Google, Accel Pick Five AI Startups After Rejecting Wave of ‘Wrapper’ Ideas
Most of the 4,000 applications added thin AI features rather than redesigning enterprise workflows.
Topics
News
Artificial intelligence startups building deeper, workflow-level products dominated the latest cohort of a startup program run by Google and venture capital firm Accel, after investors rejected a wave of applications that were little more than superficial “AI wrappers” built on top of existing models.
Five startups, K-Dense, Dodge.ai, Persistence Labs, Zingroll and Level Plane, were selected for the program, which supports early-stage companies developing AI products linked to India.
The startups can receive up to $2 million in funding from Accel and Google’s AI Futures Fund, along with as much as $350,000 in cloud and computing credits, the companies said.
The cohort was chosen after reviewing more than 4,000 applications, but most were rejected because they added thin AI features to existing software rather than redesigning workflows around artificial intelligence.
“About 70% of the applications were wrappers,” Prayank Swaroop, partner at Accel, told TechCrunch. These startups typically added AI features such as chatbots to existing software without fundamentally redesigning workflows around AI, he said.
The surge in such startups reflects a broader trend across the global AI ecosystem. As large model developers rapidly add capabilities to their platforms, investors have grown cautious about startups whose products could quickly become redundant.
Many rejected applications fell into crowded categories such as marketing automation and AI hiring tools, areas where investors said differentiation has become increasingly difficult.
“These are sectors where a lot of companies are doing very similar things,” Swaroop said, noting that startups in these categories often struggle to build defensible technology.
The accelerator’s latest cohort attracted nearly four times the number of applications when compared with earlier Atoms programs. Much of the growth came from first-time founders experimenting with AI-driven products.
The applications also showed where India’s AI startup ecosystem is currently focused.
About 62% of submissions centred on productivity tools, while 13% focused on software development and coding, meaning roughly three-quarters of applications targeted enterprise software rather than consumer products.
The five startups chosen for the cohort reflect areas where investors believe AI could drive deeper real-world transformation.
K-Dense is building what it calls an AI “co-scientist” designed to accelerate scientific discovery in fields such as life sciences and chemistry. Founded in 2025 and based in Palo Alto, the platform aims to automate the entire research process, from hypothesis generation to publication, with minimal human intervention.
Dodge.ai is developing autonomous AI agents for enterprise Enterprise Resource Planning (ERP) systems, designed to help businesses automate complex workflows across finance, operations, and supply chain processes.
Persistence Labs, founded in 2012 in Oxford by Sean Heelan, focuses on voice AI for call-centre operations, helping companies automate customer interactions and improve agent productivity.
Zingroll, a Palo Alto-based startup founded in 2023, is building an AI-powered entertainment platform that generates personalized films and shows where users themselves appear as the main characters.
Level Plane applies AI to industrial automation, targeting manufacturing processes in sectors such as automotive and aerospace.
According to Jonathan Silber, co-founder and director of the AI Futures Fund at Google, the accelerator is designed to also help improve Google’s AI models.
The program does not require startups to exclusively use Google’s models, Silber said. Many companies combine multiple models depending on the workflow.
Instead, Google wants to observe how startups deploy AI systems in real-world environments and gather feedback on model performance.
Insights from these startups can then be fed back to teams at Google DeepMind, helping refine future models.
“If a company is using an alternative model, that means Google has work to do to build the best model in the market,” Silber said.
He described the process as creating a “flywheel” between startup experimentation and AI model development, one that could shape how next-generation AI systems are built.
