While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.
It is not scaremongering in my opinion. Just that Government needed some time to understand and will do the same for any other company with such a model.
1/ Jailbreak => Rapid catchup of the industry leading to commoditization
2/ Jailbreak => 99% of internet infrastructure gets exposed to cyberattacks at a scale the world is simply not ready. Maybe <1% of internet users are using Fable, out of which <1% will use it for beyond intended use. Put yourself in the shoes of someone maintaining critical infrastructure, or millions of people working 7 days a week to run a small business. The world needs some time to adapt.
This will invariably be a problem in organizations where tokens, lines of code, PR count etc are the metrics - which happens to be in most places. I do not know if there are metrics or rewards for maintainable code, OR penalty for write code that breaks down and causes product incidents down the line. By then those engineers would have been promoted and moved on to better things.
Very well thought and written. Provisional employee or intern. Or having the candidate to come and do real work for a couple of days. The challenge imo is the big company culture vs startups. Do the things move at the pace in big companies where the teams have the ability to evaluate? Startups are a different beast however.
Is it possible that you are narrowly sizing the opportunity? While PMF does not always mean that early pioneers will be the leaders, I think the market itself goes beyond knowledge workers and developers. Agents, robots, drones etc will all use LLM or some world model.
I am rather more concerned about competition from CHINA. With how Huawei (2000 -> 2020) crushed every other telecom company and went from nobody to the most revered leader in 20 years, and with the depth of leadership in manufacturing and work culture, if China surpasses USA in AI, all US companies lose.
Both Cloudflare and Vercel have feature parity. Flags is a feature already in Vercel. While customer-first is a thing, it is also a no-brainer to start with: we use it, Vercel has it, let us build it.
I have been in this situation. A major driving force is some kind of a demand from the leadership to see the KPI for the AI adoption. And this unfortunately is the easiest one to implement.
The other aspect is virality. I think by now the implementing team should know that most people do not appreciate Claud inserting itself into the commit message. It's the job of the team to feed that to the leadership.
While Anthopic has the best model and a focussed (no disturbance, lawsuits) leadership, they got a lot of enterprise access due to AWS. It is mutual no doubt, with both sides benefitting. The culture of feedback loop of AWS customers would have helped them in getting to enterprise-ready faster. Just my hypothesis.
It is a net positive to have a technologist and hardware leader at the helm. In this era, Apple can hire the right people to build software faster. but they need a strong hardware leader at the helm to differentiate themselves. In local AI, they have a unique opportunity, but limited window of time.
Think another way, these product features are easy to build in other harnesses too. And as the open source models and the other models which are much lower cost are getting better, there will be a time when it will be justified to have a harness that can work with many models and optimize your cost and efficiency.
I think it depends on the company. In large companies, the role of PM probably won’t change that much. However, PMs who are technical and hands-on can bring significantly more value by leveraging AI tools.
There’s another path for PMs that the article and most of the comments don’t seem to mention.
Technical PMs are now in a great position to start their own companies. In the past, many were blocked or handicapped by the inability to code. With AI-assisted development, that barrier is much lower, which gives them a lot more leverage to build products themselves.
Lot of skepticism about OpenAI's survival. I am a user of both Claude (max) and Codex (plus). Some neutral points.
- Anthropic owes to AWS for their enterprise growth. Yes, their own talent as well.
- AWS investing for a purpose - solving problems with multi-agent systems - "exclusive third-party cloud distribution provider for OpenAI Frontier, which enables organizations to build, deploy, and manage teams of AI agents.". I think the multi-agent landscape will be production-ready in 2026 for solving really complex problems. AWS saw something in Codex and OpenAI's models.
- On Circular investments - if you make $100B of your revenue from ecosystem of players who spend $50B on your infra... where else would you go?
If Codex 6.0 is better than Opus 4.9, things will flip. While OpenAI has too many common enemies and trying to box them into a consumer company, they are equally enterprise focused. They need to absolutely do well with foundation model - everything else depends on that.
While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.