Many artificial intelligence startup ideas remain primarily superficial “wrappers” built on existing models. As AI model creators add more features, investors are cautious about startups that could easily become redundant.
When reviewing over 4,000 applications for the AI accelerator for Indian startups run by Google and venture firm Accel, “wrapper” ideas were prevalent. However, none were among the five startups selected for the latest cohort, Accel partner Prayank Swaroop shared with TechCrunch.
Announced in November, the AI-focused Atoms program by Google and Accel aims to support early-stage startups creating AI products related to India. Selected startups for the newest 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, according to the firms.
About 70% of the rejected applications were “wrappers” — startups adding AI features like chatbots to existing software but not reimagining new workflows using AI, Swaroop said.
Many other denied applications fell into crowded categories such as marketing automation and AI recruitment tools, which investors found to lack novelty. Startups in these areas often find it difficult to stand out, he said.
This is likely not surprising, as this year’s program saw almost four times the number of applications compared to previous Accel Atoms cohorts, with many first-time founders.
India’s expanding AI ecosystem remains largely focused on enterprise applications, which Swaroop said was evident in the applications. About 62% of the submissions were geared towards productivity tools, and 13% on software development and coding, meaning roughly three-quarters of the applications were enterprise-focused rather than consumer products. Swaroop had hoped for more ideas in healthcare and education.
Jonathan Silber, co-founder and director of Google’s AI Futures Fund, mentioned the five startups chosen align with areas where Google anticipates broader real-world AI adoption.
The program does not mandate startups to exclusively use Google’s models, Silber noted, as many companies use multiple models per workflow. The aim is to gather startup feedback on Google’s model performance in practical applications.
Insights from these startups are fed back to Google DeepMind teams to 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 selected startups are:
– [K-Dense](https://www.k-dense.ai/), developing an AI “co-scientist” to accelerate research in fields such as life sciences and chemistry.
– [Dodge.ai](https://getdodge.ai/), creating autonomous agents for enterprise ERP systems.
– [Persistence Labs](https://persistence.dev/), focusing on voice AI for call center operations.
– [Zingroll](https://zingroll.com/), building a platform for AI-generated films and shows.
– [Level Plane](https://www.levelplane.com/), applying AI to industrial automation in automotive and aerospace manufacturing.