As software systems grow more complex and AI tools generate code faster than ever before, a fundamental problem is getting worse. Engineers are drowning in debugging workspending half their time hunting down the causes of software failures instead of building new products. The challenge has become so acute that it’s creating a new type of tooling — AI agents that can diagnose production failures in minutes instead of hours.
Deductive AIa startup that emerged from Stealth Mode Tuesday, believes it has found a solution by applying reinforcement learning—the same technology that powers game-playing AI systems—to the messy, high-stakes world of production software events. The company announced that it had raised $7.5 million in seed funding CRVwith participation from Data Bricks Venturesfor , for , for , . Thomwest Venturesand Primsetcommercializing what they say "AI SRE Agent" which can diagnose and support software failures at machine speed.
The pitch resonates with a growing frustration within engineering organizations: Advanced observation tools can show that something is broken, but they rarely explain it. When a production system fails at 3 a.m., engineers are still faced with hours of manual detective work, cross-referencing logs, metrics, deployment dates, and changes to dozens of interconnected services to identify the root cause of the code.
"The complexities and interdependencies of modern infrastructure mean that investigating the root cause of an outage or incident can feel like looking for a haystack, except the haystack is the size of a football field, it’s made of a million needles, it’s constantly changing itself, and every other one doesn’t equal lost revenue, and you don’t equal lost revenue, and everyone else doesn’t know it." In an exclusive interview with VentureBeat, Sameer Aggarwal, Co-Founder and Chief Technology Officer of Kututi said:
A deduction system builds what is called a company "Knowledge graph" which maps relationships in codebases, telemetry data, engineering discussions and internal documents. When an incident occurs, multiple AI agents work together to form hypotheses, test them against live system evidence, and arrive at a root cause — simulating the investigative workflow of experienced site reliability engineers, but completing the process in minutes instead of hours.
This technology has already shown measurable impact in the world’s most demanding manufacturing environments. Dordish’s advertising platformwhich runs real-time auctions that must complete in less than 100 milliseconds, has integrated the cut into its incident response workflow. The company has set an ambitious 2026 goal of resolving production incidents within 10 minutes.
"Our ADS platform operates at speeds where manual, slow-moving probes are no longer feasible. Every minute of downtime directly affects a company’s revenue," In an interview to VentureBeat, Shahrooz Ansari, senior director of engineering at Durash said: "Deduction has become a key extension of our team, rapidly synthesizing signals across dozens of services and surfacing its insights in minutes."
Dordish The cut is estimated to have translated to more than a thousand hours of annual engineering productivity lost over the past few months, due to the root cause of nearly 100 production incidents and revenue impacts. "in millions of dollars," According to Ansari. In a location intelligence company fucresquareDeductor reduced the time to diagnose Apache Spark job failures by 90% – a process that previously took hours or days to complete in less than 10 minutes – while generating more than 5,275,000 in annual savings.
Why AI-Generated Code Debugging is Creating a Crisis
The cutoff launch timing reflects a tension in software development: AI coding assistants enable engineers to produce code faster than ever before, but the resulting software is often harder to understand and maintain.
"Web Codingfor , for , for , ." A term popularized by AI researchers Andrej Karpathirefers to the generation of code by AI assistants using natural language. While these tools speed up development, they can introduce what Agarwal describes "Redundancies, breaks in architectural boundaries, assumptions, or neglected design patterns" which accumulates over time.
"Most AI-infused code still introduces redundancies, breaks architectural boundaries, makes assumptions, or ignores established design patterns." Agarwal told VentureBeat. "In many ways, we now need AI to help clean up the mess it’s creating itself."
The claim that engineers spend half their time on debugging is not hyperbole. The Association for Computing Machinery reports that developers spend 35% to 50% time validating and debugging software. More recently, State of the Harness Software Delivery 2025 The report found that 67% of developers are spending more time debugging AI-filled code.
"We’ve seen world-class engineers spend half their time debugging instead of building." said Rakesh Kothari, co-founder and CEO of Kututi. "And as Vibe Coding produces new code at a rate we’ve never seen, this problem is only going to get worse."
