From 200+ AI agent batches to a production-grade AI CMS & MDD (Memory-Driven Development) — excited to connect with fellow builders!

Hey everyone! Excited to join Lounge.ai and connect with Ryan and all the builders here.

I'm Otávio Serra, a Systems Architect and full-stack engineer with 20+ years in the trenches. Over the past year, my daily workflow shifted almost entirely to orchestrating multi-agent coding teams (Architect, Executor, and Reviewer topologies) to build and ship production software.

Along the way, we hit two massive walls that I know many of you are also facing—and ended up building two core layers that are now synced in the Lounge member repos:

Solving "Agent Amnesia" at scale → Memory-Driven Development (MDD)

When our AI coding agents crossed 200+ implementation batches, standard context files broke down. Agents started forgetting past architectural decisions, overwriting each other's work, or blowing up token budgets.

We evolved from Spec-Driven Development into MDD (conn2flow-ai-workspace):

Strict separation between Chief Engineer Memory (read-only for agents) and Execution Memory (auto-compacted at 75 KB).
A hard 10-active-item cap on every control index (older items auto-archive with 1-line YAML summaries).
A Python CLI (mdd) + a 9-tool MCP server + VS Code extension that syncs 45+ canonical agent skills across 8 repositories.

Turning AI-assisted code into sovereign, production-ready web & SaaS infrastructure → Conn2Flow 3.0

We just shipped the 3.0 series of Conn2Flow (conn2flow on GitHub / conn2flow.com)—an open-core, AI-native modular CMS and SaaS orchestration platform (PHP 8.2–8.5, Tailwind v4, canonical REST CRUD API, built-in AI modes, and 1,880+ automated tests):

Over 1,048 indexed documentation and content pages powering an in-panel AI Assistant.
Native multi-currency Stripe subscriptions + autonomous Cloud hosting provisioning (Host Manager).
Next on our roadmap: a "Vibe Coding + Senior Technical Reviewers" Marketplace—where founders and non-technical creators can build rapidly with AI agents inside Conn2Flow and hire verified senior engineers on-demand to audit, harden, and sign off for production.

What I'm looking for here in Lounge.ai:

Fellow AI & MCP builders who want to test or collaborate on our open-source MDD framework (conn2flow-ai-workspace).
Agencies, indie founders, or SaaS operators looking for a fast, self-hostable (or managed Cloud) AI-native CMS/portal engine without vendor lock-in.
Feedback from anyone building in the AI agent memory or vibe-coding review space!

All three of our public repos (conn2flow, conn2flow-ai-workspace, and react-mapdraw-navigator) are live in the Member Repos directory, and you can check out the platform at conn2flow.com/en/

What's the hardest bottleneck you're currently hitting when moving AI-generated code into real production? Would love to swap notes!