#32 - 🚀 Buy vs build & parallel AI agents

Writer's code review decision + OpenHands Cloud

November 18, 2025

Hey readers! 🚀

This week brings a fascinating look at how AI coding tools are maturing beyond the hype cycle. From Writer's decision to buy rather than build their code review solution, to OpenHands launching a cloud platform for parallel AI agents, we're seeing the industry shift from experimentation to production-grade implementation. Plus, we've got practical insights on verification strategies, security considerations, and the real productivity numbers behind AI-assisted development.

This Week's Highlights 🎯

Buy vs build: Why Writer bought their AI code review tool explores a critical decision many teams face today. Writer chose to purchase CodeRabbit's AI code review tool rather than build an in-house solution, saving engineering resources while expanding coverage across 37+ repositories and achieving a 70% suggestion acceptance rate. – CodeRabbit

The case study reveals compelling metrics: team leads saw roughly 30% faster code reviews, and the tool caught subtle bugs like an overly permissive UUID regex that human reviewers might have missed. What makes this particularly interesting is how Writer implemented path-based rules to consolidate style guides across teams, creating consistency without heavy-handed enforcement. The VS Code and Cursor integrations, along with JIRA tracking and sequence diagrams, improved cross-team visibility and made large PR reviews faster without requiring local checkout.

"Personally, my code review time is down around 30%. When I see useful comments, I don't focus on those areas; I cover other areas and rely on CodeRabbit to have my back."

Introducing the OpenHands Cloud marks a significant milestone for the open-source AI coding agent. The platform brings browser-based AI software development capabilities with deep GitHub integration, allowing developers to start sessions from issues, get instant PR assistance, and automate pull request creation. – OpenHands

The standout feature is parallel agent capabilities, letting developers run multiple AI assistants simultaneously on different tasks. This addresses a real bottleneck in AI-assisted development: waiting for one agent to finish before starting the next task. Built on a fully open-source codebase, the platform encourages community contribution and customization, which could accelerate innovation compared to closed alternatives.

What to verify in AI-generated code tackles a critical gap in current AI coding practices. Jennifer Sand argues that traditional testing approaches fall short because AI agents work from instructions rather than understanding intent, leading to code that passes tests but introduces subtle production issues. – Codential

The article proposes an invariant-driven framework that helps AI agents understand production-ready requirements before testing begins. This is particularly relevant for concurrency bugs that appear intermittently, code that performs perfectly in test environments but collapses with real data, and systems too complex to test every combination. The key insight: verification techniques that assert what must always be true can prevent issues that testing alone cannot catch.

"AI agents produce code that looks correct very quickly, but they work from what you tell them, not what you mean."

Practical Workflows 🛠️

Vibe Coding Higher Quality Code offers battle-tested practices for maintaining code quality when heavily relying on AI-generated code. The author, drawing from experience with OpenHands, outlines five essential practices: utilizing static analysis tools like mypy and ruff, practicing test-driven development where agents write tests before code, implementing CI/CD with GitHub Actions, enabling repository customization to enforce best practices, and performing two-tier code reviews. – OpenHands

Using OpenHands to Improve your Test Coverage demonstrates practical application of AI agents for a common development task. The author used OpenHands to generate additional unit tests for a side project and discovered the tool caught a previously missed edge case involving non-English characters like "José" and "São Paulo." – Tim O'Farrell

Automating Massive Refactors with Parallel Agents introduces the Refactor SDK for breaking down large-scale refactoring projects into manageable tasks that multiple AI agents can handle independently. The approach emphasizes human oversight while leveraging AI to handle tedious work, potentially shrinking months-long projects to days. – OpenHands

Remediating CVEs at Scale with Agents addresses security vulnerability management through AI automation. The article explains how cloud-based agents can handle large CVE backlogs by running multiple agents concurrently, each working on different vulnerabilities with their own sandbox and security policy. – OpenHands

Security and Trust 🔒

Liran Tal's insights on secure AI coding emphasize integrating security tools like Snyk to enhance AI coding agents' capabilities. He promotes secure code generation, secure package selection, and secure Docker image selection, highlighting the need for continuous security testing as AI-driven code generation accelerates. – Liran Tal

The security expert raises an important question about teaching developers to use AI coding agents safely, covering both tooling and education around risks and threats. This reflects growing awareness that AI coding tools need security guardrails, not just productivity features.

Google reveals its own version of Apple's AI cloud introduces Private AI Compute, a cloud-based processing environment that brings on-device-style privacy to cloud AI. The service pairs Gemini models with protected execution environments using TPUs, Titanium Intelligence Enclaves, and encrypted connections, with a "zero access" assurance that data remains inaccessible even to Google engineers. – Muhammad Zulhusni

Productivity Reality Check 📊

AI Productivity Divide: Are Some Devs 5x Faster? examines conflicting research on AI coding productivity gains. While some studies show significant improvements (up to 5x), especially for less experienced developers, others indicate no benefit or decreased productivity in complex codebases. The author concludes realistic expectations are closer to 20% productivity gains, depending heavily on developer experience and tool usage. – Docker

"My biggest conclusion from this research is that developers shouldn't expect anything in the order of 3-5x productivity gains. Even if you manage to produce 3-5x as much code with AI as you would if you were doing it manually, the code might not be up to a reasonable standard."

AI coding assistants don't save much time: Software engineer offers a skeptical perspective from a practicing engineer. While AI tools excel at simple queries, the author expresses concern about their effectiveness in complex scenarios and warns against over-reliance that might hinder development of essential problem-solving skills. – Alain Dekker

Most developers in Southeast Asia and India use AI tools reports that 95% of developers in these regions use AI tools weekly, primarily to accelerate software development. Despite high adoption, developers maintain "pragmatic optimism," relying on human oversight for quality control. – Aaron Tan

Additional Resources 📚

Made with ❤️ by Data Drift Press

Questions, comments, or feedback? Just hit reply – we'd love to hear from you!