The repository setup that makes AI agent workflows actually work — monorepo structure, AGENTS.md, mechanical invariants, and the .claude directory that every agent session boots from.
2.The specific, repeatable process two non-traditional founders use to ship a 300k line Next.js monorepo with 3-6 AI coding agents running in parallel.
3.The actual Claude Code workflow behind Pane: discussion, planning, implementation, worktrees, and the parts we refuse to automate away.
4.How we built a repeatable software factory with Claude Code and Codex — three slash commands, custom agents, voice-first development, and the harness engineering that ties it all together.
5.Agent loops are the workflow coding agents are converging on. Pane makes them agent-agnostic, cross-platform, open source, and controllable from the runpane CLI.
6.Pat Hanrahan — Turing Award winner, GPU computing co-creator, Stanford professor — independently verified our exact development workflow: spec, read, verify.
7.A post comparing three Claude Code tools went viral. The author found the right layers but the wrong conclusion — three tools when three commands do the same job.
8.Claude Code's source code leaked. The company building the most popular AI coding agent ships code that violates every principle that makes codebases agent-readable.
9.Anthropic built fixes for every way Claude Code breaks — hallucination, context decay, lazy fixes — then gated them behind an employee flag. We built the same fixes into an open-source harness.
10.Every major coding agent shipped voice input in the last sixty days. Here's what the speed-focused framing gets wrong, and why voice works — the ramble is the signal, not the noise.
11.We built an agent pipeline that ships software almost hands-off — discussion, plan, implement, review. Here's how we turned that same machine into a business factory, and the one piece that was missing.
12.The evolution of our AI development workflow: from a human approving every step, to an autonomous pipeline with gates, to a factory that finds its own work. A candid look at what changed and why.
The evolution of our AI development workflow: from a human approving every step, to an autonomous pipeline with gates, to a factory that finds its own work. A candid look at what changed and why.
Parsa · July 10, 2026A practical guide to running Claude Code, Codex, Aider, and other AI coding agents side by side without branch conflicts. Git worktrees, agent managers, and the workflow that makes it work.
Parsa · June 25, 2026Git worktrees let AI coding agents work on the same repo without conflicts. This guide covers what worktrees are, why agents need them, and how to automate them on Windows, Mac, and Linux.
Parsa · June 25, 2026What we learned building runpane: the design decisions that make a CLI work well for AI agents, not just humans.
Parsa · June 24, 2026How we're building Pane Chat, a global orchestrator that lets you talk to one agent that creates panes, starts agents, and manages your whole workspace.
Parsa · June 20, 2026Agent loops are the workflow coding agents are converging on. Pane makes them agent-agnostic, cross-platform, open source, and controllable from the runpane CLI.
Parsa · June 18, 2026The new runpane packages give Pane a guided first-run CLI for installing the desktop app, setting up remote hosts, updating, and running diagnostics.
Parsa · June 17, 2026You can now check on your agents from your phone. Here's how we built mobile remote access for Pane and why it matters for agent workflows.
Parsa · June 10, 202670% of developers use Windows. Almost every AI agent manager is Mac-only. Here's why we built Pane for Windows, WSL, macOS, and Linux from day one.
Parsa · June 3, 2026We built an agent pipeline that ships software almost hands-off — discussion, plan, implement, review. Here's how we turned that same machine into a business factory, and the one piece that was missing.
Parsa · May 27, 2026Pane's remote daemon runs your AI agents on powerful hardware while you control them from any device. One command to set up.
Parsa · May 18, 2026Anthropic’s June 15 programmatic-credit change splits subscription usage from API/Agent-SDK usage. Here’s the full map: every popular AI coding tool, affected or not, with sources.
Parsa · May 13, 2026The actual Claude Code workflow behind Pane: discussion, planning, implementation, worktrees, and the parts we refuse to automate away.
Parsa · May 7, 2026Every major coding agent shipped voice input in the last sixty days. Here's what the speed-focused framing gets wrong, and why voice works — the ramble is the signal, not the noise.
Parsa · April 18, 2026The repository setup that makes AI agent workflows actually work — monorepo structure, AGENTS.md, mechanical invariants, and the .claude directory that every agent session boots from.
Parsa · April 16, 2026A post comparing three Claude Code tools went viral. The author found the right layers but the wrong conclusion — three tools when three commands do the same job.
Parsa · April 9, 2026Pat Hanrahan — Turing Award winner, GPU computing co-creator, Stanford professor — independently verified our exact development workflow: spec, read, verify.
Parsa · April 6, 2026Claude Code's source code leaked. The company building the most popular AI coding agent ships code that violates every principle that makes codebases agent-readable.
Parsa · March 31, 2026Anthropic built fixes for every way Claude Code breaks — hallucination, context decay, lazy fixes — then gated them behind an employee flag. We built the same fixes into an open-source harness.
Parsa · March 31, 2026How we built a repeatable software factory with Claude Code and Codex — three slash commands, custom agents, voice-first development, and the harness engineering that ties it all together.
Parsa · March 29, 2026The specific, repeatable process two non-traditional founders use to ship a 300k line Next.js monorepo with 3-6 AI coding agents running in parallel.
Parsa · March 6, 2026