Tech • AI • Robotics • Game

ENFR
TodayPlay for freeFury BloxLeaderboardShortsTop StoriesFor youTopicsVideosYT channelsArchivesSearchFavorites

Claude Code backlash: SuperPower, BMAD costs vs Base 44

AI CodingMonday, October 26, 2026· 3 videos

Briefing

Audio player

Claude Code users reject SuperPower

SuperPower is facing a sharp backlash from advanced Claude Code users who say it adds heavy prompt overhead without improving results. The main complaint is that the framework forces rigid, multi-step workflows onto tasks newer models can already complete directly. Critics argue that means slower iteration, higher token burn and more friction in everyday coding. The dispute cuts to a broader question of whether AI coding frameworks are delivering substance or mainly workflow theater.

BMAD draws token-cost criticism

BMAD is being criticized on similar grounds, with detractors calling it a large instruction layer rather than a genuine model advance. Reported figures cited include 31 skills, 274 Markdown files and more than 12,000 lines of prompt material, all of which can inflate context usage. Opponents say that extra scaffolding raises costs and latency while making debugging harder. The criticism is especially pointed among experienced users optimizing for speed, clarity and predictable spend.

Big GitHub numbers meet skepticism

Despite the backlash, both projects retain strong visibility and social proof inside AI coding circles. SuperPower is associated with roughly 300,000 GitHub stars and about 1 million installations, while BMAD is cited at around 54,000 stars. Those adoption metrics suggest broad curiosity and aggressive word-of-mouth growth. But critics say popularity does not prove better code quality, faster delivery or lower operating cost.

Base 44 beats Lovable later

Base 44 emerged as the stronger option for structural changes in growing apps after extended production-style use. The comparison focused less on first prompts and more on what happens once databases are populated, users exist and live workflows can break. In that setting, Base 44 reportedly handled deeper architectural changes more reliably than Lovable. The result points to an important divide between tools that shine at launch and tools that hold up under maintenance.

Lovable still wins fast starts

Lovable continues to score well for landing pages, polished prototypes and smaller internal tools. In the cited comparison, it often produced strong first versions in one or two prompts, making it effective for rapid validation. A third-party survey of more than 200 developers reportedly found about 85% satisfaction for landing pages, 80% for visual prototypes and 65% for simple internal tools. That positions Lovable as a strong front-end and prototyping choice even as scaling questions persist.

Lovable growth underscores market appetite

Lovable's business momentum highlights how intensely investors are backing AI app builders. In December 2025, the company raised $330 million at a $6.6 billion valuation in a round led by Capital G, with Menlo Ventures, N Ventures, Salesforce Ventures and Databricks Ventures participating. It reportedly hit $100 million ARR within 8 months and crossed $200 million four months later, while nearing 8 million users by late 2025. Those figures suggest demand remains strong even as users become more critical of long-term maintainability.

Tokens drive quality and budgets

Tokens remain the core unit behind both model behavior and billing, making them central to the debate over AI coding workflows. Models process token fragments rather than sentences, and both prompt input and generated output are charged. One cited example priced Sonnet at about $2 for input tokens and $10 for output, illustrating how verbose workflows can get expensive quickly. That makes prompt bloat from framework-heavy setups a direct budget issue, not just a usability complaint.

Context limits punish bloated prompts

Context functions as a model's working memory, and it becomes more fragile as conversations grow longer. Because tokenization varies by model and language, the same request can consume different amounts of context on GPT-4o versus other systems, with English often more efficient than French. A common rule of thumb is roughly four characters per token, with 431 characters translating to about 104 tokens. In practice, that means oversized instruction stacks can reduce reliability over time while simultaneously pushing up cost.

Videos covered

Previous briefings · AI Coding