How Deductive AI Agents Actually Investigate Production Failures
The technical approach to deduction is quite different from the AI ​​features being added to existing observation platforms e.g Datadog or The new residue. Most of these systems use large language models to summarize data or identify correlations, but they lack Agarwal’s call. "Code-aware reasoning"The ability to understand not just why something is broken, but why the code behaves the way it does.
"Most enterprises use multiple observability tools across different teams and services, so no single vendor has a single comprehensive view of how their systems behave, fail, and recover—nor are they able to pair it with an understanding of the code that defines the system’s behavior." Agarwal explained. "These are the key components of software incident resolution and it absolutely fills the gap."
The system connects to existing infrastructure using read-only API access to observation platforms, code repositories, incident management tools, and chat systems. It then continuously builds and updates its knowledge graph, mapping dependencies between services and keeping track of deployment histories.
When an alert fires, the deduction launches what the company describes as a multi-agent investigation. Different agents specialize in different aspects of the problem: one can analyze recent code changes, another examines trace data, while a third correlates the time of the incident with recent deployments. Agents share results and iteratively refine their hypotheses.
A key difference from rule-based automation is the use of deductive reinforcement learning. The system learns from every incident that led to investigative actions that led to a correct diagnosis and those that ended up dead. When engineers provide feedback, the system adds signals to its learning model.
"Each time it observes an investigation, it learns which actions, data sources, and decisions led to the correct results." Agarwal said. "It learns how to think through problems, not just identify them."
At Dordish, an increasing service issue related to a recent delay in an API was initially isolated. Investigation of the outage revealed that the root cause was actually timeout errors from a downstream machine learning platform undergoing deployment. The system connected the dots by analyzing log volumes, traces, and deployment metadata across multiple services.
"Without deduplication, our team would have had to manually correlate latency spikes across all logs, traces, and deployment dates." Ansari said. "Deduction was able to explain not only what changed, but how and why it affected productive behavior."
The company keeps humans in the loop until now
While the deductive technology could theoretically push the fix directly to production systems, the company has deliberately chosen to keep humans in the loop.
"Although our system is capable of deep automation and can push fixes into production, currently, we recommend precise fixes and mitigations that engineers can review, validate, and apply." Agarwal said. "We believe that keeping a human in the loop is essential for trust, transparency and operational safety."
However, he admitted "Over time, we understand that deep automation will come and how humans work in the loop."
Veterans of data bricks and ideas bet on reasoning over observation
The founding team brings deep expertise from building some of Silicon Valley’s most successful data infrastructure platforms. Agarwal received his Ph.D. at UC Berkeley, where he created Blink DBan efficient system for approximate query processing. He was among the first engineers Data Brickswhere he helped build Apache Spark. Kothari was an early engineer A place of ideaswhere he led teams focused on distributed query processing and large-scale system optimization.
The Investor Syndicate reflects both technical credibility and market opportunities. Beyond the CRV Max Guzorincluded participation in the round Einsteinfounder of Databricks and InScale; Ajit SinghFounder of Newtonics and Ideas ; And Ben Sigelmanfounder of Lightstep.
Instead of competing with platforms like Datadog or PegridyDeductive positions itself as a complementary layer that sits on top of existing tools. The pricing model reflects this: instead of charging based on data volume, deductible fees based on the number of incidents investigated, plus a base platform fee.
The company offers both cloud-hosted and self-hosted deployment options and emphasizes that it does not store customer data on its servers or use it to train models for other customers.
With fresh capital and early customer traction in choice companies Dordishfor , for , for , . fucresquareand Come and gocuts plans to expand its team and deepen systems reasoning capabilities from reactive analysis to proactive prevention. Near-term vision: Helping teams predict problems before they happen.
Dordish’s Ansari offers a practical validation of where technology stands today: "Investigations that were previously manual and time-consuming are now automated, allowing engineers to shift their energy toward prevention, business impact, and innovation."
In an industry where every second of downtime translates into lost revenue, there’s a rapid shift from firefighting to building that looks less like luxury and less like table stakes.