AI is solving the building problem and creating a knowing problem.
Teams can ship meaningful changes every few hours now. The people downstream of the code, product owners, leaders, salespeople, customers, are going to struggle to keep up with what the product actually does.
This is the next big scaling challenge with AI and nobody is talking about it.
Software development used to have a rhythm slow enough that downstream work fit inside it. Reviews, documentation, release comms, product owners staying current, all of it absorbed into the folds of a weeks-long process.
AI compressed the build phase to almost nothing. Nothing else compressed with it. All the work that used to hide in the folds is now exposed. The slack is gone.
Think about what happens when a team ships three meaningful changes in a day instead of a sprint.
Was each change built to spec, or did the AI drift from intent somewhere along the way? Does the new functionality solve the right problem, or just the most obvious one? Did any of it introduce patterns that conflict with how the product already works? Did it bypass architecture or security gates that exist for good reasons?
And outside engineering: do the people selling the product know what it does today? Are release notes, documentation, and customer communication keeping pace with what's actually shipping?
When development moved at sprint speed, these questions had time to get answered. Reviews happened. People stayed informed. The process had room for it. When development moves at AI speed, that room disappears. The questions don't go away. They just stop getting asked.
That's just one team. Multiply that across your entire engineering organization.
Review infrastructure has to scale with build infrastructure. Coding cycles compress to hours. Everything downstream still runs on weekly rhythms.
Automating release notes doesn't close the knowledge gap if nobody upstream understands what was built. The gates need humans who know the product, the architecture, the customer. People who can look at what shipped and answer the question that matters: should we have built this?
This isn't a new phenomenon. Organizations already experience it at quarterly scale. Lock in a quarter's worth of estimated work in a PI planning session, and two months later the teams don't know what's been done or what still needs doing. The organizational blind spot already exists, but has been easy to catch up when projects move at glacial speeds. AI just compresses the timeline from months to days, maybe even hours. The blind spot in the org gets magnified just as quickly as engineering velocity does.
And this is the problem at human speed. Teams are still writing the prompts, reviewing the output, deciding what ships. When agentic coding and automated feedback loops take over more of that process, the gap between what's being produced and what anyone understands about it gets wider, faster. We will hit a limit, either a cognitive limit or a comfort limit to how fast we can produce something without understanding it.
Organizations that recognize the knowing problem early will need more than faster reviews. They'll need a living source of truth about what the product does, current to the minute, accessible to everyone from engineering to sales. The rest will learn the hard way that nobody knows what the product does anymore.