A product idea used to prove itself through documents. A business case, a requirements spec, a six-page narrative. Stakeholders read, imagined what the product might look like and decided whether to fund it.
AI changed the economics. A working prototype that once took weeks now takes hours. Stakeholders no longer need to imagine. They see, react and decide.
The constraint did not disappear. It moved.
Product was a telling game. AI made it a showing game
The cost of proving an idea has collapsed. A team can put a working prototype in front of a stakeholder before anyone writes the business case. The feedback loop that used to take months now takes days. Ideas that would have died in a document get tested with real users before any production investment.
This is a genuine gain. But it comes with a problem.
When proof is cheap, more ideas get proved, and more proposals land on the desk with a working demo attached. The gate was not designed for this volume.
The constraint moved. The organisation has not
Before
Building was the bottleneck
After
Evaluation is the bottleneck
Andrew Ng said it plainly on the No Priors podcast in 2025: 'The bottleneck is deciding what do we actually want to build.' In Atlassian's State of Product 2027 survey of 1,000 product professionals, 86% agreed that deciding what to build matters more as AI makes shipping easier.
Recognising that the constraint has moved is not the same as redesigning for it. Most organisations have not.
The investment gate in most organisations was designed for a world where building was expensive and only a handful of ideas made it through. It is not designed for a world where every idea arrives with a working demo. In the same Atlassian survey, 69% said effort had gone into shipping faster while decision-making had not sped up, even with AI.
Music went through this shift years ago. Digital production tools and streaming platforms reduced the cost of releasing a song to near zero. Over 100,000 new tracks are uploaded to streaming platforms every day.
The bottleneck moved from the recording studio to the playlist editor. The scarce resource moved from production to curation.
The same pattern applies to product. The scarce resource is no longer the capacity to build a prototype. It is the capacity to evaluate one, decide whether it earns a place on the roadmap and kill it if it does not.
Build the muscle to say no
The organisations that gain from cheap proof are those that can evaluate and kill faster, not those that say yes to more.
I saw this play out first-hand. When my team doubled A/B experiment velocity year on year, the bottleneck moved. We could prove more, faster.
But that only mattered because we also changed how we made decisions. We shifted from opinion-based to evidence-based decision-making, built a plug-and-play experimentation toolkit and invested in the infrastructure to evaluate results, not only generate them.
The hard part was not running more experiments. It was building the discipline to act on what the experiments told us. That meant killing ideas the data did not support, even when a stakeholder had already seen a working prototype and liked it.
Saying no to a document is straightforward. Saying no to a working demo that a senior leader has already reacted to is a different conversation entirely. It is also a common one. In Atlassian's survey, 77% said executive gut instinct overrides decisions their team has based on data, at least sometimes.
That is where product judgement earns its keep.
Cheap proof does not reduce the need for product judgement. It increases it.
The first article in this series argued that AI amplifies whatever platform it is plugged into. This article asks what happens when the amplifier makes proof cheap. The answer: the bottleneck moves from building to deciding. The amplifier is only as good as the decisions it serves.
The ability to prioritise, evaluate and say no is now the most constrained capability in the stack. Organisations that treat it as a discipline and invest in it will get more from their AI spend than those that just build more prototypes.
A follow-on article will zoom out from product to the whole organisation and ask why AI budgets keep growing while the value they return does not.