Alex Finn tests OpenClaw 2.0 after reporting repeated reliability problems with earlier versions. An attempted update leaves his existing agent unresponsive, so he uses a fresh installation for the review. The new installation lets him inspect the interface, but it does not establish that his older setup, accumulated memories or skills have migrated successfully.
Alex Finn highlights imported Codex projects, automatic goals and separate conversations for delegated agents. A sample product-planning task creates research roles and visible subagent threads, but the overall task stalls rather than producing the intended demonstration. The review separates the value of seeing each agent’s work from the unresolved question of whether the complete workflow reliably finishes.
Alex Finn demonstrates forking a conversation at an earlier message to try a different implementation approach while retaining prior context. He also discusses inline widgets, while questioning whether a small chat panel is a useful place to test an application. Multiplayer access and some other release features are described rather than demonstrated, and he notes that much of the interface functionality is concentrated in the web app.
Alex Finn compares the tools through his own workflow: Hermes for configurable local work and model choice, Grokbot for tasks in cloud environments, and OpenClaw for its conversational personality and visibility into subagents. His preference is shaped primarily by reliability rather than the length of the feature list. These are personal experience reports, not a controlled benchmark or a guarantee that other users will encounter the same failures.
Watch on YouTube
