Aptos rolls out AI tooling that lets coding assistants learn Move from its documentation, with an MCP server installed via one npx command

The saving the report documents is small and concrete: one npx command ^[4] in place of feeding documentation into a chatbot by hand ^[14]. Move is niche, and the report says models trained mostly on JavaScript and Python have historically struggled with it ^[12]. The report did not include spend, install counts or any test of how often an assistant now produces valid Move, so the larger claim, that builders ship better contracts, is untested.

We think the setup saving is real and the quality gain is unproven. The thesis fails if installs stay flat, if assistants fed the LLMs.txt files write Move no more reliably than assistants without them, or if faster drafting only lengthens the review queue, the risk cryptobriefing.com raises when it says an assistant that writes contracts faster can also write mistakes faster ^[9].

The skills are the part that touches money. They cover secure Move V2 contracts, security audits and tests with 100% coverage ^[7], and the report calls smart contract bugs expensive and permanent once deployed ^[10]. Full coverage means every line executed under test (whether the assertions check the right thing is a separate question), and cryptobriefing.com says the skills do not replace human scrutiny ^[9].

The counter-thesis favours Aptos. Move is built around resource safety and formal verification, and resource safety means assets cannot be accidentally copied or quietly deleted ^[11], so in our view one category of drafting error meets resistance from the language before it reaches an auditor.

The report describes no Aptos-trained model. The work it lists is a server, documentation feeds and a skills set, with the skills and the server documentation carrying update dates 25 days apart ^[13].