As software systems become more complex and AI tools generate code faster than ever, a fundamental problem is growing: Engineers are drowning in debuggingwho spend up to half their time tracking down the causes of software bugs instead of building new products. The challenge has become so acute that a new category of tools is emerging: AI agents that can diagnose production failures in minutes instead of hours.
Deductive AIa startup emerging from stealth mode on Tuesday believes it has found a solution by applying reinforcement learning — the same technology that powers AI systems for playing games — to the messy world of high-stakes production software incidents. The company announced it has raised $7.5 million in seed funding led by CRVwith the participation of Databricks Ventures, Thomvest VenturesAnd PrimeSetto commercialize what it calls “AI SRE agents” that can diagnose and help resolve software errors at machine speed.
The pitch reflects growing frustration within tech organizations: Modern observation tools can show that something has broken, but rarely explain why. When a production system goes down at 3 a.m., engineers still have to do hours of manual detective work, comparing logs, metrics, deployment histories, and code changes across dozens of interconnected services to identify the root cause.
“The complexity and interdependencies of modern infrastructure mean that investigating the root cause of an outage or incident can feel like looking for a needle in a haystack, except the haystack is the size of a football field, made of a million other needles, constantly rearranging itself and on fire – and every second you don’t find equals lost revenue,” said Sameer Agarwal, co-founder and chief technology officer of Deductive, in an exclusive interview. with VentureBeat.
Deductive’s system builds what the company calls a “knowledge graph,” which maps relationships between codebases, telemetry data, technical discussions and internal documentation. When an incident occurs, multiple AI agents work together to form hypotheses, test them against live system evidence, and converge on a root cause – mimicking the investigative workflow of experienced site reliability engineers, but completing the process in minutes instead of hours.
The technology has already shown measurable impact in some of the world’s most demanding manufacturing environments. DoorDash’s advertising platformwhich runs real-time auctions that must be completed in less than 100 milliseconds, has integrated Deductive into its incident response workflow. The company has set itself the ambitious goal of resolving production incidents within 10 minutes by 2026.
“Our advertising platform is operating at a pace where manual, slow investigations are no longer feasible. Every minute of downtime directly impacts the bottom line,” said Shahrooz Ansari, Senior Director of Engineering at DoorDash, in an interview with VentureBeat. “Deductive has become a critical extension of our team, quickly synthesizing signals from dozens of services and surfacing the insights that matter in minutes.”
DoorDash estimates that Deductive has caused approximately 100 production incidents in recent months, which Ansari says translates into more than 1,000 hours of annual technical productivity and a revenue impact “in millions of dollars.” At location information company SquareDeductive has reduced the time to diagnose failed Apache Spark jobs by 90%, completing a process that previously took hours or days in less than 10 minutes, while saving more than $275,000 annually.
Why AI-generated code is causing a debugging crisis
The timing of Deductive’s launch reflects a growing tension in software development: AI coding assistants allow engineers to generate code faster than ever, but the resulting software is often harder to understand and maintain.
“Vibe coding”, a term popularized by AI researcher Andrei Karpathyrefers to the use of natural language prompts to generate code via AI assistants. While these tools speed up development, they can introduce what Agarwal describes as “redundancies, breaks in architectural boundaries, assumptions, or ignored design patterns” that accumulate over time.
“Most AI-generated 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 clean up the mess that AI itself creates.”
The claim that engineers spend about half their time debugging is not an exaggeration. The Association for Computing Machinery reports that developers are spending money Spending 35% to 50% of their time validating and debugging software. More recently, Harness’s state of software delivery 2025 The report shows that 67% of developers are spending more time debugging AI-generated code.
“We’ve seen world-class engineers spend half their time debugging instead of building,” said Rakesh Kothari, co-founder and CEO of Deductive. “And because vibe coding is generating new code at a rate we’ve never seen before, this problem will only get worse.”
How Deduction’s AI agents actually investigate production errors
Deductive’s technical approach differs substantially from the AI features being added to existing observation platforms such as Data hound or New relic. Most of these systems use large language models to summarize data or identify correlations, but they lack what Agarwal calls “code-aware reasoning”: the ability to understand not only that something has broken, but also why the code behaves the way it does.
“Most companies use multiple observability tools across different teams and services, so no vendor has a single, holistic view of how their systems behave, fail and recover – nor are they able to tie that to an understanding of the code that defines system behavior,” Agarwal explains. “These are the key ingredients for solving software incidents and it is exactly the gap that Deductive fills.”
The system connects to existing infrastructure using read-only API access to observability platforms, code repositories, incident management tools, and chat systems. The knowledge graph is then continuously built and updated, mapping dependencies between services and tracking deployment history.
When an alert goes off, Deductive launches what the company describes as a multi-agent investigation. Different agents specialize in different aspects of the problem: one analyzes recent code changes, another examines trace data, while a third correlates the timing of the incident with recent deployments. The agents share findings and iteratively refine their hypotheses.
The critical difference from rule-based automation is Deductive’s use of reinforcement learning. The system learns from each incident which investigation steps led to the correct diagnoses and which were dead ends. When engineers provide feedback, the system incorporates that signal into its learning model.
“Every time it observes a study, it learns which steps, data sources and decisions led to the right outcome,” Agarwal said. “It teaches how to think about problems, not just point them out.”
At DoorDash, a recent latency spike in an API initially appeared to be an isolated service issue. Deductive’s investigation found that the root cause was actually timeout errors from a downstream machine learning platform that was being implemented. The system connected these dots by analyzing log volumes, traces, and metadata across multiple service deployments.
“Without Deductive, our team would have had to manually correlate the latency spike across all logs, traces, and deployment histories,” Ansari said. “Deductively could explain not only what changed, but also how and why it affected production behavior.”
The company is keeping people informed – for now
While Deductive Solutions’ technology could theoretically push straight to production systems, the company has consciously chosen to keep people in the loop – at least for now.
“While our system is capable of deeper automation and could push solutions to production, we currently recommend precision fixes and solutions that engineers can review, validate and apply,” Agarwal said. “We believe having a human in the loop is essential for trust, transparency and operational security.”
However, he acknowledged that “over time we think there will be deeper automation and the way people operate will evolve over time.”
Databricks and ThoughtSpot veterans are betting on reasoning about observability
The founding team brings deep expertise in building some of Silicon Valley’s most successful data infrastructure platforms. Agarwal received his Ph.D. at UC Berkeley, where he created BlinkDBan influential system for processing estimated queries. He was one of the first engineers Databrickswhere he helped build Apache spark. Kothari was an early engineer at ThoughtSpotwhere he led teams focused on distributed query processing and large-scale system optimization.
The investor syndicate reflects both technical credibility and market opportunity. In addition to CRVs Max Gazarthe round included participation from Ion Stoicafounder of Databricks and Anyscale; Ajeet Singhfounder of Nutanix and ThoughtSpot; And Ben Sigelmanfounder of Lightstep.
Instead of competing with platforms like Data hound or PagerDutyDeductive 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, you are charged deductively 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 – a crucial assurance given the proprietary nature of both the code and behavior of the production system.
With fresh capital and early customer acquisition at companies like DoorDash, SquareAnd TenderDeductive plans to expand its team and deepen the system’s reasoning capabilities, from reactive incident analysis to proactive prevention. The short-term view: helping teams predict problems before they happen.
DoorDash’s Ansari offers a pragmatic endorsement of where the technology is today: “Surveys 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, that shift from firefighting to construction is looking less and less like a luxury and more like a table stake.
#Deductive #Saved #DoorDash #Tech #Hours #Automating #Software #Debugging