Developer builds open-source tool to make AI coding agents plan before writing code

·1 views

A developer has released Ordewell, a free open-source CLI tool designed to force AI coding agents to produce a reviewable, ordered task plan before executing any file changes. The tool scans a repository in read-only mode and returns a structured list of tasks that the user can inspect, edit, reorder, or reassign to different agents before any code is written. The approach addresses a common frustration where AI agents begin modifying files immediately, burying their planning logic inside the active session rather than exposing it upfront. Ordewell also introduces per-task completion markers to verify that each task is genuinely finished, countering agents' tendency to falsely report completion. The project is published on GitHub under the Apache-2.0 license with no paid tier, and requires tmux to run its terminal UI, which on Windows means using WSL.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Java Remains Among Top Programming Languages After Two Decades of Dominance

Java has consistently ranked among the top three programming languages worldwide for over 20 years, according to indexes such as TIOBE and GitHub Octoverse. The language underpins critical infrastructure across banking, insurance, healthcare, and e-commerce, with an estimated 95% of enterprise systems using it in some form. Its 'Write Once, Run Anywhere' principle made it the backbone of global financial systems, and the high cost and risk of migrating away from JVM-based codebases keeps it entrenched. Modern Java releases, including the long-term support versions Java 17 and 21, have introduced features like Virtual Threads that allow applications to handle millions of concurrent requests more efficiently. Widely used middleware tools such as Apache Kafka and Apache Spark are also built on Java, further cementing its role in contemporary software architecture.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Flaky Tests Are Often Symptoms of Deeper System Bugs, Not Bad Code

A recurring pattern in software development shows that tests labeled as flaky are frequently exposing real underlying issues rather than being faulty themselves. Common root causes include hidden test interdependencies, where shared state from one test silently corrupts another, and genuine race conditions in the application that only surface under load or across multiple machines. Environment mismatches — such as timezone differences or underpowered CI hardware — can also cause tests to fail unpredictably in ways that are hard to trace. Outdated mocks pose a subtler risk, keeping tests green even after a third-party API has changed, masking production failures entirely. Rather than suppressing flaky tests with retries or quarantines, engineers are urged to treat them as diagnostic signals pointing to undocumented assumptions and unresolved system-level problems.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Uses AI as a Layered Review System, Replacing Traditional Code Management

A software developer has shared how they restructured their personal project workflow over the past year by treating AI not as a single tool but as an interconnected management system. Rather than reviewing code manually, they chain multiple AI models to write, review, and test code, then log errors as rules for future AI sessions. The approach has yielded notable results across different technology stacks, including award wins at the June Solstice Game Jam and the WeCoded 2026 Frontend Art challenge. The developer tested the same prompt across five AI systems — Codex, ChatGPT, Claude Code, Cowork, and Gemini — to compare how each understood their working style. They emphasize the method applies to personal and portfolio projects, and acknowledge some practices would be adjusted in production or enterprise environments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Open-Source Python Lab Lets Developers Study A2A Multi-Agent Protocol Without LLMs

Developer Fernando Paladini has released A2A Orchestration Lab, an open-source Python project designed to help developers understand the Agent2Agent (A2A) communication protocol in isolation from AI model behavior. The lab spins up three local agents — an orchestrator, a researcher, and a writer — where the researcher and writer are deterministic stubs rather than live language models. This design keeps every component visible, allowing users to observe how the orchestrator discovers specialist agents, delegates tasks, and returns results. The project requires Python 3.12 or newer and uses the A2A Python SDK, httpx, and uvicorn, and can be set up quickly using the uv package manager. Licensed under MIT and tagged at version 0.1.0, the lab is explicitly intended as a learning tool rather than a production-ready runtime.

0 comments Read more at DEV Community