TraeCode

The trait that decides it is restraint. Completion appears, you accept it or you keep typing, and nothing tries to take over the file. For developers who find agent panels distracting, that is a genuine preference rather than a limitation, and the whole-function suggestions are good enough to earn their keystrokes.

What you will notice within a month is the ceiling: no multi-file work, no terminal, no delegation. Pick it if completion is all you wanted. Pick Tabnine if you want the same thing with a clearer account of where your code goes.

The limitation is commercial and it will not be fixed. Everything a developer would want next lives in the paid editor from the same vendor, so the plugin's gaps are a segmentation strategy, and no amount of adoption changes that. A team that grows into needing more is a team being migrated, which is the intended outcome.

What it does right is the beta feature. Predicting the next edit location during a refactor addresses a real friction that completion alone does not, and that idea is worth more than the rest of the feature list.

1. The published record is a marketplace listing. It enumerates capabilities and describes none of them, so how completion is ranked, how the explanation is grounded and how a generated test is validated are all unstated. 2. Documentation this thin makes every entry a claim.

2. The absence of a benchmark is consistent rather than evasive, because nothing here asserts a capability that would require one. That is the most favourable thing available to say about the evidence base.

Folding an established extension into a newer brand converts installed users into awareness of the product the company actually wants to sell. Nearly two million existing installs make that a cheaper acquisition channel than any advertising budget, and it explains why a free tool with no revenue line continues to get engineering attention.

Moat: the parent's scale and its model line, neither of which belongs to this product. Likely path: continued existence as a funnel, ending whenever the paid editor no longer needs one. Position: use it, expect nothing, and read the deprecation notice when it comes.

Zero cost is the beginning and the end of the finance discussion. What it costs me is control: installation is a personal choice on each workstation, there is no administrative surface to enforce or remove it, and the only way I discover who has it is by auditing extensions.

Source leaves the editor continuously to a cloud service, and I have no published data handling terms to attach to that. Nothing runs in a pipeline, so there is no throughput number to justify accepting the exposure. Not yet: I would block it at the extension policy before I approved it.

The configuration surface is an extension settings page and the model behind it is not even named, so I cannot reason about what is reading my buffer, let alone change it. My provider key has nowhere to go and inference does not happen on my hardware.

There is no repository, so no fork exists to outlive the vendor, and no protocol interface, so nothing I have built can reach it. It is a free tool that costs me the only currency I care about, which is the ability to change my mind later.