Devanshu Kumar Singh’s Post

Zero is an AI engineering OS. It runs locally. Built in Rust. It sits across your entire stack. You can pull in work directly from GitHub Issues, GitLab, Jira, Sentry, Bitbucket. Zero turns all of that into one thing: a case. One system, one state machine, one place where work actually happens. A Sentry error, a Jira ticket, a GitHub issue, it doesn’t matter. Same workflow. Same agent. Same proof-gated resolution. Here’s how it works in practice. You open a case (or import a case from Github, Sentry, Jira..etc). An agent picks it up and starts running sessions. It proposes tasks, writes code in isolated git worktrees, and nothing touches your branch without your approval. Every task needs proof before it can close. Tests, logs, screenshots, whatever makes it real. If a proof is missing, the case simply doesn’t resolve. That’s enforced at the system level, not a guideline people ignore. Cases move through a real state machine: OPEN → INVESTIGATING → IMPLEMENTING → VERIFYING → RESOLVED No manual tracking. Transitions happen automatically. You can see the full timeline, why something moved, what it’s waiting on, and what’s blocking it. You also get signals: health, staleness, blockers, decision hints. The system tells you exactly what’s going on instead of making you guess. Under the hood, it’s all real infrastructure. SSE streaming, isolated worktrees, GitHub PR integration, built-in terminal, Monaco editor, multiple agents running in parallel, and shared memory injected into every session. It ships as a native macOS app on Apple Silicon using Tauri. Intel and Linux are coming. Private beta is live at zero.polymathlabs.xyz. DM me for an invitation code.

I've been using Claude Code as my daily coding companion for months. It's great at writing code — but the moment I need it to check a Jira ticket, read a Slack thread, or pull context from Confluence, it hits a wall. The problem isn't capability. It's authentication. Every enterprise system sits behind SSO. There's no simple way for an AI agent to authenticate as you and access the tools you use every day — Outlook, MS Teams, Slack, Jira, Confluence. You end up copy-pasting context manually, which defeats the purpose of having an AI assistant. So I built sig — an open-source CLI that handles this. The idea is simple: → You log in once through your browser (normal SSO flow) → sig stores the credentials encrypted (AES-256-GCM) → Your AI agent uses them through scripts, without ever seeing the raw tokens What this looks like in practice: I tell Claude Code "summarize my unread Slack messages" — it calls a script, sig injects the auth, and I get a summary. Same for "check my Outlook inbox", "what's the status of JIRA-1234", or "find the latest meeting notes in Confluence". The key design decision: credentials never enter the AI's context window. sig can run as a local MITM proxy that injects auth at the network layer — the AI agent doesn't even know credentials exist. Currently supports: • Outlook, MS Teams, Slack — for daily communication • Jira, Confluence — for project management • Any cookie-based or OAuth2 system — extensible It's fully open source (MIT): https://lnkd.in/ghdUQ277 If you work with AI coding agents and keep hitting auth walls with your work tools, this might save you some headaches. Happy to chat about the approach.

Well, it's live! https://biomelab.dev/ I've building this project to make my workday more prolific. If you are into #agentic development workflows, using git worktrees for parallel task, you probably want to take a look at this project, which helps me in monitoring the ongoing work. Spoiler alert: it uses Docker, Inc Coding Agent #Sandboxes under the hood for adding safety to your workflows 😉

I kept running into the same problem across different projects. Start a new ML pipeline, API, or POC → first 30–60 minutes go into just setting up folders and files. Not the modeling. Not the logic. Just structure. And every time: Slightly different folder naming Missing files discovered later Inconsistent layouts across teams Extra back-and-forth during onboarding (“where does this file go?”) It’s a small inefficiency, but it compounds quickly in real-world projects. So I built a simple solution: Scaffold Gen A CLI tool where you define your project structure once (using tree text, YAML, or JSON), and generate it instantly. Instead of explaining structure: → You codify it → Version control it → Reuse it Problem → Manual, inconsistent project setup slows down development and introduces avoidable errors. Solution → Define structure declaratively and generate it via CLI. Impact → Faster project bootstrap (seconds instead of minutes) Standardized structure across teams Easier onboarding Cleaner, more reproducible workflows This is especially useful in: ML/MLOps pipelines Microservice-based APIs Rapid prototyping / hackathons Sometimes the most valuable improvements are not complex models, but removing friction from everyday workflows. GitHub: https://lnkd.in/dSzHGfEP Curious — how do you standardize project structures in your team?

