https://iceshore.ai/en/radar/ 2026-11-07T00:00:00Z IceShore Radar Editorial Team https://iceshore.ai/en/radar/rn-1046/ 2026-11-07T00:00:00Z

Researchers studied 7,581 pull requests written by GitHub Copilot's coding agent with public session logs. In 58%, no developer stepped in while the agent worked; when developers did, the exchange was brief, a median of two rounds. After the first intervention, the agent's work trail dropped significantly, by 20.8 iterations.

https://iceshore.ai/en/radar/rn-1048/ 2026-11-07T00:00:00Z

A single-author case study tracked the author's own markdown repository edited daily by three AI agent tools under one rule set. Of 118 recorded failures, stale documentation was the largest class, at 48. An AI-run audit found about half of known stale statements, with at least 93% of its findings correct, and the author urges retesting old rules.

https://iceshore.ai/en/radar/rn-1052/ 2026-11-07T00:00:00Z

CodeMeepo, from an arXiv preprint, splits a whole-repository language migration among eight specialized AI roles and checks each milestone against a fixed test oracle the agents cannot edit. Across nine projects and three language pairs it passed 18% more tests than a leading multi-agent baseline using the same model.

https://iceshore.ai/en/radar/rn-0010/ 2026-11-06T00:00:00Z

Blitzy, which builds an AI platform for large company codebases, announced a collaboration with Microsoft. Blitzy says its platform maps older, poorly documented systems, then hands the work to many AI agents that build from a specification a person has approved. It will also bring Microsoft's GitHub Copilot and Azure services into its platform.

https://iceshore.ai/en/radar/rn-1056/ 2026-11-06T00:00:00Z

Researchers ran 9,400 experiments with two coding agents to test 42 ways of deleting old conversation history to save money. Under flat per-token pricing, estimated costs fell in 26 of the 42 setups. Under pricing that discounts cached input, none lowered costs, because trimming breaks the cached prefix and agents often re-fetched what was deleted.

https://iceshore.ai/en/radar/rn-1047/ 2026-11-06T00:00:00Z

An arXiv preprint analyzed 14,320 Java unit tests written by four AI models under four prompting styles. Incorrect assertions caused 15.6% of test failures, and the tests rarely checked null, negative or empty inputs, showing a bias toward the happy path. The gaps appeared even in simple functions.

https://iceshore.ai/en/radar/rn-0006/ 2026-11-05T00:00:00Z

Google published a research reference design for AI agents that respond to production outages on their own. In simulated Black Friday sale outages, the time until the first mitigation fell from 15–45 minutes to 1.8 seconds. Permanent changes to how the agents are organized still require sign-off from site reliability engineers.

https://iceshore.ai/en/radar/rn-0007/ 2026-11-05T00:00:00Z

GitHub made Copilot approvals generally available for its AI code review. When admins turn the feature on, an approval from Copilot counts toward a repository's required approvals. Adding commits after Copilot signs off cancels its approval, so a fresh review is needed.

https://iceshore.ai/en/radar/rn-0008/ 2026-11-05T00:00:00Z

In a post on its engineering blog, Inkitt described a system in which AI agents take small, low-risk code changes from the initial request to a draft pull request. Steps such as checking the work and repairing failed automated test runs get at most two attempts each. A person still has to mark the pull request ready for review and approve it.

https://iceshore.ai/en/radar/rn-1049/ 2026-11-04T00:00:00Z

A preprint describes StateMirror, which checks an agent's planned commands against the live repository before they run, catching references to missing files or packages. In 4,398 logged sessions such errors cost a median of three extra steps. On 79 live tasks it cut execution time 15.8% and token use 11.8% without lowering the solve rate.

https://iceshore.ai/en/radar/rn-1040/ 2026-11-04T00:00:00Z

Researchers built 205 tasks from cases where users said they regretted not saving something an agent later changed. Six frontier models finished 86.1% of tasks on average but backed up the affected files, environments or settings beforehand only 15.2% of the time. An add-on backup layer the team built covered every such action.

https://iceshore.ai/en/radar/rn-1043/ 2026-11-04T00:00:00Z

A preprint study of 22,854 widely used, active GitHub projects found instruction files for coding agents, AGENTS.md or CLAUDE.md, in 26.3% of them and in 59.6% of the most-starred. Of 341,852 statements analyzed, 74.0% of directives gave no reason, and projects split on whether to disclose AI use.

https://iceshore.ai/en/radar/rn-1053/ 2026-11-03T00:00:00Z

SpeC2Rust, described in an arXiv preprint, has an AI model write specifications for each module of a C codebase, then translate and repair the code against them. All eight test projects compiled in Rust and passed their original end-to-end tests with no unsafe code. Against an earlier AI-based approach, cost fell 66.1% and input tokens 84.7%.

https://iceshore.ai/en/radar/rn-1042/ 2026-11-03T00:00:00Z

Researchers had seven AI models translate code across six language pairs on three benchmarks and hand-labeled 13,919 cases where the output strayed from the original. They sort the errors into three types, including semantic conflicts and lost project context, and propose a fix that retrieves similar past translations.

https://iceshore.ai/en/radar/rn-1059/ 2026-11-03T00:00:00Z

Linus Torvalds, creator of the Linux kernel, told a Register columnist that arguments against using AI for commit messages are misguided, and that Linux wants capable developers to use AI as a tool. He said submitters who send AI-made patches they don't understand should be discouraged. The Register published the remarks on Oct. 23.

