The terminal is a piece of software most developers use constantly and almost nobody has questioned in decades. Warp rebuilt it, adding block based output where each command and its result form a discrete unit, modern text editing, and a set of AI features that have grown from command suggestions into a full agent mode. The documentation covers all of it clearly, and the product is more interesting than the AI framing initially suggested. Blocks are the underrated feature and the documentation explains them well.
Being able to select, copy, share or rerun a single command and its output, rather than scrolling through an undifferentiated stream of text, is a genuine improvement that you notice within an hour. It has nothing to do with AI and it is the change most likely to make you keep using the product. Command generation from natural language works better than I expected. Describing what you want and getting the correct invocation of a tool with obscure flags is a real saving, particularly for the commands you use rarely enough to never remember.
The documentation is honest that generated commands should be read before running, which is the correct advice and easily ignored when the suggestion looks plausible. Agent mode is where the product has moved and the documentation handles it responsibly. The agent can run commands, read output and iterate towards a goal, which is powerful and obviously risky in a terminal with your permissions. The material covers approval settings, boundaries and what the agent can and cannot do without asking, and that coverage is more thorough than a lot of comparable products offer.
Anyone enabling it should read that section rather than accepting defaults. Migration documentation is genuinely helpful. Existing shell configurations, aliases, prompt customisation and the accumulated setup that developers are reluctant to abandon are addressed properly, which lowers the cost of trying it. That attention to switching cost is sensible product thinking and the docs reflect it.
Privacy is where I want more than the documentation provides. AI features require sending context to remote services, and in a terminal that context can include file paths, command history, output containing credentials and details about internal infrastructure. The documentation explains the settings and the data handling in general terms, and it does not confront the question as directly as it should. Anyone working on sensitive systems needs to understand exactly what leaves the machine, and getting that clarity takes more digging than it ought to.
The account requirement is a cultural obstacle that the documentation cannot solve. Requiring a login to use a terminal offends a reasonable instinct among developers, and while the reasons are understandable given the architecture, it is a genuine reason people refuse to try it. Resource usage is heavier than a conventional terminal, which is the expected cost of a modern application and noticeable on older hardware or with many windows open. The documentation does not really address performance expectations.
Three point eight. Well written documentation for a terminal that is genuinely better than the incumbent even before the AI features, with responsible coverage of agent permissions and insufficient directness about what context leaves your machine. Try it for the blocks and read the privacy material before you point it at anything confidential.