AWS IDE that doesn't suck
#006 Specs are back (and AI made them good)
Hi everyone!
Two announcements this week actually matter for production teams: AWS shipped an Agentic IDE that doesn't suck, and Cognition (yes, the DEVIN people) acquired Windsurf in a move that somehow makes perfect sense.
Newsletter Rebrand Update: This newsletter is now "Agentic dev 3.o" and will focus exclusively on software engineering powered by Neural Networks. No more broad AI topics that don't directly relate to Agentic development or AI applied to the SDLC. I don’t want to bring another general AI newsletter like we need another JavaScript framework.
AWS Kiro Enters IDE Warzone
Yes, even AWS decided to fork VSCode (oh shocking I know). Meet Kiro, an Agentic IDE that implements spec-driven development using EARS (Easy Approach to Requirements Syntax) notation. They've built something that finally bridges the gap between "Claude Code, write me an app" and actual enterprise software development.
What This Actually Does: Kiro give you two Agent mode, Vibe and Spec, follow the white rabbit to get a plan, and generate specifications that don't sound like they were written by a sleep depraved engineer.
The free tier gives you 50 Agentic executions with Claude 4, which is very low compare to other solutions but AWS make it clear that their GPU are not for giveaway.
Why This Matters: Most "AI coding tools" skip the specification phase entirely and jump straight to generating code that "works" (in the loosest sense of the word). Kiro actually give you the choice to Vibe code or be serious through proper requirements gathering, technical design, and task planning. It's like having a product manager who actually knows what they're doing.
Steering Rules That Don't Suck
Kiro introduces "Steering" concepts that look like Cursor rules or CLAUDE.md files at first glance. But here's the difference: clean segregation between Product, Structure, and Tech rules with content that's actually precise. No more "just enough" details mixed with the usual unnecessary technical garbage.
The UX includes a dedicated panel for Steering rules, Specifications, and MCP tools. AWS has never been known for stellar UX, but they're actually crushing it here.
The Specification Workflow Actually Works
Requirements Phase: I tested this with Notionvibe, a SaaS that I'm building (go register on waitlist btw). The output quality genuinely surprised me. It’s cleaner and more precise than my custom PRD agent using Claude Code. The iteration capability is solid; I was able to add acceptance criteria and edge cases through two rounds without the AI losing context or generating conflicting requirements.
Design Phase: This phase generates technical implementation details including project architecture, data structures, schemas, and testing strategy. The quality here is impressive. This is a perfect example of what happens when you build proper Agentic workflows instead of just throwing prompts at a model and call it a wrapper. AWS clearly developed specialized agents with advanced context engineering and specification best practices as training examples.
Task List Phase: All tasks get detailed in a tasks.md file with bullet points that reference specific requirement sections. Each task includes a "Start task" button that triggers the Agentic workflow for development. The implementation is elegant: each specification gets its own directory with three markdown files, and the IDE UI makes everything easily accessible.
What This Solves: The eternal problem of specifications that exist only in someone's head, change every week, and somehow never match what gets built. Finally, a tool that enforces the discipline most teams skip.
MCP Integration
MCP Protocol has significant flaws, but its simplicity for exposing tools made it massively adopted across the industry. Beyond the ones I've built, I rarely use third-party MCPs, but Linear MCP has become a genuine time-saver, especially with Claude Code slash commands. In Kiro, it works well with detailed tool calls that show exactly which tool was invoked with which parameters.
Windsurf × Cognition: The Acquisition That Actually Makes Sense
The startup acquisition drama finally settled: Cognition acquired Windsurf after OpenAI and Google lost the bidding war. Yes, the same Cognition behind DEVIN—that AI developer poping in Slack channel that was supposed to replace us all but mostly just generated memes about broken commits.
The Reality Check: A year ago, we were all (rightfully) skeptical about DEVIN's capabilities. But with Windsurf's team now backing serious development of background tasks and Agentic workflows, this acquisition has genuine potential. The combination makes strategic sense: Windsurf's solid IDE foundation plus Cognition's Agentic development experience.
What This Means for Developers: Expect significant improvements in Windsurf's Agentic capabilities, background task execution, and workflow integration. The team that built a genuinely competitive coding assistant is now backed by the resources to implement more sophisticated agent architectures.
The Production Reality: Can Kiro Replace Your Current Setup?
For Individual Developers: If you're currently cobbling together Claude Code with custom slash commands and jumping between Cursor for manual coding, Kiro offers a unified workflow that actually enforces best practices. The specification-driven approach will slow you down initially but pays off later.
For Technical Leaders: This is the first AI coding tool that addresses the specification problem instead of ignoring it. The enforced structure means junior developers can't skip straight to implementation without thinking through requirements and design.
Cost Considerations: The free tier's 50 executions will give you a realistic evaluation period. Beyond that, expect enterprise pricing that reflects AWS's position in the market. Factor in the time saved on specification rework and requirements clarification when calculating ROI.
Keep in mind Kiro is still in “Preview” and lot of things can break and change in coming weeks. The Enterprise support is not yet there but given AWS experience with codewhisperer and amazon q they should pull of something good.
Looking Ahead: The Specification Renaissance
The common thread across these announcements is the industry finally acknowledging that good software starts with good specifications. We've spent two years watching AI tools generate impressive demos that fall apart in production because nobody bothered defining what "good" actually means.
Tools like Kiro represent a maturation of AI-assisted development—moving beyond "generate code that compiles" to "generate systems that solve actual problems." The Windsurf acquisition suggests this trend will accelerate as teams with serious engineering chops get resources to build proper agentic workflows.
The Pattern: Successful AI development tools are shifting from raw code generation to structured problem-solving workflows. Expect more tools that enforce methodology rather than just automating implementation.
What to Watch: As these specification-driven approaches prove their value, we'll likely see similar workflows appear in other IDEs. The question is whether they'll match the thoughtful agent design AWS demonstrated, or just bolt specification prompts onto existing chat interfaces.
Until next time, keep building—with proper specifications this time.
—Pierre
P.S. I love how AWS named their IDE "Kiro" when they could have gone with something sensible like "CodeSpec" or "AWS-Cloud9-SpecDriven"