With its Wishlist Summer 2025, the Californian accelerator reveals a structured vision of the future of startups in the era of artificial intelligence.
Each season, Combinator publishes a list of ideas that he wishes to see emerging in his program. That of summer 2025 exceeds the exercise of style. It constitutes a strategic orientation document for the founders, investors and operators of the ecosystem. The whole is dominated by an ambition to reconstruct, from bricks of artificial intelligence, complete systems (professions, organizations, products) capable of directly competing the existing models.
The rise of IA agents
Several proposals highlight a paradigm shift in software design. These are no longer tools that combinator wishes to see emerging, but agents, capable of acting independently instead of the user. The processing of emails, the management of an agenda, the telephone response or the organization of administrative tasks would no longer fall under the user or a human assistant, but of an agent with contextual memory and a capacity for action.
Tom Blomfield, former founder of Monzo, imagines a personal assistant who does not organize tasks, but accomplishes them. Pete Koomen, co -founder of Optimizly, evokes a structural trend: in each company, each employee will soon be able to create his own agent to automate recurring tasks. These scenarios are today technically possible by the recent progress of language models.
Verticized startups from the outset
One of the most striking proposals comes from Jared Friedman, a partner of Y combinator, rather than creating a tool for a traditional sector, with for example an agent for legal firms. And suddenly, why not directly create a new structure that integrates these agents with its value chain? A form of radical verticalization where the startup does not sell a product to an actor in place, but itself becomes this actor, reinvented by AI.
This “Full-Stack” model implies a more capitalist approach, more complex to operate, but potentially more defensible in the long term. It is part of a desire to break with the classic B2B sales cycles and to reduce dependence on large companies as first customers.
The priority given to verticals with strong inertia
Y Combinator emphasizes several sectors considered ripe for a deep transformation via AI. This is the case of health, where Gustaf Alströmer points out the colossal weight of administrative costs. This is also the case for personal finance, where human biases and the cost of individual advice give way to automated solutions.
Education is another key sector, considered difficult to transform but essential in the long term. Harj Taggar and Tom Blomfield evoke the possibility of creating personalized tutors capable of adapting to the rhythms and learning methods of each student, through multimodal interfaces integrating text, image, audio and animation.
Scientific research, finally, is perceived as a strategic territory. Diana Hu calls for designing tools to accelerate modeling, discovery and experimentation in complex fields such as chemistry, material physics or industrial optimization.
Robotics on the eve of its “GPT moment”
Unlike other sectors, robotics has not yet experienced a rupture linked to generative AI models. But for Diana Hu, this step is close. The trigger would be the integration of perception and decision models sufficiently efficient to allow robots to operate in non -marked real environments.
Rather than creating robots themselves, YC is interested in software tools that allow others to design them more easily. This indirect approach could promote the emergence of new technical standards, in particular for industrial or agricultural uses.
A call to a new generation of founders
Finally, several publications in the program express a desire to expand the profile of the founders. Aaron Epstein highlights designers, perceived as endowed with an ability to translate a coherent user experience product intention, without which even advanced technology can remain ineffective.
Jared Friedman revives, for his part, a call for the creation of independent research laboratories. He recalls that Openai was born in YC Research and affirms that the program can also support long scientific projects, without imperative of immediate marketing. The notion of “startup” thus extends to exploration structures, capable of laying the technological bases of the next waves.
