Klarna saved $60 million and broke its company. The missing layer is what I'm calling intent engineering + 2 prompts to find yours

Klarna’s Q3 2025 earnings revealed its AI agent now does the work of 853 full-time employees and has saved the company $60 million. But six months earlier, its CEO had already admitted publicly that the AI strategy had cost Klarna something far more valuable — and he’s still trying to buy it back.

This is not another “AI was overhyped” story. It’s the opposite. The AI worked too well. And that distinction — between AI that fails and AI that succeeds at the wrong thing — is the most important unsolved problem in enterprise AI right now. It’s bigger than context engineering, though it includes it. It’s bigger than prompt engineering, which now looks like a warm-up act. The right name for what’s missing is intent engineering: the discipline of making organizational purpose — goals, values, tradeoffs, decision boundaries — machine-readable and machine-actionable so that when you deploy an autonomous system, it optimizes for what your company actually needs, not just what it can measure.

The disorienting part is worth sitting with. MIT reports 95% of generative AI pilots fail to deliver measurable impact. Gartner predicts 40% of agentic AI projects will be cancelled by 2027. And yet the money has never flowed faster — Deloitte found the majority of enterprises putting 21–50% of their digital transformation budgets into AI automation. These numbers don’t contradict each other. They describe organizations that have solved “can AI do this task?” and completely failed to solve “can AI do this task in a way that serves our organizational goals?”

That’s the intent gap. Klarna is the clearest case study. Its AI agent handled 2.3 million conversations in its first month and cut resolution times from eleven minutes to two. Then the CEO went on Bloomberg to explain why the strategy had backfired — and started hiring humans back. The full story is more nuanced than the headline, and it illustrates exactly why “the AI worked” and “the AI failed” can both be true at the same time.

The progression matters: prompt engineering told AI what to do in a single session. Context engineering — where the industry is right now — tells AI what to know. Intent engineering tells AI what to want. And almost nobody is building for it yet.

Here’s what’s inside:

• The three disciplines. Prompt engineering told AI what to do. Context engineering tells AI what to know. Intent engineering tells AI what to want — and almost nobody is building for it yet.

• The Klarna post-mortem and the Copilot stall. How $60 million in AI savings destroyed the thing that actually made customers stay — and why 90% Fortune 500 Copilot “adoption” producing just 3.3% paid uptake is the same failure at different speed.

• Three missing layers. Unified context infrastructure, coherent AI workflow architecture, and organizational alignment frameworks — what each one prevents, and what it costs to skip.

• Why this is fractal. The same intent failure breaking enterprise AI deployments is breaking your personal AI workflow right now. The fix is the same architecture at a different scale.

• The prompts. An intent audit and a delegation framework you can run this week to find out whether your agents actually know what you’re optimizing for.

I want to show you how this happened — and why the fix is architectural, not inspirational.