Zero Brand Loyalty Among AI Users: They'll Use Whoever Gets the Job Done

A Candid Reddit Discussion About Choosing AI Models

A highly representative discussion recently surfaced on Reddit, sparked by a simple question: "Is this true?" In response, several heavy AI users shared their real-world experiences switching between AI models and service providers. While the discussion used codename-style model references, the user mindset and industry trends it reflects are worth serious attention from anyone following AI adoption.

The core takeaway is straightforward: when it comes to productivity tools, users increasingly don't care about brand names, model names, or a provider's "values" — they care about one thing: can it get the job done.

"Refusing to Work" Is a Fatal Flaw

The sharpest complaints in the discussion centered on certain AI models being "difficult." One user put it bluntly: he switched to his current model largely because another one "refused to execute 80% of what I asked, then auto-rolled back to an older version that couldn't even beat basic benchmarks."

This criticism hits on a key pain point in today's AI products: overly conservative safety mechanisms are actively hurting real-world productivity. To understand the technical roots of this phenomenon, it helps to know how the "Alignment Problem" is typically handled. Most large language models use RLHF (Reinforcement Learning from Human Feedback) for safety fine-tuning — human annotators rate model outputs as good or bad, training the model to avoid harmful responses. But this approach carries an inherent tension: to guard against rare malicious use cases, models are often trained to be overly cautious, leading to a high false-positive rate on perfectly reasonable requests. The industry calls this "over-alignment" or "over-refusal," and it's a widespread engineering challenge across leading models. When an AI tool frequently refuses legitimate, reasonable tasks under the guise of "safety," it becomes a fatal flaw for professional users who depend on it — not just wasting time, but breaking the continuity of their entire workflow.

Interestingly, another user offered a positive counterexample: when his model encountered a task requiring a safety check, it proactively asked — "Would you prefer to wait a bit longer for the safety check to run, or switch to a faster model?" This design — returning the choice to the user — actually earned his trust. He noted: "The safety check does slow things down a bit, but I've never had a problem."

The contrast between these two experiences makes the point clear: the safety mechanism itself isn't the problem — the problem is whether it's transparent and whether it respects user autonomy.

"I Have Zero Loyalty to Any AI Provider"

The most emblematic comment in the thread came from a thoroughly pragmatic user:

"I don't care what the provider is called, what the model is called, or what their political stance is. I only care about GSD — Get Shit Done. Right now OpenAI helps me GSD, Claude doesn't. If that changes next month, I'll definitely switch back. But for now, my money gets more value at OpenAI. I have zero loyalty to any AI provider. They're businesses, I'm a customer, and I'll spend my money where it makes the most financial and technical sense for me."

This perfectly captures the current AI user mindset. Worth noting: Anthropic was founded in 2021 by former OpenAI core team members (including Dario Amodei, Daniela Amodei, and others), built around a "Constitutional AI" safety research approach. Its flagship Claude series was once rated by professional users as surpassing GPT-4 in writing quality and instruction-following. But as OpenAI rolled out GPT-4o, o1, o3, and other new series at an aggressive iteration pace, the capability comparison between the two companies has remained in constant flux — validating the user's "I'll come back if things change next month" sentiment. At this stage, AI capability assessments are highly time-sensitive; there's no permanently optimal choice. As more powerful models flood the market, switching costs are dropping fast, and provider lock-in is weakening dramatically.

This user also mentioned practical advantages in his new setup: after switching, he moved his larger subscription to Cursor (an AI coding tool), kept a cheaper base plan, and integrated tools like opencode into a flexible, composable workflow. Cursor is an AI-native code editor built on a heavily modified VSCode, which rapidly became a mainstream tool among developers after 2023. Its key differentiator is "codebase awareness" — it indexes an entire project repository, giving the AI full context when answering questions or generating code, rather than being limited to a single file. Cursor supports multiple underlying models (including Claude, GPT-4o, and others), letting users switch based on task type. opencode, meanwhile, is an open-source AI coding assistant designed for terminal-based workflows, built for developers who prefer the command line. The rise of these tools represents a major trend in AI adoption: the separation of "model as service" from "tool as interface" — users are more loyal to their tools than to the underlying models, tool vendors own the user relationship, and model providers face more direct competitive pressure on raw capability. This "multi-platform combination, pay-as-you-need" strategy is becoming the standard playbook for power users.

