Valentina Halasi
Deploying
AI Agents
at Scale
The loop
Prompt
→
n rounds
LLM
Tool call
Result
→
Answer ✓
Two flavors
Workflow
Fixed steps
Fetch data
Parse
Summarize
Report
Agent
Figures it out
LLM
Action
Result
decides
Real systems
Blend both
HTTP in
Parse
+ auth
Agent
Think
Act
Observe
Render
to JSON
HTTP out
Multi-agent
Pull Request opened
Orchestrator
READS PR · PLANS TASKS · SPAWNS AGENTS
PARALLEL
Code Analyzer
📁 reads diff
bugs · style · logic
Test Runner
⚡ runs test suite
coverage · regressions
Sec Scanner
🔒 checks CVEs
deps · secrets · vulns
Reviewer Agent
synthesises all findings · writes PR comment
✓ Review comment posted
Prototype
↓
Production
8
problems to solve
↓ click a problem to see which components solve it
crew ai amp
langsmith
mastra cloud
crewship
Thank you!
Questions?
x.com/valentinahalasi
linkedin.com/in/valentinahalasi
================================================================ -->
crew ai amp
langsmith
mastra cloud
crewship