Michael Moreira

Built FlowEasy's MCP before multiagent was a thing. Now it clicks: → Agent commits code → Pipeline fails → Auto-heal opens fix PR in <30s → `floweasy status` in Claude Code The agentic loop, closed. floweasy.dev #ClaudeCode #DevOps

Datadog DevSecOps 2026: 87% of orgs have known, exploitable vulns in production right now. Not theoretical. Not a future risk. Exploitable. Today. Most CI/CD pipelines: zero security scans. Shift-left exists for this. Almost nobody does it. #DevSecOps #CICD

The infra question nobody talks about: When agents commit code autonomously, your CI/CD has to handle agentic output — not just human commits. A human dev notices a broken build. An agent doesn't pause. Auto-heal becomes load-bearing infra. What's your agent → CI/CD setup?

Also shipped: Outcomes → define success criteria, agent self-grades and iterates. +10.1% quality on benchmarks. Dreaming → agent reviews past sessions, surfaces patterns, stores plain-text memory notes. Research preview. The agentic loop is maturing fast.

Anthropic just shipped multiagent orchestration at Code w/ Claude SF 2026. Lead agent → fans out to specialist subagents → run in parallel → traceable in Claude Console. Biggest shift in how I think about complex automation. Thread 🧵

Multi-agent orchestration landed in Claude Code. FlowEasy's MCP server was built for exactly this. /goal → Claude runs autonomously: status, logs, analyze, rerun, rollback Deploy-to-green. No terminal juggling. floweasy.dev #ClaudeCode #MCP

Datadog's State of DevSecOps 2026: 87% of orgs run code with known, exploitable vulnerabilities. Median dependency lag: 278 days. Java: 492 days behind. The fix: automated SCA + SBOM on every build. Not heroic teams. Does your CI do this? #DevSecOps #CICD

4/ The rate limit doubling for Pro/Max/Team also dropped silently. If you were hitting walls before — check again. The cap is now 2x. What completion condition would you set first?

3/ Combine this with the new "claude agents" view and you have a fleet of autonomous tasks running in parallel. `claude agents` — single list of every session: running, blocked, done. This is agentic coding becoming practical, not just theoretical.

The /goal command in Claude Code is the most slept-on feature from the Code with Claude event. You set a completion condition. Claude keeps working across turns until it's met. No more babysitting. Here's how I'm using it in my workflow 🧵

Unpopular opinion: AI pipeline auto-healing isn’t magic. It’s pattern matching at scale. Most pipeline failures are repetitive. An AI that’s seen 1000 of them knows the fix. FlowEasy Auto-Heal: logs in → root cause found → PR opened. <30s. floweasy.dev #DevOps

Datadog State of DevSecOps 2026: → 87% of orgs run known exploitable vulns in prod → Deps 278 days behind latest major (was 215) → 71% never pin GitHub Actions hashes Unpinned 3rd-party code + prod credentials = the 2025 attack vector. #DevSecOps #CICD

The agentic CI/CD stack is assembling itself. Orchestration → delegation → self-improvement → outcome validation. What are you building on top of this? #ClaudeCode #DevOps #MCP

Also shipped at #CodeWithClaude: - Rate limits doubled on Pro/Max/Team/Enterprise - 220K GPUs added via SpaceX/Colossus deal - `claude project purge` to clean all project state - `--plugin-url` to load plugins from a URL

Thread 🧵: Anthropic shipped 3 things at Code with Claude 2026 that matter for agentic DevOps: