Replit made OpenAI the default for its agent and kept Claude as an option. Your automations need that same switch
One of Anthropic's biggest customers changed its default model, and the change happened in a setting, not a rebuild.
What happened
Speaking at TechCrunch Disrupt, Replit CEO Amjad Masad said OpenAI would replace Anthropic as the default provider behind Replit Agent. He cited cost, partnership and performance, saying Replit's internal and public benchmarks show OpenAI's Sol and Astra models doing better at building apps from scratch. Masad noted Replit had been, until recently, a top five Anthropic customer.
Afterwards he clarified that customers can still choose Claude, and that "Replit remains model agnostic." TechCrunch reports that a person familiar with the partnership disputed his on-stage comments and said it was still in place, and points out that most customers stick with the default.
Replit's docs show how that works in practice. Intelligent routing picks a model for the task, Free Mode always uses routing, and Core and Pro users can pick a primary model in Power or Max mode.
My take
I am not taking sides on which model is better. The point is that Replit could change providers because the model was already a setting.
Most client automations I audit are the opposite. The model name is hard coded in a dozen n8n nodes or Zap steps, the prompts were tuned for one vendor, and nobody knows what would break on a switch. Then a price change or an outage lands, and the switch takes a week.
What makes a switch cheap:
- Keep the provider and model in one config value or credential that every step reads.
- Keep a small test set of real inputs per step, so you can compare two models in an afternoon.
- Ask for structured output with a fixed schema, which travels between vendors better than free text.
- Have a fallback model ready for outages.
- Remember that defaults matter. Whatever you set is what the client's team will use.
Model choice will keep moving. Your automations should not need surgery each time.
Swappable models, test sets and fallbacks are built into the lead routing, CRM, reporting and agent workflows I deliver. Examples are at romielwillautomate.dev.
More posts
- MiniMax M3.1 Flash costs $0.15 in and $0.60 out per million tokens, but its hallucination rate is 93%. Give it a way to say unknownOct 20, 2026
- Attackers aimed cheap AI agents at mid-sized online stores. Successful breaches took about 23 hoursOct 20, 2026
- n8n ran a full AI stack on one NVIDIA DGX Spark with no cloud API. Invoice extraction took under a second with thinking offOct 20, 2026
