<https://devfeed.tech/tags/knowledge-graph.md>), [knowledge-graphs](<https://devfeed.tech/tags/knowledge-graphs.md>), [memory](<https://devfeed.tech/tags/memory.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>)

### AI overview

The key to keeping a growing knowledge graph trustworthy is to separate entity resolution, which handles naming, from deduplication, which determines identity. Treating both as one fuzzy check can silently merge distinct real-world entities and erode trust in the graph. The excerpt introduces a five-step pipeline that takes a document or conversation turn and produces canonical, deduplicated nodes wired into the graph: an LLM first extracts entities and relationships as triplets, guided by a POLE+O, Facts, and Preferences ontology so it extracts only relevant entity types. The excerpt ends during this step, so it does not establish the remaining stages or the two safety nets mentioned.

### Source excerpt

The resolution, deduplication, and review pipeline that keeps agent memory usable as it grows.

## Stop Chasing the Perfect Ontology

DevFeed: [Stop Chasing the Perfect Ontology](<https://devfeed.tech/articles/stop-chasing-the-perfect-ontology-18303.md>)

Original publisher: [Read original article](<https://www.decodingai.com/p/ship-a-knowledge-graph-ontology-in-5-minutes>)

Author: Paul Iusztin

Published: 2026-05-26T05:00:51Z

Content type: opinion

Language: en

Sources: [Decoding ML](<https://devfeed.tech/sources/decoding-ml.md>)

Topics: [schema design](<https://devfeed.tech/topics/schema-design.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [schema-evolution](<https://devfeed.tech/topics/schema-evolution.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-memory](<https://devfeed.tech/tags/agent-memory.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [approach](<https://devfeed.tech/tags/approach.md>), [article](<https://devfeed.tech/tags/article.md>), [articles](<https://devfeed.tech/tags/articles.md>), [base](<https://devfeed.tech/tags/base.md>), [brain](<https://devfeed.tech/tags/brain.md>), [build](<https://devfeed.tech/tags/build.md>), [content](<https://devfeed.tech/tags/content.md>), [content-creation](<https://devfeed.tech/tags/content-creation.md>), [core](<https://devfeed.tech/tags/core.md>), [data](<https://devfeed.tech/tags/data.md>), [extend](<https://devfeed.tech/tags/extend.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [generic](<https://devfeed.tech/tags/generic.md>), [ontology](<https://devfeed.tech/tags/ontology.md>)

### AI overview

The author recommends starting a knowledge-graph memory with a small, fixed but extendable POLE+O model, rather than trying to design a complete ontology upfront. The model uses generic noun types plus Preferences and Facts for information that does not fit those nouns. As you explore real data, add subtypes where the generic categories clash with it. The author argues that this lowers the barrier to getting started while allowing the graph to better represent the domain over time; an inadequate ontology can make GraphRAG perform worse than simple RAG. The excerpt does not spell out POLE+O's noun categories or implementation details.

### Source excerpt

Start with a fixed, generic base and extend only when your data demands it.

[Next page](<https://devfeed.tech/sources/decoding-ml.md?cursor=WyIyMDI2LTA1LTI2VDA1OjAwOjUxKzAwOjAwIiwgImM0OTZhMmFkLTVhMDQtNDY0My1iNWQzLTk0YTBmY2JlOTY4NiJd>)