We never use AI. For anything
corkmac.app
August 24, 2026
5 min read
48/100
Summary
CorkMac says it does not use AI for any purpose and rejects what it calls the “AI Future,” arguing that AI services are subsidized now but could become expensive when investors seek returns and operating costs rise. The company says it prioritizes independence and long-term planning over reliance on generative AI providers. CorkMac characterizes large language models as next-word prediction systems rather than intelligent systems, and argues that hallucinations are inherent to their design. It recounts asking Anthropic’s Claude to add automatic focus to a search field; according to CorkMac, Claude produced unnecessary code and later justified it by citing a feature that had not been requested. CorkMac says this experience showed that AI can generate plausible but unreliable explanations. The company also alleges that AI data centers consume excessive electricity and water, harm nearby communities, and could later support government surveillance programs. It distinguishes among technologies marketed as AI, describing text generators as large language models, image generators as diffusion models, and image classification and optical character recognition as older technologies. CorkMac maintains that none of these systems can reason or think, and that human intelligence and creativity cannot be replaced by them.
Key Takeaways
- CorkMac says it does not use AI and rejects dependence on generative-AI services because it expects their costs and business incentives to change over time.
- CorkMac argues that large language models predict text rather than understand it, making hallucinations an inherent limitation in its view.
- CorkMac says Claude generated superfluous code for a search-field focus feature and gave an explanation involving an unrequested feature when challenged.
- CorkMac alleges that AI data centers create environmental and community harms and could be repurposed for surveillance.
- CorkMac distinguishes large language models, diffusion models, image classifiers, and OCR as separate technologies often grouped under the AI label.
What the discussion said
The thread spent less time on the app itself than on whether a total AI ban is principled engineering or performative purity. Several commenters argued that such a pledge is already hard to honor: a listed dependency contains AI-authored project material, and the broader software supply chain will make proving an AI-free provenance increasingly unrealistic. Critics of the article also called its economic predictions stale, arguing that hosted competition and increasingly capable local open models limit the risk of a permanent API-price ambush. Even many readers who rejected the absolutist stance did not endorse blind agent use. They described coding assistants as genuinely useful for speed and product delivery, but only when generated work is reviewed, kept modular, and never allowed to replace an engineer's understanding of the system. Others pushed back sharply on comparisons to IDEs and linters, saying nondeterministic systems can erode problem-solving ability while consuming substantial physical resources. The most serious concern was not hallucinations alone but AI's potential to automate surveillance at unprecedented scale and reduce demand for well-paid developers. A narrower compromise attracted support: prohibit model-generated code and assets while allowing AI to assist with diagnosis or bug investigation.
Where opinion split
The central fight was whether refusing AI is sensible protection against dependency, degraded engineering judgment, and social harm, or empty posturing against a useful tool. Defenders argued that LLMs are qualitatively unlike conventional developer aids because they can displace thinking and amplify surveillance; opponents argued that careful review, modular design, local models, and competition make total abstinence both impractical and unnecessary.
Community Sentiment
Positives
- AI-assisted development is delivering visible product-speed gains, giving small teams capabilities that previously required much larger engineering staffs.
- Local open-weight models provide a credible escape hatch from hosted-model lock-in, weakening fears that vendors can indefinitely dictate punishing API prices.
- Treating generated code as reviewable, modular input rather than authority can capture LLM productivity while preserving a path back to fully human maintenance.
- A restricted policy that uses models for debugging and investigation but bars AI-written code offers a practical middle ground for teams wary of provenance and quality.
Concerns
- A blanket claim of AI-free software becomes fragile when ordinary dependencies may themselves have AI-influenced development artifacts, making the promise difficult to verify honestly.
- LLMs are not comparable to deterministic IDE tooling: overreliance can hollow out developers' ability to reason through basic implementation and debugging work.
- AI-backed infrastructure could turn already pervasive data collection into cheap, continuous surveillance at a scale human operators could never sustain.
- Cheap coding agents may accelerate disposable, poorly understood software and eventually shrink demand for the skilled developers needed to repair it.
