Waiting for the next model is the strategy

I think the most expensive mistake in enterprise AI right now is building too fast. John Stecher, CTO of Blackstone, and Scott Worland, CTO of AI Engineer, said they now explicitly price a six-month wait into roadmaps because the next model often makes a hard build trivial.

That is not patience for its own sake. It is math on cost and capability curves that both men have watched bend hard in the last year.

Norm AI learned it the expensive way. Worland said the company, founded in 2023, spent early 2024 adding structure to make models usable in regulatory work, sequencing them, checking consistency, distilling outputs into what looked like a large map-reduce system with agent scaffolding. It worked, and it was backed by tens of thousands of hand-labeled evaluations from experts. Then agentic approaches arrived and much of that code was obsolete, Worland said, which is why the engineering leadership rented a Sprinter van after a snowstorm to upstate New York to design a platform from scratch around agents as co-developers.

Stecher, who runs all technology at Blackstone and advises its portfolio on AI, made the parallel case from the buy side. He said the firm wins on agents only when the data story is right, with a cohesive strategy to move data from transactional systems into something agents can actually use, and then it optimizes for straight-through processing and injects intelligence where humans did rote decisions. "Having an agent be able to go off and do these things for me and renovate my code and my source code has made all the difference in the world for the velocity that we deliver at," Stecher said. He pushes teams to chase proof of value, not slide decks, and to stop counting tokens. "Leverage tokens, which like I think it's a fallacy to be like counting tokens in some way, shape, or form. It's better to think about producing value," he said. In financial services, he added, the accuracy bar is unforgiving. "If you're trafficking in financial services, like the duty to be accurate is like up here, and hitting that is like something like we think about all the time."

The best case against waiting is that you learn nothing while you wait. Both teams concede it implicitly, which is why they invest heavily where wait-and-see fails. Norm AI keeps human approval gates around its harness, sandbox and egress primitives because client data is highly sensitive, and both firms treat evaluation sets as the durable asset that survives a model swap. Without that plumbing and those scorecards, a new model has nowhere to plug in.

I think their claim survives that concession because they are not arguing for inertia. They are arguing for conduits, not dams, as Worland put it. That distinction matters against the funding context that still rewards pure legal AI platforms. Norm AI raised $120 million in a Series C at a $1.2 billion valuation led by Khosla Ventures, with Blackstone, Bain Capital Ventures and Craft Ventures participating, a deal that shows investors still pay up for explainable compliance automation rather than thin wrappers around a frontier model.

If you asked me what to build now, I would say build your evals, your data plumbing and your human approvals, and deliberately delay the clever orchestration you think you need today.