Sonarly Wants to Fix Your Production Bugs Before You Wake Up

The Macro: On-Call Is Broken and Everyone Knows It

Being on-call is the worst part of software engineering. I have done it. You have done it. Everyone who has shipped production code at scale has experienced the 3am page that turns out to be a duplicate of something you already fixed last Tuesday, or a flapping alert that nobody bothered to tune, or a genuine incident buried under seventeen false positives.

The tooling for dealing with this is better than it was ten years ago. Datadog, Sentry, Grafana, PagerDuty. These are good products. But they are fundamentally alert delivery systems. They tell you something is wrong. They do not tell you why. They definitely do not fix it.

There is a growing category of companies trying to change that. Rootly handles incident management workflows. FireHydrant does similar things. Shoreline automates runbooks. But most of these tools stop at orchestration. They help you respond faster. They do not actually diagnose root causes or write code to fix them.

The obvious question is whether AI is finally good enough to close that loop. Can you go from alert to diagnosis to fix without a human in the middle? That is a massive claim, and most teams would be skeptical. But the underlying models are getting genuinely capable at code understanding, and the monitoring data that feeds into these systems is highly structured. If there is a domain where autonomous AI has a real shot at working, production debugging might be it.

The Micro: Two French Engineers Chasing Autonomous Fixes

Sonarly connects to your existing monitoring stack and does three things. First, it triages alerts to remove noise and duplicates. Second, it investigates logs, traces, metrics, and code to find root cause. Third, it either generates a ready-to-merge fix PR or recommends updating your alerting rules so the noise goes away permanently.

The founding team is Dimittri Choudhury and Alexandre Klobb, both from France. Alexandre started freelancing at 16 and previously cofounded Meoria, a career guidance app that reached 100K French students. Dimittri cofounded a school-finding app that also scaled to 100K users. They are a two-person team from the YC Winter 2026 batch, working with Jared Friedman.

The product is live and currently free. I think the free pricing is smart at this stage. The hardest part of selling developer tools is getting engineers to actually try them, and engineers are notoriously resistant to adding new things to their stack. Free removes that friction entirely.

What makes Sonarly interesting is the claim that it builds “a living map of your production system.” If that actually works, it means the tool gets smarter over time as it processes more alerts and understands more about your specific architecture. That is a meaningful moat. Your monitoring data and system topology become training data that makes the product better for your specific environment.

The Verdict

I think Sonarly is swinging at the right pitch. The pain is real, the timing is right, and the approach is sound. Connecting to existing tools rather than replacing them is the correct go-to-market for developer infrastructure. Nobody wants to rip out Datadog. Everyone wants something that sits on top of Datadog and makes it more useful.

The risk is trust. Engineers do not trust automated fixes to production code. That is a cultural barrier, not a technical one, and it will take time to overcome. The first time Sonarly pushes a bad fix to production, the team that deployed it will rip it out and never come back.

In 30 days, I want to see how many fix PRs have been merged without modification. That is the metric that matters. Not PRs generated. PRs merged. In 60 days, the question is whether alert noise actually goes down for teams using the product. In 90 days, I want to know the false positive rate on root cause analysis. If Sonarly correctly identifies root cause 80% of the time, this company is going to be very valuable. If it is 50%, it is just another alert triage tool with a nice UI.