Weekly Newsletter Digest - AI Native Dev
- News: 4 original pieces and 6 curated items for AI Native developers.
- Landscape: 7 new tools added.
- Podcast: Is Your Team Ready for AI-Driven Modernization?
- Events: Paris & London meetups on AI-native dev
1. News
AI Native Dev Originals 📌
Three AI IDEs, one spec-driven challenge: Cursor, Windsurf, and Copilot all managed to deliver working builds, both greenfield, and brownfield. The real separation came in ergonomics, UI polish, and how much hand-holding each agent needs. Cursor nailed polish and multi-file edit, but sometimes lost its thread. Windsurf shined in context retention and auto-project scaffolding. Copilot ran servers fastest and wrote the strongest tests, but required more human nudges. The takeaway: each IDE has its own style, and your fit depends on what your needs are. The bigger story, though, is what’s missing - none of these tools truly treat specs as first-class citizens.
Paul Duvall learned the hard way that custom slash commands in Claude Code aren’t much use if they vanish when you switch machines or projects. His fix was to treat them like real code: version control everything, bundle commands into an npm toolkit, and anchor them with a CLAUDE.md file for context. The result is portable, reproducible workflows - from /xgit for commits to /xsecurity for safety checks - that survive across projects and teams. The takeaway: customization is powerful, but only if it’s persistent.
AWS is taking a page from Claude Code’s playbook, rolling out customizable agents in its Q Developer CLI. Instead of manually reconfiguring prompts and contexts each time, devs can now predefine task-specific assistants through simple JSON files, scoping permissions, tools, and context. The move builds on Amazon Q’s integration with MCP, letting teams not just connect external MCP servers but decide how each agent should use them. It’s a push toward modular, interoperable assistants; though, as early adopters point out, stitching multiple agents together still requires manual effort. For now, it’s a meaningful step: Q is becoming less of a single AI companion and more of a platform to shape your own.
Anthropic is giving devs more say in how Claude Code communicates. Beyond the default terse mode, you can now pick explanatory (narrated reasoning and tradeoff analysis) or learning (a guided pair-programming style with TODOs for you to fill in). You can even scaffold your own styles with /output-style. Combined with Claude’s subagents, which shape what the assistant does, output styles shape how it collaborates. Early users like the flexibility, though some note Claude’s default cheeriness is hard to suppress.
AI Native Dev Radar 📡
- Claude Code: Now in Beta in Zed.
- Why I’m Against Claude Code’s Grep-Only Retrieval?
- Anthropic raises $13B Series F at $183B post-money valuation.
- Definitive guide to context engineering fundamentals.
- Using Q CLI to validate the implementation of Kiros specs.
- Community-driven open standards for AI context across platforms.
2. AIND Landscape
This week, we’ll be adding 7 new tools to the Landscape:
- Codeplain, spec-driven, production-ready, code generation in plain language.
- Cclsp, a non-IDE-dependent LSP integration for Claude Code.
- Claude code lsp, a file-based type checker that integrates with Claude Code to provide real-time diagnostics for programming languages.
- Serena, a coding agent toolkit providing semantic retrieval and editing capabilities.
- mcp-language-server, gives MCP-enabled clients access to semantic tools like get definition, references, rename, and diagnostics.
- GitHits, an agentic code search that distills public repos.
- Command Center, turns code changes into interactive walkthroughs.
3. AI Native Dev Podcast
In this episode of AI Native Dev, Birgitta Böckeler (Distinguished Engineer at ThoughtWorks) draws a sharp line that many teams blur: AI-in-products vs. AI-for-building-software. Her focus is the latter: how AI reshapes the software delivery lifecycle itself.
Birgitta shares that ThoughtWorks’ Technology Radar now sees over half of new entries tied to generative AI. Interestingly, no single vendor tool earned an “Adopt,” but one technique did: RAG. It’s a telling signal and a pointer for engineering leaders to invest in practices that reduce hallucinations and ground outputs in real project context.
The conversation then dives into the rise of agentic coding - tools that don’t just autocomplete a function, but can span multiple files, run tests, issue terminal commands, and coordinate feedback loops. That shift redefines the dev’s role: from primary author to orchestrator.
“You aren’t just using an AI assistant; you’re assembling an agentic system tailored to your migration pattern.”
Takeaways for leaders:
- Separate AI inside your product from AI inside your workflow.
- Invest in repeatable, observable guardrails: tests, evals, and review gates.
- Favor techniques (like RAG) over vendor lock-in.
- Think orchestration, not abdication; devs are managers of these new loops.
A must-listen if you’re figuring out how to make AI productive across coding, refactoring, and legacy migrations.
05:28 Building with agents
11:33 Legacy migration
24:45 Reliable validation
28:24 Success expectation
32:35 Changes in legacy modernisation
39:13 The humans in the loop
4. AIND Events
AI Native Dev is hitting the road this September with two exciting community meetups - Paris and London- bringing together engineers, tool builders, and thought leaders shaping the future of AI-native software development.
🇫🇷 Paris - AI Native Dev Meetup at AI Engineer Fair (Sept 22, 19:00, Theodo Cloud)
Explore how AI is reshaping the software development lifecycle.
- Insights from our very own Patrick Debois (DevOpsDays founder, co-author of The DevOps Handbook)
- Practical lessons on AI-assisted workflows and tools
- Networking with peers building the future of engineering
🇬🇧 London - Expert Panel: Taming Agents with Specifications (Sept 23, 18:00, Tessl HQ)
A deep dive into Spec-Driven Development (SDD) and how it keeps agentic coding on track.
- Panel featuring Guy Podjarny, Simon Maple, Don Syme, and Alan Pope
- Honest discussion on the promise, pitfalls, and future of SDD
- Time to connect, share stories, and swap lessons from the frontier of AI-native dev
Looking forward to seeing you there!