Two prominent CEOs leading multi-billion-dollar AI companies are challenging the prevailing narrative about AI’s ability to quickly automate workplace tasks. Glean CEO Arvind Jain and Databricks CEO Ali Ghodsi shared candid insights on the “Bg2 Pod” podcast, revealing that deploying AI effectively is far more complex than many business leaders assume.
Jain, whose AI startup Glean raised $150 million at a $7.2 billion valuation in September, described his own failed attempts to automate internal workflows. He attempted to use AI to automatically identify and document employees’ weekly priorities for leadership review. Despite Glean having “all the context inside the company to make it happen,” the seemingly simple automation project hasn’t worked. The company also invested in building and fine-tuning a custom model for a specific use case, but that effort “didn’t really pan out,” forcing Glean to pivot back to existing foundation models that proved easier to deploy.
Ghodsi, whose company Databricks recently raised over $4 billion at a $134 billion valuation, emphasized that successful AI implementation isn’t as simple as unleashing agents and expecting immediate results. He characterized making AI useful within organizations as “an engineering art” that requires careful evaluation, production work, and strong technical teams to support it.
Both executives reframed the conversation around AI project failure rates. Jain noted that the commonly cited statistic that “95% of projects fail” is actually a positive indicator. “When you’re actually experimenting with new technology, if all of your projects are failing, that means you’re not trying enough,” he explained. This perspective suggests that high failure rates indicate healthy experimentation rather than fundamental problems.
The human element remains critical in AI deployment, according to both leaders. Ghodsi previously stated at a June conference in San Francisco that even as AI agents become more prevalent, “there will be a human overseeing and approving every step.” He predicted that workers will increasingly become supervisors of AI systems rather than being replaced by them.
This view aligns with perspectives from AI research pioneers like Yoshua Bengio, one of the “godfathers of AI,” who recently emphasized that human qualities will become more valuable as machines automate routine tasks. “The human touch is going to take more and more value, as the other skills become more and more automated,” Bengio said on “The Diary of a CEO” podcast.
Key Quotes
It has all the context inside the company to make it happen. I thought AI would ‘magically’ do the work.
Glean CEO Arvind Jain describing his failed attempt to automate employee priority documentation, revealing that even AI companies with comprehensive internal data struggle to achieve simple automation goals.
It’s not just you can just unleash the agents, and it just works.
Databricks CEO Ali Ghodsi cautioning against oversimplified expectations for AI deployment, emphasizing that successful implementation requires sophisticated engineering and careful oversight.
You hear these 95% of projects fail. That’s actually what you want. When you’re actually experimenting with new technology, if all of your projects are failing, that means you’re not trying enough.
Arvind Jain reframing high AI project failure rates as a positive indicator of healthy experimentation rather than a sign of fundamental problems with the technology.
I think in a few years, yes, we’ll have agents in many, many places, but there will be a human overseeing and approving every step, and you’re on the hook when you approve, when you click, ‘OK.’ We all become supervisors.
Ali Ghodsi outlining his vision for the future of work with AI, where humans transition into supervisory roles rather than being replaced entirely by automation.
Our Take
The willingness of these AI CEOs to publicly acknowledge their automation failures is refreshingly honest in an industry often characterized by inflated promises. Their experiences reveal a critical gap between AI’s theoretical capabilities and practical implementation challenges. What’s particularly striking is that these failures occurred within AI-native companies with access to cutting-edge technology and top talent—suggesting that traditional enterprises will face even steeper challenges. The pivot from custom models back to foundation models at Glean also signals an important trend: the economics and practicality of pre-trained models may outweigh the benefits of customization for most use cases. This reality check could actually benefit the AI industry long-term by setting more realistic expectations and encouraging sustainable investment strategies rather than boom-bust cycles driven by unrealistic hype.
Why This Matters
This candid assessment from two CEOs running companies valued at over $140 billion combined represents a significant reality check for the AI industry and businesses investing heavily in automation. Their experiences challenge the hype cycle surrounding AI’s immediate transformative potential and provide crucial guidance for organizations planning AI deployments.
The admission that even AI-native companies struggle to automate their own workflows has profound implications for enterprise AI adoption. It suggests that businesses should adjust their timelines and expectations for AI ROI, focusing on incremental improvements rather than revolutionary transformation. The emphasis on human oversight also reshapes the conversation around AI and employment—rather than wholesale job replacement, the future may involve humans supervising and directing AI systems.
For investors and stakeholders, this transparency about failure rates and implementation challenges provides a more realistic framework for evaluating AI projects. The message that 95% failure rates are normal and even desirable during experimentation phases could help companies avoid premature abandonment of AI initiatives. This perspective shift from “AI will automate everything quickly” to “AI requires careful engineering and human collaboration” may ultimately lead to more sustainable and successful AI implementations across industries.