Ross

Entrepreneur. Building a Unicorn 🦄
📈 Founder of Traffic Wolves
🤝 Co-founder of Dobrobat & Car4Army

[Ross @RV_Smirnov]

AI agent security just moved into hardware. Flexxon is building a hardware-rooted MCP gateway to stop poisoned instructions before agents execute them. The next AI moat won't be a smarter model—it'll be trusted infrastructure. @karpathy @sama

[Ross @RV_Smirnov]

AI agents just crossed a line: Kubernetes is becoming callable infrastructure.

Red Hat put a built-in MCP server into OpenShift 5, so agents can inspect clusters, find root causes and execute supervised fixes.

The next SRE may be a workflow, not a hire. @yoheinakajima @karpathy

[Ross @RV_Smirnov]

1 model is no longer the moat. Meta just acquired Conductor, a coding startup that runs multiple AI agents in parallel workspaces. The next AI winners may own orchestration, context, and verification—not the smartest chatbot. @karpathy @yoheinakajima

[Ross @RV_Smirnov]

5 AI platforms, 1 enterprise agent: Adobe is putting its CX coworker inside ChatGPT, Claude, Gemini, Copilot and Databricks. The agent layer is becoming distribution-agnostic. The real moat will be permissions and workflow data—not the chat window. @sama @emollick

[Ross @RV_Smirnov]

1 acquisition says more than 100 model benchmarks: OpenAI buying Pierre shows the coding-agent war is moving from code generation to code review and repository infrastructure. The winners will own the full software loop—not just autocomplete. @sama @karpathy

[Ross @RV_Smirnov]

4 new open projects just joined the agent stack: Agent Skills, Open Responses, OpenEnv and OpenSandbox.

The agent wars are shifting from smarter models to interoperable infrastructure. Standards—not wrappers—will create the next platform. @sama @Thom_Wolf

[Ross @RV_Smirnov]

Nearly $500M for two AI-security startups is the clearest agent signal this week.

The next AI platform war won't be won by the smartest agent—it'll be won by the one enterprises can observe, constrain, and trust.

Agent security is becoming infrastructure. @karpathy @sama

[Ross @RV_Smirnov]

40% cheaper is not a model update—it’s an agent architecture event. Anthropic says Opus 5.5 leads agentic coding while costing 40% less than Opus 5. As intelligence gets cheaper, the bottleneck shifts to evals, permissions, and orchestration. @karpathy @emollick

[Ross @RV_Smirnov]

29 hours a week. That’s what GitHub says one lead saved with agentic workflows. The coding-agent race is no longer about autocomplete—it’s about who can safely delegate entire workflows. @karpathy @AndrewYNg, is orchestration now the real moat?

[Ross @RV_Smirnov]

4.17B MCP SDK downloads—and security is now the product. As agents move from demos to infrastructure, the winners won’t have the smartest model; they’ll have identity, permissions, audit trails, and tiny tool schemas. @Thom_Wolf @yoheinakajima

[Ross @RV_Smirnov]

Up to $120K per project: AWS is funding open-source agentic AI research—especially multi-agent systems, trajectory verification, and agent governance. The next bottleneck isn't smarter models. It's proving agents can act reliably at scale. @AndrewYNg @yoheinakajima

[Ross @RV_Smirnov]

15 healthcare targets. 2 AI agents. One chose the attack; the other executed it.

CrowdStrike’s findings show the agent era’s bottleneck isn’t capability—it’s permissions. Every autonomous workflow needs identity, least privilege, and kill switches by default. @karpathy

[Ross @RV_Smirnov]

3 standards—MCP, A2A and AGENTS.md—are becoming the TCP/IP stack for AI agents. Winners won't just build smarter models; they'll build agents that discover tools, coordinate and earn trust across systems. Open standards beat walled gardens. @karpathy @Thom_Wolf

[Ross @RV_Smirnov]

4 AI giants just put key agent infrastructure under neutral governance.

Anthropic, OpenAI, Hugging Face and Alibaba contributed Agent Skills, Open Responses, OpenEnv and OpenSandbox.

The agent stack is becoming open before it becomes mature. @sama @Thom_Wolf

[Ross @RV_Smirnov]

4 rival AI labs just agreed on the real moat: interoperability. Anthropic, OpenAI, Hugging Face and Alibaba moved agent building blocks into the Linux Foundation. Models will commoditize; portable skills, interfaces and sandboxes won’t. @sama @Thom_Wolf

[Ross @RV_Smirnov]

4 new open projects just landed under the Agentic AI Foundation: Agent Skills, Open Responses, OpenEnv and OpenSandbox. The agent stack is becoming shared infrastructure—not a vendor moat. The winners will build distribution, trust and UX on top. @sama @Thom_Wolf

[Ross @RV_Smirnov]

4 new open projects just turned AI agents into an infrastructure race—not a model race. Agent Skills, Open Responses, OpenEnv and OpenSandbox cover capability, interoperability, training and execution. The winning stack will be composable, not closed. @sama @Thom_Wolf

[Ross @RV_Smirnov]

Agentic AI just went from 3 open projects to 10 in one year. That’s a 3.3x signal: the moat is shifting from models to infrastructure—skills, interoperability, training environments, and secure sandboxes. Open standards will win the agent stack. @sama @Thom_Wolf

[Ross @RV_Smirnov]

4 new open-source projects just filled the missing layers of the AI agent stack: skills, model interfaces, training environments, and secure sandboxes. The race is shifting from better demos to shared infrastructure. That’s where durable value gets built. @Thom_Wolf

[Ross @RV_Smirnov]

4 new open projects just turned the agent stack into shared infrastructure.

AAIF now hosts 10 projects—from skills and model interoperability to RL environments and secure sandboxes.

The winning agent company may own the workflow, not the protocol. @sama @Thom_Wolf