AI Coding Tools

Code completion, agents and review tools that work inside your editor or terminal.

15 tools

Start with your coding workflow

Separate editor completion, terminal coding and coordinating several agents before choosing a tool. These guides explain the trade-offs; check each vendor for current pricing and features.

Terminal coding agents, and how to actually choose one
Nine tools, three families, and the one question that decides between them. Written for people who have to pick, not people collecting options.
Pi vs opencode — a product and a toolkit, wearing the same clothes
Both are MIT-licensed TypeScript terminal agents you point at your own model. One ships plan mode and permissions; the other deliberately ships neither. That single choice decides which one you want.
herdr vs tmux and Zellij — the terminal problem that starts at your third agent
A multiplexer keeps terminals alive. It cannot tell you which of them is waiting for an answer. That single gap is the entire case for herdr, and it is both narrower and more interesting than the pitch.

This is the category where AI has changed daily practice most concretely, and it has moved through three distinct generations in a short time: autocomplete that finishes the line you are typing, chat that answers questions about your code, and agents that read a repository, make changes across several files and run the tests. Which generation a tool belongs to matters far more than its benchmark numbers.

What to look at when choosing

Completion, chat and agent are three different products

Completion is low-friction and helps most with boilerplate. Chat is good for understanding unfamiliar code and for questions you would otherwise search. Agents can carry out a multi-step change on their own — which is the biggest leap and also the one that needs the most supervision. Buying an agent when you wanted better autocomplete is a common and expensive mismatch.

How much of your codebase it can see

A tool that only sees the open file will suggest code that ignores the helper you already wrote and the conventions your team follows. Repository-wide indexing is what makes suggestions fit the code around them. On a small script this is invisible; on a large codebase it is the difference between useful and irritating.

Where the code goes, and what your employer allows

Most of these tools send code to a server. Policies differ on retention, training and whether an enterprise tier changes any of it, and some organisations restrict which tools may touch their source at all. Check this before adopting a tool into daily work rather than after, because reversing that decision is awkward.

When you probably do not need one

These tools amplify whatever judgement you already have; they do not substitute for it. Generated code that looks right and passes a quick glance is the main hazard, and the risk is highest exactly where you know the domain least — which is also where you are most tempted to accept it. If you cannot review the output, you are not saving time, you are deferring it.