https://iceshore.ai/en/radar/rn-1057/ 2026-11-02T00:00:00Z

A preprint introducing the FeatureLoop benchmark tested 12 frontier models on 61 human-reviewed tasks from 34 repositories, with requirements changing mid-project. The best model finished 72.1% of tasks, but only 15.4% of failed stages recovered within two rounds of corrective feedback. The authors also saw incomplete repairs and regressions.

https://iceshore.ai/en/radar/rn-1035/ 2026-11-02T00:00:00Z

A benchmark of 43 tasks in mature enterprise codebases, checked against 750 expert-audited points, measures whether coding agents fulfill full product requirement documents. Across 11 model and tool setups, the best complete-delivery rate was 14.0%, by OpenAI's GPT-5.6 Sol; missing integration was a main cause.

https://iceshore.ai/en/radar/rn-1045/ 2026-11-02T00:00:00Z

GitHub said on Oct. 23 that its app for Microsoft Teams now brings in Copilot, its AI coding assistant. When a bug report lands in a Teams chat, users can mention GitHub to have Copilot investigate, fix the bug and open a pull request for the team to review. The old GitHub app for Teams will be retired on Nov. 16.

https://iceshore.ai/en/radar/rn-1044/ 2026-11-01T00:00:00Z

Research firm Forrester released its 2027 technology and security predictions on Oct. 21. It expects AI-assisted coding and patching to outpace teams' capacity to test, and forecasts that a bug or regression in AI-generated code will trigger an outage spanning several industries worldwide.

https://iceshore.ai/en/radar/rn-1039/ 2026-11-01T00:00:00Z

Aimed at agent-driven migrations, a preprint generated 849 end-to-end test cases for a 33,429-line COBOL program to check its Java translation, finding and fixing 61 defects in the code generator. When the authors planted 450 defects, the check caught 47.6%, including every deleted write in executed code, but no rounding errors.

https://iceshore.ai/en/radar/rn-1038/ 2026-11-01T00:00:00Z

On his site Oct. 21, software author Martin Fowler recommended an essay by Cindy Sridharan on debugging with AI agents. Her view, which he summarized approvingly, is that an agent should first gather evidence about how the system behaves and only then form hypotheses from the code. Fowler noted that a human is still needed to steer the agent.

https://iceshore.ai/en/radar/rn-1041/ 2026-10-31T00:00:00Z

OpenAI says GPT-6.1 Sol correctly migrated COBOL programs to languages such as Java or Python without changing their behavior 86.6% of the time at a low reasoning setting, up from 5.4% for GPT-5.6 Sol, and 97.9% with more reasoning. The tests are OpenAI's own. BleepingComputer notes that real migrations would still need expert review and testing.

https://iceshore.ai/en/radar/rn-1032/ 2026-10-31T00:00:00Z

InfoQ reports Grafana Labs wants teams to measure AI agent behavior, such as hallucination and task fulfilment, and manage it with service-level objectives and error budgets. Its example target: 95% of conversations judged as fulfilled over 30 days, with the rest as the error budget.

https://iceshore.ai/en/radar/rn-1033/ 2026-10-31T00:00:00Z

On an AWS blog, map company HERE Technologies said it uses multiple Kiro agents to turn customer requirements into Jira tickets that keep the traceability automotive audits demand. Based on early internal measurements, it expects to cut the investigation phase for new tasks by more than 50%.

https://iceshore.ai/en/radar/rn-1058/ 2026-10-30T00:00:00Z

LY Corp. said on its tech blog that AI agents now draft requirements and designs for its Japanese shopping site, a service with over 3,000 repositories and 27 years of history. The team first organized domain knowledge and system diagrams for the AI, and senior engineers review each stage. Planning time fell from about two months to roughly one.

https://iceshore.ai/en/radar/rn-0004/ 2026-10-30T00:00:00Z

Japanese internet company CyberAgent said on its engineering blog that it used generative AI to catalog about 200 AWS Lambda functions, small cloud programs running in production and managed by multiple people. It refactored and redeployed 38 of them. Risky steps such as production deploys required explicit approval from a human engineer.

https://iceshore.ai/en/radar/rn-1027/ 2026-10-30T00:00:00Z

Amazon Web Services said Oct. 15 that Resilience Hub, its service for checking how applications withstand failures, now starts a DevOps Agent investigation to watch each resilience test and report symptoms, root cause, impact and recovery actions.

https://iceshore.ai/en/radar/rn-0001/ 2026-10-29T00:00:00Z

Research firm Gartner said on Oct. 26 that by 2030, 50% of organizations will use AI agents that need minimal human involvement in IT operations, up from nearly 0% in 2026, Japan's ITmedia reported. In a Gartner survey, 58% of IT operations leaders said moving AI into production was an important priority.

https://iceshore.ai/en/radar/rn-0002/ 2026-10-29T00:00:00Z

GitHub's annual Octoverse report says that as of September, more than 40% of pull requests, the proposed code changes submitted for review, began with a developer directing an AI agent, up from 17% in March. The report describes code review as the way developers shape what the agents return.

https://iceshore.ai/en/radar/rn-0003/ 2026-10-29T00:00:00Z

Amazon Web Services published a walkthrough of AWS DevOps Agent, its AI operations tool, tracing a failure across separately managed Terraform stacks. In the example, an infrastructure team's tighter network rule cuts an app team's function off from its database, and the agent uses alarms, dependencies and change logs to point to the likely cause.