QCon AI New York 2025: Execution Beyond the Hype
Five practical lessons on risk, architecture, product engineering, and multiagent systems for companies operating AI at real scale.
In New York, at QCon AI, I joined one of the most technical and insightful AI conferences today.
The strongest message I took away was not about models.
It was about execution: moving beyond hype and understanding how companies operating at real scale are building advantage.
1) AI changes enterprise risk whether you use it or not
AI reshapes the risk surface of the entire business.
Leading in AI is first about leading risk, decision quality, and governance in a context where historical playbooks are often insufficient.
2) LinkedIn at global scale: the key lesson was cultural and architectural
- AI as a new engineering execution model
- A clear flow: Intent -> Plan -> Execute -> Validate -> Output
- Strong emphasis on validation, trust, and quality, not only speed
3) Meta: progressive transformation, not a “big bang”
The journey started small and scaled to hundreds.
Core learnings:
- Progressive tool integration
- Clear guardrails for AI-generated code
- Significant investment in education and human review
- AI amplifies productivity, but does not replace judgment
4) Netflix: fine-tuning treated as production product engineering
- Well-scoped use cases
- Clear success metrics
- Cost, latency, and reliability designed in from the start
5) Multiagents, memory, and guardrails
A cross-cutting theme: agents do not need “more intelligence”, they need more truth.
- Well-defined memory
- Strong semantics (ontologies and semantic layers)
- Testing, automated validation, and anti “AI slop” practices by default
It was also great to connect with Brazilian leaders from Nubank, Will Bank, Decolar, and others, all sharing the same view: this transformation is technical and strategic at the same time.
For deeper reading: