Agriculture is fundamental for human survival through food production and is performed in ecosystems that, while simplified, still operate along ecological principles and retain complexity. Agricultural plants are thus part of ecological systems, and interact in complex ways with the surrounding terrestrial, soil, and aquatic habitats. We discuss three case studies that demonstrate how agricultural solutions to pest and weed control, if they overlook important ecological and evolutionary factors, cause "surprises": (i) the fast emergence of resistance against the crop-inserted Bt-toxin in South Africa, (ii) the ecological changes generated by Bt-cotton landscapes in China, and (iii) the decline of the monarch butterfly, Danaus plexippus, in North America. The recognition that we work with complex systems is in itself important, as it should limit the belief in reductionist solutions. Agricultural practices lacking eco-evolutionary understanding result in "surprises" like resistance evolution both in weeds and pest insects, risking the reappearance of the "pesticide treadmill"-with increased use of toxic pesticides as the follow-up. We recommend prioritization of research that counteracts the tendencies of reductionist approaches. These may be beneficial on a short term, but with trade-off costs on a medium-to long-term. Such costs include loss of biodiversity, ecosystem services, long-term soil productivity, pollution, and reduced food quality.