➤ This week's Crossing guest is Kevin Ding, founder and CEO of Pyromind. The space Pyromind operates in is often called RL as a Service — reinforcement learning as a service, or post-training as a service. The company recently completed an angel round, with investors including Hillhouse, Baidu Venture, BlueRun Ventures, Atypical, and others.
But in Kevin's view, RL as a Service is just the starting point, not the destination. He said on the show that Service only solves half the problem; to truly get Agents to continuously improve in production environments, you need to string training, rewards, feedback, deployment, and data flyback into an automatically looping pipeline — what Pyromind is betting on: AutoRL.
We also discussed Pyromind's real-world deployment scenarios: industrial quality inspection, process parameter optimization, GUI Agents, Coding Agents, and those production problems inside enterprises with clear ROI, sufficiently rich data, and relatively well-defined good/bad standards.
At the same time, this episode tackles a very practical question: how can enterprise AI avoid being project-based consulting?
Kevin was candid that on first entry into a customer scenario, FDE work is unavoidable: you need to understand data formats, evaluation baselines, business processes, and reward signals. But Pyromind's goal is to make this work gradually decrease across similar modalities and similar scenarios, turning one-time delivery into a horizontally replicable AutoRL pipeline.
Kevin shared in depth their new work, PyroDash: a large-small model collaborative architecture composed of a 4B small model, a strong base model, and a collaboration engine. Kevin believes this corresponds to a larger thesis: the world of needs is pluralistic, and not all problems should be collapsed into a single centralized large model.
This episode is for anyone who cares about AI Agents, post-training, enterprise intelligence, industrial AI, and AI infrastructure.
🎬 Our video podcast will be released simultaneously on Koji's WeChat Channels, Douyin, Xiaohongshu, Bilibili, YouTube, and other platforms.
📒 The transcript will be published on the Crossing WeChat official account.
00:23 Lightning round: age, alma mater, MBTI, zodiac sign, one-sentence intro, funding status, revenue and profit, team size, pre-founder experience, have you found PMF?
01:55 Why does the AI endgame look more like an Agent swarm than a single super foundation model ruling everything?
03:44 Why has RSI become important this year?
05:21 Which half of the problem does RL as a Service solve?
08:32 How does Pyromind select the customers best suited for AutoRL?
10:16 Why is FDE unavoidable on first entry into a scenario, and how does subsequent work decrease?
22:13 How can industrial customers calculate clear ROI?
23:22 What three criteria determine whether to take on an enterprise need or pass?
26:29 Why can't enterprise AI escape the impossible triangle of cost, efficiency, and privacy, and how does post-training crack it?
27:03 How does PyroDash let a 4B Worker Model collaborate with a Base Model, balancing effectiveness and cost?
29:23 Harness can improve process and speed, but what key enterprise needs does it still fail to solve?
34:02 How do two FDEs support more than a dozen B2B customers, and what work does Pyromind pull back into the product layer?
38:20 As models get better with training and training needs decrease, does Pyromind make more or less?
40:58 China SaaS is hard to build — why does Pyromind believe AI ToB can recalculate ROI?
42:06 Cloud providers can also do post-training and control compute — will Pyromind get swallowed?
47:47 Small model downloads still dominate on Hugging Face — why won't all needs collapse into one Base Model?
Subscribe to Crossing: 🚦 We track the industry shifts and new entrepreneurial opportunities brought by the new wave of AI technology.
🚦 "Crossing" was Steve Jobs's metaphor for Apple — standing at the intersection of technology and liberal arts, where great products are born. AI is transforming every industry. We seek out, interview, and bring together a new generation of AI founders and active builders in the AI era. Together with them, we explore and embrace the new changes, the new possibilities.
👦🏻 Host Koji: I founded Crossing, launched AI Hacker House — a community space for a new generation of AI founders — and serve as Venture Partner at ZhenFund. I believe technology, especially AI, is the greatest value creation opportunity of our generation. Koji on Jike, Koji's website
View this episode's transcript on Xiaoyuzhou