Visual summary of operating lessons from Charlie O’Neill.

Lessons from Charlie O’Neill

Charlie O'Neill is an AI researcher who leads Base Labs at Baseten and previously co-founded Parsed. These lessons are drawn from their own published essays and an X article. — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "Every model has an intelligence ceiling: the fraction of economically valuable work it can automate at a cost and quality where you’d be irrational not to deploy it". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "This is a lower bar than human-level performance". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "most of that ceiling can only be reached through specialisation". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "most enterprise work lives in the long tail: proprietary edge cases, domain-specific data formats, error modes that only surface in production". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "The bottleneck is task-specific training data and the RL loops to learn from it". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "Over the course of numerous customer engagements, we’ve found a surprising amount of what we call “hard” work is actually “specific” work, and specificity is what specialisation buys you". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "I want to be honest about where this argument doesn’t hold". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay "The Michael Scott Paper Company of AI", O'Neill wrote: "a sub 80B model fine-tuned on a client’s claims data will outperform GPT-5.2 on that client’s claims workflow, not because it’s a smarter model, but because it’s been trained on the actual distribution it needs to perform on". — The Michael Scott Paper Company of AI, Charlie O'Neill (February 14, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "Modern LLMs are amnesiacs". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "continual learning is inseparable from specialization". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "The cost of remembering scales with the scope of what you’re trying to remember. Narrow the scope, and the problem becomes tractable". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "Generality has overhead, and that overhead grows faster than linearly with scope". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "The brain solves continual learning by not attempting generality in the first place". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "The intractability of continual learning in generalist models is the invisible hand pushing deployment toward narrow experts". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "The monoliths will be impressive demos. The zoo will be the product". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2026 essay on continual learning, O'Neill wrote: "This is engineering, not research". — Continual learning and the post-monolith AI era, Charlie O'Neill (January 26, 2026).

In their own 2025 essay on machine-learning engineers, O'Neill wrote: "once you’ve validated the problem and see early signs of product-market fit, the game changes". — Your MLEs are brilliant, but you're giving them the wrong job, Charlie O'Neill (September 1, 2025).

In their own 2025 essay on machine-learning engineers, O'Neill wrote: "The hard part about LLMs is that optimising them requires understanding everything end-to-end". — Your MLEs are brilliant, but you're giving them the wrong job, Charlie O'Neill (September 1, 2025).

In their own 2025 essay on machine-learning engineers, O'Neill wrote: "We want to commoditise Big Lab talent so anyone can access it as a nicely packaged product". — Your MLEs are brilliant, but you're giving them the wrong job, Charlie O'Neill (September 1, 2025).

In their own X article on training in-house models, O'Neill wrote: "For the last 5 years, I have spent all day, every day, taking a general LLM and teaching it to do specific things". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "the reward signal for being good at coding inside Cursor lives inside Cursor, and nowhere else". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "Correctness is discovered through product iteration". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "The labs cannot follow, and the reason is organisational". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "Frontier labs are organised to serve one model to many customers. Specialisation requires the inverse". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "Every user session generates training data. Every training run produces a better model. Every better model attracts more users and more data". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "the only defensibility that survives the collapse of software cost is owning a model trained on signal no one else can see". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).

In their own X article on training in-house models, O'Neill wrote: "The question every app layer company is now asking is no longer ‘how do we use AI?’ It is ‘how do we resist commodification to deliver better results for customers?’". — Why every serious AI company is training its own model, Charlie O'Neill, X article (May 13).