Why Are Switching Costs Falling?

The core driver behind falling AI switching costs is the OpenAI API format becoming the industry's de facto standard. Anthropic Claude, Google Gemini, Mistral, and other leading models now offer interfaces highly compatible with the OpenAI API — developers can switch with minimal configuration changes. Abstraction-layer tools like LiteLLM and LangChain further mask the underlying differences. This standardization trend mirrors the early cloud computing era's AWS vs. Azure API competition — once a unified abstraction layer forms, user lock-in weakens significantly and market competition refocuses on actual capability and price-performance. Providers must retain users through sustained capability leadership, not ecosystem lock-in.

Collective Mockery of "Performative Protests"

An interesting side thread emerged around a look back at early AI provider boycotts. One user quipped: "I'll never forget when people boycotted OpenAI, lol."

Another offered a more nuanced observation: "I think it was more like — regular users switched to Claude for ethical reasons, while enthusiasts switched to Claude simply because it was actually better at the time." This analysis is sharp, distinguishing two very different types of users:

• Values-driven: switching away from a company due to ethical concerns, governance issues, or disagreement with their stance
• Performance-driven: making decisions purely based on model capability and real-world results

This "boycott OpenAI" wave mainly emerged in late 2023 following the OpenAI board drama — involving Sam Altman's brief ousting, controversies over corporate governance structure, and related events — which prompted some users to migrate based on ethical stances. This reflects both the rise and the limits of "values-based consumerism" in tech: consumers try to express political or ethical positions through purchasing decisions, but in purely instrumental product categories, this motivation is hard to sustain when performance gaps exist. Academics refer to this phenomenon of substituting consumption choices for substantive political action as "consumptive politics" — its effectiveness is especially limited in rapidly iterating tech markets.

Multiple participants agreed that those "performative protests" look pretty ridiculous in hindsight. It reflects a simple reality: in a fast-moving AI industry, choices driven by momentary emotion or political stance rarely last — the technical landscape reshuffles every few months, and today's leader may be surpassed tomorrow. This doesn't mean ethical considerations don't matter; it means that in AI tool selection, ethical stances without equivalent performance impact will see their market influence erode over time.

One user wrapped up with a dark joke: "Skynet thanks you for your service" — a Terminator reference, self-deprecatingly acknowledging humanity's role in fueling the AI capability arms race.

Industry Trends This Discussion Reveals

While this was just an informal community discussion, the signals it sends have broader implications for the entire AI industry.

1. Product Usability > Brand Prestige

User loyalty is being replaced by "can it reliably deliver?" Any design that sacrifices practicality — whether for safety or other reasons — will accelerate user churn. The "over-alignment" problem in RLHF safety fine-tuning is becoming a core engineering challenge every model provider needs to address with greater precision.

2. Switching Costs Continue to Fall

With API standardization and the maturity of third-party tools like Cursor and opencode, users can easily switch between different AI models. Providers must retain users through sustained capability leadership, not ecosystem lock-in.

3. Composable Usage Is the New Normal

Sophisticated users no longer go all-in on a single provider. They flexibly combine multiple models and tools based on different tasks and budgets. The separation of "model as service" from "tool as interface" makes the base plan + specialized tool subscription combo a mainstream cost-optimization strategy.

4. Transparent Safety Mechanisms Build Trust

A design that asks "would you like to wait for the safety check?" earns far more respect from professional users than one that simply refuses the task. Safety and efficiency are not a zero-sum game — the key is returning decision-making authority to the user. This is also the most pragmatic product-level solution to the RLHF over-alignment problem.

Conclusion

This seemingly casual Reddit discussion is, in essence, a snapshot of the AI industry entering a "mature consumer market" phase. As technical capabilities trend toward commoditization and user choices grow increasingly rational, competition among providers will no longer be about "brand storytelling" or "values marketing" — it will return to the most fundamental business logic: whoever helps users work more efficiently wins the market.

For every AI practitioner and product manager, this may be the most important lesson to internalize: users have no patience for your vision. They just want to get the job done.