Many artificial intelligence startup ideas are still little more than superficial “wrappers” built on top of existing models. But as the AI model makers add more features, investors are wary of startups that could become so easily unnecessary.

Case in point: when reviewing more 4,000 applications for the joint AI accelerator for India startups run by Google and venture firm Accel, “wrapper” ideas dominated. But none of them were among the five startups for the latest cohort, Accel partner Prayank Swaroop told TechCrunch (pictured above).

Announced in November, the AI-focused Atoms program by Google and Accel aims to back early-stage startups building AI products linked to India. Startups selected for the latest cohort will receive up to $2 million in funding from Accel and Google’s AI Futures Fund, along with up to $350,000 in cloud and AI compute credits from Google, the firms said.

Roughly 70% of the rejected applications were “wrappers” — startups that layered AI features such as chatbots on top of existing software but “were not reimagining new workflows using AI,” Swaroop said.

Many of the remaining applications that were denied, Swaroop said, fell into crowded categories such as marketing automation and AI recruitment tools, areas where investors saw little novelty. Startups in those sectors often struggle to differentiate themselves, he said.

This isn’t, perhaps, surprising. This year’s program received nearly four times the applications than previous Accel’s Atoms cohorts — with many first-time founders.

India’s growing AI ecosystem remains largely focused on enterprise applications and Swaroop said the applications reflected that. About 62% of the submissions focused on productivity tools and another 13% on software development and coding, meaning around three-quarters of the applications were enterprise software ideas rather than consumer products. (Swaroop had hoped to see more ideas for healthcare and education.)

Jonathan Silber, co-founder and director of Google’s AI Futures Fund, said the five startups selected aligned closely with areas where Google expects AI to see deeper real-world adoption.

The program does not require startups to use Google’s models exclusively, Silber said, noting that many companies combine multiple models depending on the workflow. The goal, he said, is to gather feedback from startups on how Google’s models perform in real-world applications.

Insights from those startups can then be fed back to Google DeepMind teams to help improve future models, creating what Silber described as a “flywheel” between startup experimentation and AI development. “If a company is using an alternative model, that means Google has work to do to build the best model in the market,” he told TechCrunch.

This year’s startups selected are:

• K-Dense, which is building an AI “co-scientist” to accelerate research in fields such as life sciences and chemistry;
• Dodge.ai, which develops autonomous agents for enterprise ERP systems;
• Persistence Labs, which focuses on voice AI for call centre operations;
• Zingroll, which is building a platform for AI-generated films and shows;
• Level Plane, which applies AI to industrial automation in automotive and aerospace manufacturing.