The core message of the session is that Prosus is moving toward an AI-native organization because it can do more with the resources it already has, bring products to customers faster, and create a direct economic difference. The speaker rejects the idea that this shift is mainly about trend-following, arguing instead that AI is already capable enough for much more autonomous work and that “the limiting factor is actually us.” The talk frames AI adoption as a progression from chats, to simple agents, to agents doing more work, and says the organization is already operating close to the destination many people assume is still far away.

The speaker defines an AI-first organization through four characteristics. First, humans and agents collaborate inside the same teams, changing team size, roles, and organizational design. Second, “everybody’s a builder,” because domain experts can use AI tools to create powerful applications without depending only on software developers. Third, agents can connect information across systems and people’s knowledge, making the organization queryable across its many internal sources. Fourth, organizations can become self-improving through loops where technology continuously improves while interacting with customers, rather than being paused, upgraded, and redeployed in separate cycles.

The first concrete initiative is Project REST, which explores the limits of autonomy through agent-controlled vending machines. The agents know objectives, make orders, decide prices, determine whether customers get what they ordered, and compete with each other to learn what performs best. The speaker connects this experiment back to restaurants: restaurants should keep doing what they are great at, while Tokam can “do the rest.” The second initiative is the introduction of agent “night shifts,” where agents are launched in the evening with GPU power and a task. One example involved improving a production model that predicts which customers should receive discounts; over roughly fourteen hours, the agent reduced error more than colleagues could in that short time. The third initiative extends this into a software factory staffed only by agents, including planners, developers, verifiers, testers, and mergers, which can complete complex tasks over six to eight hours.

The closing section expands from internal use to ecosystem scale. After using agents with 40,000 employees, Prosus opened Tokam to 5,000,000 partners, and with Zapier is opening it to customers. The speaker says Europe, Latin America, and India can lead in AI and innovation rather than follow, and that Prosus believes it can contribute to that. The talk ends by thanking the Tokhan, Tokhan Clo, Zapier, iFood, OLX, PayU, Despegar, and broader ecosystem teams, followed by a small live moment showing Zapier and Taos Berserk bringing food, reinforcing the claim that the work functions in practice and at scale.