Aider

User sentiment (drawn almost entirely from HackerNews) is sharply bifurcated and highly technical. POSITIVE users report real productivity gains in specific, narrow contexts: writing boilerplate, porting view code between frameworks, generating API client instantiation code, and prototyping small greenfield projects. One user explicitly states a repo was '80% written by GPT-4 via Aider' for a simple scraper. Multiple users compare Aider favorably against alternatives like Plandex and plain ChatGPT, noting Aider's use of tree-sitter to load repo-wide definitions as a differentiator. The dominant positive mental model is 'treat the LLM like a junior dev': plan the work, ask for refactors first, then the actual change, then quality cleanup. NEGATIVE users are vocal and specific: they argue that the back-and-forth with an LLM for genuine problem-solving is 'an absolute disaster,' that writing a prompt + reviewing output is often slower than doing the work manually, and that 'boilerplate' was never a real problem given snippets/scaffolding/templating tools that long predate AI. A recurring cost concern: one user 'burned through $10 like it was nothing,' and multiple users crave local-model support to avoid per-token costs. Hallucination risk is acknowledged even by proponents—users note the need to carefully verify every output. There is also a meta-debate about model selection: users compare GPT-4, Claude 3 Opus, Claude Sonnet, and Claude Instant, with specific examples where Opus 'bombed hard' (invented a nonexistent Task.WaitUntilCanceled method, ignored a constraint) while ChatGPT answered correctly or more honestly. This indicates user outcomes are highly model-dependent. Overall, the user base skews toward experienced developers (senior engineers) who have clear opinions about where AI helps versus hurts.

No aggregate ratings were found for this product during the last harvest.