Code is merged and deployed by agents in the middle of the night. The decision about *what* to build still takes three meetings and a Slack thread.

This is the decision-speed gap, and it's where quality goes to die.

We've spent the last two years optimising the execution layer. Agents write code, run tests, open PR's, deploy infrastructure. The tooling is genuinely impressive. But the organisational layer above it hasn't changed at all. Approval chains designed for humans moving at human speed now sit on top of systems that move at machine speed. The result isn't faster delivery. It's faster delivery of the wrong thing, or fast systems waiting on slow decisions.

I'll introduce the ISEE framework (Intent, Structure, Execution, Evidence) as a way to think about where agents actually fit in your decision architecture. Intent is the hardest part: who decides what matters, and how fast can that decision propagate? Structure is how you encode constraints so agents don't need to ask. Execution is the part we've already solved. Evidence is what closes the loop, the part almost everyone skips.

This isn't theory. ISEE is built into working software — Git-Ape, an open-source multi-agent platform that plans, validates, and deploys infrastructure through exactly these four stages. The framework came from watching where that system needed human judgment and where it didn't.

I'll show where most organisations are jammed (it's not execution), what actually needs to change (it's not the agents), and why the teams that get this right will outperform by a structural margin — not because their agents are better, but because their decisions are faster.

Key takeaways:
1. The bottleneck in most agent-enabled organisations isn't execution speed — it's decision speed. Optimising the wrong layer makes the gap worse, not better.
2. The ISEE framework (Intent, Structure, Execution, Evidence) gives teams a practical model for identifying where human judgment is essential and where it's just friction.
3. Organisations that redesign their decision architecture around agent capabilities will outperform those that bolt agents onto existing approval chains, and the gap will compound.