In his latest video covering the Claude Opus 4.6 announcement, Nate B. Jones describes how the bottleneck is shifting to human clarity of intent and judgment. I agree, and this means the core of product management, deciding what to build to succeed in the marketplace, is about to get a lot more important.
Anthropic said Claude produced a roughly 100,000-line C compiler using 16 agents across nearly 2,000 Claude Code sessions. While currently in beta and limited to the Claude Developer Platform, Opus now also supports a 1M-token context window, a roughly 5x jump in context size from Opus 4.5, making it more feasible for AI to carry far more project context as it works, not just isolated files.
These leaps in AI capability to handle large projects preview a near future where writing code is less of the bottleneck. In such a future, deciding what product will succeed in the market and what features will differentiate that product from the competition becomes an even more central challenge. And as it's easier to add features, it will be tempting to create feature bloat. Product managers will need to use great care to decide what features not to build to keep products simple to understand and use.
The Core of Great Product Management
If AI automates requirements drafts, backlog grooming, baseline competitive research, and synthesizing customer feedback, the remaining product management work becomes problem framing, decision-making, and go-to-market alignment. This is about answering what we build first, what we do not build at all, and how we define success.
Can AI automate this too? Never say never with AI. It can access field notes, but to dig deeply, someone needs to sit down with customers and ask the right questions to comprehend their problems and the competition's weaknesses relative to those problems. Ultimately someone has to be accountable for setting the strategy and making risk tolerance tradeoffs over the long haul.
This isn't just about ideation vs. execution. Systematically gathering data from the marketplace and defining and iterating the product vision based on that data IS execution. But it's execution plus judgment about what will be successful in that marketplace at its core.
The 1% Advantage
So back to Nate B. Jones's point: The career differentiator in the coming AI age will be judgment, which comes from deep experience. Having 1% better "taste" than everyone else in some particular niche, be it the personal stylist who creates 1% better fashion, the chef who creates 1% better recipes, or the product manager who picks a product strategy that customers find 1% more useful than competitors' products.
What Comes Next: The AI Supervisor
A related challenge will appear as we create workflows where AI monitors usage patterns, evaluates bottlenecks, proposes and prioritizes improvements, and even implements them — all without human involvement.
What is the product manager's role in such a world? I'm thinking it's to be the AI supervisor and coach who sets the guidelines under which it does this so the outcomes are positive for users. And the challenge becomes even more acute: not bewildering users, and keeping the product from becoming ridiculously complex with features and options.
Consider how fast the analytics → product improvement cycle could become. Today it takes weeks or months from seeing a usage pattern to shipping a fix. When AI can close that loop autonomously — observing a drop-off, hypothesizing a cause, generating a fix, A/B testing it, and shipping the winner — the gating factor shifts from engineering bandwidth to how fast customers can absorb change. The product manager's role becomes less about deciding what to build and more about governing the rate and direction of continuous product evolution.
