After two decades of building systems like Apache Spark, BlinkDB, and large-scale ML and observability platforms at ThoughtSpot and Facebook, we’ve seen one truth: our ability to build has outpaced our ability to reason.
Every new abstraction: containers, orchestrators, managed services, even AI-generated code, helps us move faster but makes it harder to truly understand what we have built. The recent major outages at AWS, Meta, and Google Cloud all point to the same cause: not missing data, but missing reasoning. We don’t lack observability. We lack systems that can reason across all that data, connecting actions to consequences in real time.
And this gap does not just cause outages; it quietly slows teams down. Every new feature adds complexity, increasing the operational burden until engineers spend more time maintaining what exists than building what’s next. The distance between all the data we collect and our ability to make sense of it has become the defining bottleneck in software reliability.
This is why we started Deductive: to bring reasoning back to reliability.
The Core Insight: Reliability is a Reasoning Problem
Modern systems rarely fail because a single component fails. They fail because everything around that failure reacts, often correctly in isolation but unpredictably in combination. Each automation, retry, or safeguard behaves as intended, yet together they can amplify the problem instead of containing it. Reliability depends on how systems reason about change. As dependencies multiply, small updates, deployments, or policy shifts can trigger side effects that no tool or team can fully anticipate. Observability can surface symptoms but can’t show how actions interact across services, code paths, and environments.
The result is a gap between reacting to a failure and fully understanding it. The problem isn't missing data. We need systems that can connect cause and effect, and understand how local actions ripple out and impact systems as a whole.
The Technological Breakthrough: Reasoning Under Uncertainty
If reliability depends on reasoning, then the next generation of systems must learn to operate and adapt under uncertainty. Perfect understanding is impossible at the scale of modern software. Every signal is partial, every failure unique, and every decision must be made with incomplete information. In such environments, the only way forward is adaptability, with systems that learn continuously from outcomes, infer cause and effect, and refine their behavior through feedback.
Deductive is the first platform to apply reinforcement learning to reason about production behavior. Our AI SRE Agents learn from every incident which actions restore stability and which amplify failure. They do not rely on static playbooks or brittle rules. They continuously learn, adapt, and improve with experience, just like the best engineers, but at machine speed.
When something breaks, Deductive does not just surface correlations. It forms hypotheses, tests them against live evidence, and presents conclusions with proof — diffs, traces, and metrics that explain exactly why something broke and how to fix it. Over time, the system evolves from reactive investigation to proactive prevention, learning patterns of reasoning that no single human could hold in their head.
From the Builders of BlinkDB, Databricks, and ThoughtSpot
Sameer earned his Ph.D. in Computer Science from UC Berkeley, where he created BlinkDB, a pioneering system for approximate query processing that enabled fast, data-driven reasoning over petabytes of information. As one of the first engineers at Databricks, he helped build and scale Apache Spark, shaping the foundations of modern data infrastructure.
Rakesh was one of the early engineers at ThoughtSpot, where he helped build one of the most powerful search and AI-driven analytics platforms in the industry. He later led core engineering teams focused on distributed query processing, intelligent caching, and large-scale system optimization. His career has been defined by building systems that bring human intuition and machine intelligence closer together.
We met over a decade ago while working on some of the hardest data and reasoning problems in distributed systems. Across our careers, we have seen the same challenge repeat itself in every generation of software: as systems scale, understanding collapses. Deductive is our attempt to fix that, humbly but ambitiously, by putting reasoning at the core of reliability.
What Comes Next
Deductive is already running in production at some of the world’s most sophisticated engineering organizations, identifying root causes across hundreds of incidents with remarkable precision and speed. What begins with incident analysis will expand to prevention, performance optimization, and change risk prediction, enabling systems that learn continuously and improve on their own.
Our mission is to help teams focus on building, not firefighting.
We are excited to finally share what we have been building and to announce our launch out of stealth. We are also proud to share that Deductive has raised $7.5 million in seed funding led by Max Gazor at CRV, with participation from Databricks Ventures, Thomvest Ventures, and Primeset, along with prominent angels, entrepreneurs, and technology pioneers, including Ion Stoica (Founder, Databricks and Anyscale), Ajeet Singh (Founder, Nutanix and ThoughtSpot), Abhinav Asthana (Founder, Postman), Clint Sharp (Founder, Cribl), Ben Sigelman (Founder, Lightstep), and Amit Singhal (former SVP, Google Search).
If you are curious about what we are building, please book a demo. If you are passionate about shaping the future of reliable systems through cutting-edge AI research, we would love to hear from you!
