It has almost always been requirements or rather lack of clarity / not understanding actual business needs, that has caused projects to run into trouble, very rarely have I seen a technical implementation lead to a project failing.
I've worked on projects with Startup Founders, and also experienced Product Owners in established corporates, who were unable to express the business needs and context - a dev team can't build what the business cannot describe.
AI can help with this.
I've been building a framework for my own development, using Claude, Claude Code, using Agents, Skills etc. One of the most valuable aspects of this is an Agent + Skills to capture and document requirements, asking questions in a kind of 'Pair Programming for Product Owners' approach, that at each phase presents information gathered and prompts for validation, round-tripping until business context, scope and feature descriptions are as close to complete as possible, then creates and updates Epics and User Stories in Jira.
Jira is then used as specifications for development using TDD in a workflow in Claude Code with Compound Engineering.
The Product Owner Agent can be run multiple times during development, reading and understanding progress in Jira as the system is developed.