The Lab Mindset: Why Speed is the Only Strategy in the AI Era
Moving from rigid product cycles to experimental loops in a world of shifting models
The era of the five-year product roadmap is dead. In the time it takes a traditional corporate committee to approve a design document, a new model release has already rendered the underlying assumptions obsolete. For companies trying to build in the age of generative AI, the old way of working—defined by long cycles of planning, development, and deployment—is a recipe for irrelevance. The new requirement is not better planning, but better experimentation. This is the distinction between a manufacturing company and a laboratory. A factory seeks to minimise variance; a lab seeks to find it. If you are not running dozens of small, cheap experiments every week, you are not building for the future; you are merely managing the decline of your current relevance.
The 'Almost Possible' Framework
Josh Woodward, who leads Google Labs, operates on a principle that should be standard for any agency or tech firm: looking for the 'almost possible'. This is the space where a technology is just functional enough to be interesting, but not yet polished enough to be a product. Most companies avoid this space because it is messy, unquantifiable, and full of failure. But this is exactly where the next decade of value will be captured. In the AI era, the goal is to find these edges—the moments where a model can suddenly perform a task that felt impossible six months ago—and to build around them before the capability becomes a commodity. This requires a culture that prizes the ability to spot a signal in the noise of rapid technical shifts.
The goal is not to build a perfect product, but to build the capacity to find the right product through rapid, repeated failure.
This shift requires a fundamental change in how we value roles. We used to value the specialist who could execute a complex plan with precision. Now, we value the generalist who can navigate ambiguity and the researcher who can turn a technical breakthrough into a user experience. The lines between engineer, product manager, and designer are blurring. The most successful teams are those that function as small, autonomous units capable of seeing a new capability in a model and immediately prototyping a use case for it. They don't wait for permission; they wait for data.
- Lower the cost of failure by prototyping with existing models rather than building from scratch.
- Set strict 'kill criteria' for experiments to avoid the sunk cost fallacy.
- Prioritise speed of learning over speed of shipping.
- Reward the identification of 'almost possible' capabilities, even if they don't lead to a product.
The hardest part of this transition is knowing when to quit. In a traditional business, a project that fails to meet its KPIs is a disappointment. In a lab, a project that fails to meet its KPIs is a successful experiment if it provides clear data on why it failed. The danger is the 'zombie project'—the idea that is kept alive because of the ego of its creator or the budget already allocated to it. A lab team survives by being ruthless about killing ideas that don't show early signs of product-market fit. In the AI race, the winners won't be those who build the most features, but those who learn the fastest.
In a high-velocity environment, your competitive advantage is your rate of learning, not your ability to execute a plan.