Manufacturing processes have evolved over time from handicrafts, where unique items are created on a one-off basis, to mass production, where standardised products are produced in large quantities at lower cost. Although the quality of handmade products varies according to the expertise of the craftsman and the variability in the raw materials, they are generally considered to be more precisely made than similar mass-produced items, due to the greater attention to detail by highly skilled workers. In the meantime, computer-aided methods have emerged. Numerical simulations provide a tool for improving product performance, with the ability to take into account various sources of uncertainty and develop robust designs accordingly. The result is an inevitable trade-off between performance and robustness that cannot usually compete with unit production and may not be sufficient for high performance applications. A new paradigm has first emerged with the concept of the digital twin which aims to improve the fidelity-to-data between simulations and reality. We are adapting this concept by proposing self-design manufacturing with the objective of combining mass production and custom design modifications in the same process. This idea is part of an emerging paradigm that we refer to as "physical intelligence" which is complementary to the computational intelligence underlying the virtual prototyping approach used today. To demonstrate this new approach, an additive manufacturing process is coupled with a decision algorithm and online vibration instrumentation to design a high-performance product. Experimental results from the design and fabrication in a conventional 3D printer of a beam-shaped vibration absorber - a passive control device highly sensitive to detuning - inserted in a plate will illustrate the potential of the proposed strategy for high-precision manufacturing requiring optimal efficiency.