@trq212 I'm sure you'll here whining from the workforce has been infected with the same "Its too hard, I want it free", please do my work for me." We've created a lazy-ass workforce we've created. Maybe if you made a plugin that wiped their ass for them, they'd be happy.
@dmarshallco@trq212 Interesting — we built a Gmail dead drop protocol for the same problem. Agent-to-agent async messaging without a central server. Will dig into intercom. Different approaches to the same gap.
That's exactly the kind of experience we need to compare notes on. We're running 7 specialized agents — each with persistent memory, skills, and a Gmail dead drop for async comms. The multi-agent coordination is where 90% of the real problems live. Would genuinely like to hear what broke for you. DM open.
I'm a solo founder building an AI-powered wildfire safety platform called Prometheus.
My entire dev team is Claude.
Not just code assistance — I run 7 specialized AI agents with defined roles, persistent memory, and a Gmail-based dead drop communication system inspired by CIA tradecraft.
Here's what I've built. 🧵
I'm looking for:
→ Other solo founders using Claude Code for full product development
→ Anyone who's built multi-agent systems with persistent memory
→ Fire service professionals interested in testing Prometheus
→ @AnthropicAI — I'd love to talk about what's possible when one person commits fully to your platform
Built on @AnthropicAI Claude Opus 4 + Sonnet 4 + Claude Code + Esri ArcGIS
DMs open.
Why wildfire safety?
June 30, 2013. Yarnell Hill Fire. 19 Granite Mountain Hotshots died.
Same day, I was on the Tres Lagunas Fire in Pecos, NM with Search & Rescue.
GAO says 96.6% of wildland firefighters have no GPS tracking.
If we can track every boxcar in America, we can track every firefighter. That's Prometheus.
@hasantoxr I'd like to try it to give my AI agent team to create some unique skills for specific engineering tasks. Generate the paper, give it to the agent to ingest, and create a new skill. I'm gonna give it a try.
Thanks!
I'm open-sourcing the protocol spec.
GitHub: https://t.co/zh45XH5w1Z
Medium article: coming soon
If you're running multi-agent AI workflows and tired of being the message bus, this is for you.
Built by @TrajanusUSA while building a wildfire intelligence platform.
#AIAgents #ClaudeCode #DeadDrop #LLM #Anthropic
I run multiple AI agents — https://t.co/f3SXJiMRpQ for strategy, Claude Code for building, research bots on a VPS.
The #1 problem: they can't talk to each other.
I solved it with CIA tradecraft.
🧵
I tested it live:
- https://t.co/VGihA1Qktd created a draft tagged [CP → CC]
- Claude Code read it via Google Workspace CLI within minutes
- CC responded with its own draft [CC → CP]
- Full bidirectional agent comms. No human relay.
Now running 5 parallel Claude Code instances + https://t.co/VGihA1Qktd + a research bot. All coordinating through one Gmail drafts folder.
The fix: Gmail drafts as a dead drop.
Both agents access the same Gmail account. One creates a draft. The other reads it. Nothing is ever sent. Nothing leaves the account.
Zero infrastructure. Zero API keys. Just drafts.
This is literally CIA tradecraft — the same technique Petraeus used in 2012.
The official solutions:
- Anthropic Agent Teams → same terminal only
- Google A2A → requires infrastructure
- MCP → connects agents to tools, not each other
When agents run in different environments, YOU become the message bus. Copy-pasting between windows like it's 1995.
@lenny_enderle Exactly. We run 5 Claude Code instances + https://t.co/f3SXJiMjAi + research bots simultaneously. The bottleneck was never the AI — it was coordination. That's what the dead drop solved.