SpecMem: How Kiroween in San Francisco Sparked the First Unified Agent Experience and Pragmatic Memory for Coding Agents

Source: Dev.to

Discovering a Workspace

Wandering the Embarcadero, I needed a spot to charge my phone and MacBook. A quick search pointed me to the AWS Builders Loft. The front desk mentioned an event upstairs, but I decided to try the space anyway. On the second floor I immediately felt the buzz of an AI coding hackathon. After registering on Luma and showing my passport, I learned the event was called Kiroween—a playful blend of “Kiro” and “Halloween.”

Kiroween Hackathon

I had heard of Kiro during a prompt‑engineering talk in London by Ricardo Sueiras, but I’d never tried it. Skeptical of VS Code forks and AI‑driven IDEs, I usually stick to CLI‑based editors. Still, the vibrant atmosphere at the AWS Loft—breakfast, free Wi‑Fi, and a crowd of Silicon Valley builders—made me give Kiro a chance.

Kiro and Spec‑Driven Development

Kiro markets itself as a structured AI coding platform built around spec‑driven development (SDD). Recent buzz includes GitHub’s SpecKit, AWS’s launch of Kiro, and several Bay‑Area and London startups promoting SDD. The community is still debating whether SDD is genuine innovation or hype, with viewpoints from ThoughtWorks, Marmelab, and the AI Engineer Code Summit (hosted by Dex Horthy of HumanLayer).

Having used TDD/BDD since 2012 (RSpec, Cucumber) at companies like AOL and BCC, I was curious how an IDE could embed SDD as a core workflow. After a quick setup—creating a new Python project with uv—Kiro generated a project structure where specifications are written in the same language as the code, reminiscent of RSpec/Cucumber but fully integrated.

Workflow Stages in Kiro

1. Requirements (Product Owner / Business Analyst)

   • Write high‑level user stories and acceptance criteria using the familiar “As a … I want … So that …” format (Gherkin‑style).

2. Design (Technical Architect)

   • Define the system’s technical architecture and overall plan, not the implementation details. This step feels more like an “Architect” or “Plan” phase rather than UI design.

3. Tasks (Developers)

   • Implement concrete tasks, view changes in real time, and watch modular execution directly in the IDE—similar to pulling tasks from JIRA but without leaving the development environment.

This three‑stage flow makes Kiro feel like a behavior‑driven IDE, bringing classic BDD concepts into the era of AI agents. It reminded me of The RSpec Book and The Cucumber Book (both 2013), now refreshed with agentic capabilities.

Lightweight and Focused

Despite my wariness of heavyweight AI IDEs, Kiro surprised me with its performance. After years of using VS Code and its many forks, I’ve grown accustomed to the trade‑off between feature richness and resource consumption. Kiro feels notably lighter while still offering AI‑enhanced development.

Reflections on Hackathons and Platform Choices

In recent months, many AI hackathons feel more like user‑acquisition campaigns funded by venture capital, offering modest prizes and platform credits in exchange for engagement. The authentic spirit of pre‑AI hackathons—builders collaborating purely to create and learn—seems to be fading. As a founder, I’m increasingly selective about which events to join, weighing the value of building on someone else’s platform against the potential IP complexities.

Closing Thoughts

My accidental detour in San Francisco led to the first hands‑on experience with a unified agent experience (AX) and pragmatic memory for coding agents. Kiroween and Kiro demonstrated that spec‑driven development can be a practical workflow for AI‑augmented coding, bridging the gap between traditional BDD practices and the emerging world of autonomous agents.