Unprompted Intelligence in High-Stakes AI with Hebbia CTO Aabhas Sharma

In this episode of the Shift AI Podcast, Aabhas Sharma, CTO at Hebbia, joins host Boaz Ashkenazy for a deep conversation on what it actually takes to deploy AI in high-stakes institutional environments — and why the bar is far higher than most tools can meet. Aabhas shares his path from running engineering at Postmates through its Uber acquisition, to a five-year stint as CTO at fintech startup Found, to his current role at Hebbia which began as a year-long advisory engagement before he joined formally. From there, the conversation dives into the core problem Hebbia solves: knowledge workers at the world's top investment banks, asset managers, and private equity firms buried in thousands of documents, needing multi-step analysis that existing tools simply can't handle. The discussion covers how Hebbia thinks about the interplay between deterministic software and large language models using LLMs only where generative reasoning is truly needed and keeping everything else rules-based, auditable, and fast. Aabhas explains the company's "forward deployed banker" model, where former investment bankers and transaction lawyers embed with customers to encode each firm's unique institutional methodology into Hebbia's platform. Boaz and Aabhas also explore what trust actually means in practice in regulated industries: explainability at the clause level, consistency across runs, and deliberate human checkpoints built into every workflow. They dig into the adoption dynamics at junior versus senior levels, the governance challenges that come with agentic systems, and why earning autonomy must be incremental in high-stakes environments. The episode closes with a striking observation: firms layering AI on top of existing processes are just moving bottlenecks. The firms that will pull ahead are the ones willing to redesign the factory floor, asking what it would look like to build the organization from scratch with AI today. This episode is essential listening for enterprise technology leaders, founders building in regulated verticals, and anyone thinking seriously about the gap between individual AI productivity and firm-level outcomes. --- Chapters [00:00] Aabhas's Path to CTO at Hebbia [02:01] First Jobs and Early Tech Experience at Purdue [03:24] What Hebbia Does and How the Company Evolved [05:16] Why Finance is the Next Frontier for AI [05:55] What Makes High-Stakes Environments Different [09:57] Deterministic vs. Non-Deterministic Architectures [12:00] What Trust Really Means in Practice [13:52] Explainability, Consistency, and Human Checkpoints [14:58] Adoption Dynamics: Junior vs. Senior Financial Professionals [17:30] The Change Management Challenge at the Senior Level [18:08] Agentic AI and the Shift from Reactive to Autonomous [19:42] Autonomy Must Be Earned Incrementally [20:46] Security, Enterprise Controls, and the Governance Layer [22:37] Cloud Deployments, Model Management, and Customer Trust [24:44] The Security Hire from Bridgewater [25:22] The Gap Between Individual Productivity and Firm-Level Outcomes [27:22] The Textile Mill Analogy and Institutional AI [28:07] What Will Be Obvious in Two Years [30:18] Two Words for the Future of Work: Unprompted Intelligence --- Connect with Aabhas Sharma LinkedIn:   / aabhassharma   Connect with Boaz Ashkenazy LinkedIn:   / boazashkenazy   Email: [email protected]