How to avoid AI’s ‘valley of despair’

AI can make one step of a job blazingly fast and still leave the whole process crawling. That’s the core argument behind the latest warning about AI’s “valley of despair”: companies have sped up isolated tasks without rebuilding the approvals, reviews and handoffs around them.

A proposal can be drafted in seconds, but then it may sit for days waiting on pricing and sign-off. Code can be produced in minutes, but testing, security and deployment can still stretch into days or weeks. The result is familiar to anyone who has watched software promise to transform work while the business process quietly refuses to move.

The numbers are sobering. Forrester Research found that only 15% of AI decision-makers said earnings before interest, taxes, depreciation and amortization rose over the past 12 months. Fewer than one-third could clearly connect AI spending to bottom-line change. That doesn’t make skepticism cynical; it makes it normal. AI should be treated like any other capital expense, with expected returns, deadlines and milestones attached.

The article’s answer is not more enthusiasm, but process redesign. Companies need to stop asking only how to automate existing work faster and start asking whether some of that work should exist at all. Useful targets are specific business outcomes: less churn, a shorter lead-to-proposal cycle, better forecast accuracy or faster fraud detection.

That also means measuring the right things. Leaders should expect usable ROI data within about six months, while real redesign work can take 90 to 120 days before testing and change management even begin. And they need to watch costs at the case level, because agentic systems can vary wildly depending on the use case, the number of model calls and the amount of context involved.