Writer, the enterprise generative AI company, didn’t start as an open source shop. But according to co-founder and CTO Waseem Alshikh, the path from Hugging Face user to paying customer to open source model contributor wasn’t just a marketing pivot — it was a survival strategy. In a conversation with Hugging Face’s Jeff Boudier, Alshikh laid out why Writer is now putting its own models into the open, and his reasoning cuts against the grain of much of the current AI hype.
Alshikh is blunt about what he sees as the biggest misconception in generative AI right now: the belief that bigger models automatically mean better outcomes. Companies are burning enormous compute budgets chasing parameter counts when, as he puts it, the real work happens in how a model is tuned and served for a specific task. That’s a familiar refrain from the fine-tuning era of smaller transformer models, but hearing it from someone running production LLM workloads at scale gives it more weight than the usual hot takes on social media.
The conversation also digs into the practical economics of serving LLMs. Writer is taking a hybrid approach, running workloads on both GPUs and CPUs depending on the use case. That’s a detail worth paying attention to — most enterprise AI discussions assume GPUs are non-negotiable, but Writer’s production reality suggests CPUs still have a meaningful role when efficiency and cost matter. Alshikh frames it as a question of matching the hardware to the inference task rather than defaulting to the most expensive option available.
Hugging Face’s Expert Acceleration Program gets a nod as part of Writer’s journey, though the specifics of what the program delivered remain thin in this interview. What’s clearer is the strategic logic: contributing open source models gives Writer credibility in the developer community while also creating a feedback loop that improves their commercial offerings. It’s a playbook we’ve seen before, but Writer’s execution — moving from consumer to contributor — signals how enterprise AI companies are starting to treat open source less as a charity case and more as a distribution channel.