If you’re learning to code, You’ve probably heard these two terms everywhere: Git and GitHub At first, they feel confusing. Commands. Terminals. Repositories. Too many new words. But here’s the simple way to think about it: → Git = Your version control system It tracks every change you make in your code → GitHub = Your online storage + collaboration space It helps you store, share, and work with others Think of it like this: Git is your local memory GitHub is your online backup + portfolio It’s all about: → Tracking your progress → Avoiding mistakes → Collaborating with others → Showing your work publicly If you’re serious about AI, Data, or Tech This is not optional. Every serious developer uses it. Takeaway: Don’t just learn how to code. Learn how to manage your code. Quick question: What part of Git or GitHub confuses you the most right now? 👇

 Claude Hooks Mastery — Automate Your Dev Workflow People ask claude following tasks multiple times:  "Claude, also run the linter."  "Claude, format that file."  "Claude, update the changelog."  Hooks fix that. They run YOUR scripts automatically every time Claude writes a file, runs a command, or finishes a task. No prompts. No reminders. It just happens.  A few things you can wire up in under an hour:  — Auto-format and lint every file Claude touches (Prettier, ESLint, Black)  — Block rm -rf, git push --force, and git reset --hard BEFORE they execute  — Scan every commit for leaked .env values and API keys  — Slack ping when a long task finishes  — Auto-update CHANGELOG.md on every source edit  #ClaudeCode #AIEngineering #DeveloperTools #AIAcademy #DevWorkflow

GitHub Copilot and Claude Code may both use slash commands, but they are not the same thing. Here is the simplest way to think about it: Built-in GitHub Copilot slash commands Controlled by GitHub and Microsoft Available based on your VS Code version, Copilot features, and extensions Designed for built-in actions inside Copilot Chat Examples: /explain, /fix, /tests, /doc, /new, /help Custom slash commands in Claude Code User-defined command patterns Used to shape how Claude responds Helpful for formatting, tone, structure, reasoning, and analysis Examples: /TLDR, /ELI5, /CHECKLIST, /SWOT, /COMPARE, /STEP-BY-STEP Claude Code skills Reusable automations for tasks you do often Great for turning repeated workflows into commands Examples: /review, /security, /optimize, /a11y, /test-plan My takeaway: GitHub Copilot slash commands = built-in product features Claude Code custom commands = flexible response controls Claude Code skills = repeatable workflow automation This distinction matters because many people see “slash commands” and assume they all work the same way. They do not. #GitHubCopilot #ClaudeCode #AI #DeveloperTools #VSCode #SoftwareDevelopment #Productivity #GenAI #Coding

I was asked a few times "Can I use 𝗚𝗶𝘁𝗛𝘂𝗯 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 like 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲"? My answer is: OF COURSE ! Just like Claude Code, GitHub Copilot(GHCP) can work in a "terminal like" mode (using GHCP CLI): It can plan workflows, Invoke tools automatically, read files, call APIs, query DBs, execute commands - all WITHOUT building the orchestration logic yourself. So what advantages does 𝗚𝗶𝘁𝗛𝘂𝗯 𝗖𝗼𝗽𝗶𝗹𝗼𝘁(GHCP) have over 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲? ✅ GHCP supports all the leading models (Claude, OpenAI, Gemini) including automatic model routing option, while Claude Code supports only Claude models. ✅ GHCP has native DevOps integration - especially if you use GitHub (it is GitHub workflow aware). ✅ In addition to the "Terminal mode" like Claude Code, it can work inside all popular IDEs: Microsoft IDEs (VSCode, VS), JetBrains IDEs, Eclipse, Xcode, Vim/Neovim. ✅ Enterprise Governance & Security - unlike Claude Code, you can define who can access GitHub Copilot, which features are enabled, what models are permitted, how policies are enforced across orgs and more. ✅ Comes with enterprise cost control & usage monitoring Bottom line: If you are an independent vibe coder - 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 is a good choice. If you are a vibe-coder in an organization or a professional developer who also wants to vibe-code and needs an enterprise-grade tool - 𝗚𝗶𝘁𝗛𝘂𝗯 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 is the right tool for you. #GitHub #GHCopilot #GHCP #MicrosoftAI #ClaudeCode

I have been working on both extensively and here’s my perspective. #GitHubCopilot is the accelerator. When the path is clear, Copilot keeps you in flow with fast inline completions, strong pattern recognition and seamless IDE integration. It’s perfect for features, tests and predictable day‑to‑day coding. #ClaudeCode is the problem‑solver. It thrives in ambiguity — reading large repos, reasoning across files, debugging complex systems and planning multi‑step changes. It’s built for refactors, migrations and architecture-level thinking.

Will add: it works out of the box to create dataverse columns off a simple json schema. Whenever I need a new table I generate a json schema create the base table connect to the MCP server and have copilot create all the fields.