Bilingual

Mouse: Precision Editing Tools for AI Coding Agents
21 days ago
- HIC Mouse is an AI file-editing tool using coordinate-based editing to improve surgical accuracy for AI agents.
- It offers staged changes with atomic rollback, allowing approval or cancellation of risky edits before saving.
- The tool provides embedded guidance, including suggestions, risk assessment, and file structure, to compensate for AI limitations.
Eclipse Enclave
20 days ago
- AI coding agents need autonomy to run commands, install packages, and edit files efficiently, but granting this directly on host machines poses risks like system damage, secret leaks, and interference.
- Eclipse Enclave provides a sandboxed runtime with isolated containers, filesystems, process trees, and network stacks, allowing multiple agents to work in parallel without interference via separate Git worktrees.
- The platform includes a sidecar gateway to restrict and log outbound network traffic, along with features for auth, configuration, history persistence, and a control center for managing agents.
- Its isolation, logging, and policy infrastructure offers audit trails, network access logs, and dependency records, helping organizations comply with regulations like the EU AI Act and Cyber Resilience Act.
- Eclipse Enclave separates agent execution from identity, treating agents as pluggable workloads to enable community focus on isolation, observability, and governance, while allowing users to mix agents and editors.
A verification loop 4x'd DeepSeek's intelligence, matching Opus at 1/7 the cost
18 days ago
- Verification loops significantly improve coding agents' ability to complete multi-step web development tasks by catching bugs early.
- Using IronBee (a verification layer) with DeepSeek (low-cost model) increased task completion from an average of 6.8 to 17 out of 20 tasks, matching the performance of Claude Opus (frontier model).
- DeepSeek with IronBee achieved similar results to Opus alone but at about 1/7th the cost ($2.4 vs. $15 per run), showing verification can reduce reliance on expensive models.
- The study emphasizes that verification helps most with bugs that only appear when the running application is tested, especially in sequential tasks where early errors compound.
- Limitations include using only one project and model pair initially, with plans to expand testing to more models, projects, and a cost-matched ablation study for broader insights.
A new study just debunked the biggest fear about AI and open source
18 days ago
- A study on AI coding agents in GitHub repositories found no significant decline in newcomer participation, contrary to fears that AI would push beginners out.
- While code complexity increased (e.g., cognitive complexity rose about 11% in Python projects), it did not deter new contributors, and the active contributor base grew.
- The research had limitations, including focusing on established projects with pre-AI history and measuring adoption via configuration files rather than actual tool usage.
- Despite AI-generated code being more complex, newcomer retention held steady, and the main concern shifts to the growing maintenance burden on existing maintainers.
- The study suggests future work should explore AI reliance levels and projects 'born with AI,' emphasizing the impact on maintenance efforts over contributor numbers.
Benchmarking coding agents on Databricks' multi-million line codebase
17 days ago
- Databricks developed an internal coding benchmark using real engineering tasks from their codebase, evaluating AI coding agents on performance and cost.
- The benchmark identified three distinct capability tiers among models, indicating that high-cost models aren't always necessary for common tasks, with models like GLM 5.2 offering competitive quality at lower costs.
- Token costs often poorly predict overall task expenses due to variations in reasoning efficiency, with differences in harness performance highlighting the importance of context management.
- The team emphasized the need for custom benchmarks over public ones like SWE-Bench to accurately reflect internal needs and ensure optimizations don't hinder developers.
- Using merged pull requests as a foundation, tasks were rigorously filtered and specified to create a reliable benchmark, with manual validation to avoid issues like leaked solutions from git history.
- Findings support using cost-effective models for routine tasks and intelligent routing to balance efficiency and capability, with plans to expand the benchmark and automate model selection.
A Study of Microsoft's Early 2026 Rollout of Claude Code and GitHub Copilot CLI
12 days ago
- The study examines the adoption, retention, and impact of command-line AI coding agents like Claude Code and Copilot CLI at Microsoft in early 2026.
- Findings indicate that initial adoption spread mainly through social networks among engineers.
- Retention was linked more to engineers' coding activity levels rather than demographic factors.
- Adopters merged approximately 24% more pull requests compared to their baseline, using merged PRs as a proxy for output.
- The productivity lift persisted over a four-month observation period.
- The results suggest CLI coding agents are not uniformly adopted or mere novelties, and peer influence is crucial for rollout strategies.
- Organizational token costs can be high, making accurate assessment of adoption and impact essential to avoid wasted investment.
MicroVM sandbox on Win, Mac, Linux, with policy engine
12 days ago
- Docker Sandboxes allow running AI coding agents in isolated microVM environments, each with its own Docker daemon, filesystem, and network.
- Agents can perform tasks like building containers, installing packages, and modifying files without affecting the host system.
- The sbx CLI is free for commercial and personal use, with only organization governance requiring a paid subscription.
- Organization admins can manage network and filesystem policies centrally via the Docker Admin Console for uniform rules across all developer machines.
- Installation instructions are provided for macOS (via Homebrew), Windows (via winget), and Linux (via curl and apt-get), including steps for Linux KVM setup.
- After installation and login, users can launch an agent (e.g., 'sbx run claude') from their project directory.
- Additional resources cover agents, customization, architecture, security, CLI reference, troubleshooting, and FAQs.
- Feedback and issues can be submitted via GitHub at github.com/docker/sbx-releases/issues to influence future development.
Stop Using OpenCode
5 days ago
- OpenCode is a popular open-source AI coding agent with 161k stars on GitHub.
- The author criticizes OpenCode for poor design, security vulnerabilities, and usability issues.
- Annoying aspects include inefficient prompt caching, flawed context pruning, broken subagent interactions, and a buggy text UI.
- Alarming security issues include trivial bypasses of command filters, unsafe default permissions, and a history of CVEs allowing remote code execution.
- The author advises users to stop using OpenCode and calls for better security practices in coding agents.
- A postscript discusses local LLMs, noting they avoid cloud dependence but still pose risks to code integrity.

Hasty Briefsbeta

ai coding agents

Mouse: Precision Editing Tools for AI Coding Agents

Eclipse Enclave

A verification loop 4x'd DeepSeek's intelligence, matching Opus at 1/7 the cost

A new study just debunked the biggest fear about AI and open source

Benchmarking coding agents on Databricks' multi-million line codebase

A Study of Microsoft's Early 2026 Rollout of Claude Code and GitHub Copilot CLI

MicroVM sandbox on Win, Mac, Linux, with policy engine

Stop Using OpenCode