News, research and customer case studies from LogicStar.
A coding agent can turn a ticket into a green pull request in minutes. We split the merge decision into narrow checks, let fixed rules decide, set the strictness by the blast radius of a change, and bring in a person only where the agent gets stuck.
Blog
We ran three coding agents through four gateways to see what a gateway controls, what it measures correctly and where it misleads you, from budgets that overshoot to telemetry that leaks prompts and data residency set hop by hop.
We compared a stronger model, a different agent, answers from the author and a read-only investigation before any code. On 19 fresh tickets, the investigation went ahead on 9 of 10 wanted changes and held back 7 of 9 requests an author had already declined.
We ran OpenSpec, Spec Kit, Superpowers and Claude Code plan mode on a real ticket, then planted drift and contradictions. For one feature we couldn't separate the code, and plan mode cost about half. What differs most is the record each leaves for the next agent.
PROBE measures whether coding agents find the bugs maintainers fix. Across 20 repositories and 2,457 historical bugs, the best single run recovers 20.8%, even though most sampled findings are real defects.
We built TestSuiteBench to measure whether coding agents can turn a specification into a test suite that catches bugs. Their tests are valid but weak: the best agent catches 55% of injected bugs, while implementations built from the same specifications pass up to 93% of the reference tests.
Why automating the most visible part of the job did not automate the largest one: where the bottleneck moved, and how to tell whether your agents are actually getting better.
LogicStar reviewed an open-source EU AI Act RAG service and opened 38 public GitHub issues. What was found, what was fixed, and what the public record does and does not prove.
Field crews use the app offline, on site, where a crash means a lost record. How LogicStar found the bugs that mattered, including an authentication vulnerability, and turned production noise into root-cause fixes.
When your product's promise is that local behaves like production, a parity gap is a broken promise. How LogicStar finds the correctness gaps hand-written tests miss and ships them as merged fixes.
LogicStar found 2x more unique high-impact bugs than Claude Code with LogicStar-generated rules, at less than one-third the cost per unique high-impact bug.
An anonymized medical application was preparing for wider hospital rollout after being built with substantial AI assistance. In the first 24 hours after LogicStar was set up, the application owner fixed 23 production-relevant issues across frontend and backend workflows.
.png)
Most teams focus on what AI coding agents can do. We focus on where they fail. Explore the six research benchmarks that shaped LogicStar's approach to building reliable autonomous software engineering.
We analyzed the leaked Claude Code using LogicStar AI and found 10+ critical security issues, including remote code execution and permission bypasses. Learn what this means for developers.
Coding agents can write code. But can they decide what actually matters in large software systems? We explore why the next generation of AI software tools must move beyond patch generation toward architectural judgment.
We scale distillation of Agentic Coding Capabilities efficiently, to train a family of best-in-class coding models.
LogicStar Autonomously Finds and Fixes A Real Bug in Our Production Code
AI coding agents accelerate development but also drive up complexity and technical debt, causing early productivity gains to fade. Self-Healing Software closes this gap by automatically detecting and fixing issues as fast as new code is generated. LogicStar provides this capability, keeping codebases healthy and velocity sustainable.
We optimized the OCI layer structure of code execution environments to improve storage and distribution at scale
The L* agent achieves state-of-the-art results on SWE-Bench Verified using an ensemble of cheap agents and strong validation
The L* Agent achieves a new state-of-the-art of 84% on SWT-Bench Verified
LogicStar AI Successfully Completes SOC 2 Audit
LogicStar AI Joins the ETH AI Center as an Affiliate! 🚀
BaxBench: Can LLMs Generate Secure and Correct Backends?
A TechCrunch article about us titled LogicStar is building AI agents for app maintenance
SWT-Bench Benchmarking CodeAgents' Test Generation Capabilities
INSAIT and ETH Zurich Entrepreneurs launch LogicStar AI, a new Agentic AI startup
A Benchmark for Testing and Validating Bugfixes
LogicStar, building the AI agent for fully autonomous application maintenance, raised a $3m round led by Northzone.
We are looking for passionate software engineers to join our team
We are excited to announce the launch of LogicStar AI, our startup to revolutionize application monitoring.
With the team behind the first AI for code.
Backed by frontier coding agent research with ETH Zurich.
Cookie settings
By clicking 'Accept all cookies', you agree to storing cookies on your device to enhance site navigation, analyze site usage and assist in our marketing efforts as outlined in our cookie policy.