3 tools reviewed
Hoplite (YC S26, Launch HN 49 points) deploys cloud coding agents that work in isolated per-thread sandboxes, run tests, drive a real browser against a live preview URL, and open reviewed pull requests. Full review: how it differs from Cursor cloud agents and Codex Cloud, pricing ($82.50/seat Pro), the no-upcharge-on-tokens model, MCP server and CLI, and the founder's answers to HN criticism.
Prime Agent is Prime Intellect's open-source coding harness built on two research abstractions: the Recursive Language Model (RLM), which treats context as variables and subagent delegation as function calls inside a persistent IPython REPL, and the Continual Harness, which lets the agent CRUD its own prompts, skills, memory, and subagents mid-task. We review the architecture, the /refine self-improvement pipeline, the ARC-AGI-3 saturation claim, and the HN debate over whether self-modifying harnesses still matter now that frontier models caught up.
Warp took the agent built into its terminal and shipped it as a standalone CLI that runs anywhere — Ghostty, iTerm2, VS Code, Windows Terminal. The differentiator is a tmux-like multiplexing architecture: persistent sessions that survive directory changes, agents that drive full-screen apps like sqlite and gdb, SSH sessions with no remote binary install, and cloud-agent handoff that can delegate to other harnesses like Claude Code and Codex. Review of the mux architecture, the $18/month pricing, and the HN debate about whether Warp's AI push broke the terminal people loved.