Samuel Kelemen
Principal Security Engineer at Lovable
Samuel Kelemen
Principal Security Engineer at Lovable
Samuel Kelemen is a Principal Security Engineer at Lovable, where he builds infrastructure to secure AI-generated applications at scale, and foundational security primitives.
His work spans identity and access, automated remediation, across both the platform, and apps created on it.
Upcoming conference sessions featuring Samuel Kelemen
Disposable Code and Other Permanent Decisions
Working Effectively with Agentic Engineering
We already knew code was a liability. Now we can generate it faster.
When producing an implementation becomes dramatically cheaper, what should change about software engineering? We can build tools we will use once, turn more ideas into experiments, and compare eight implementations instead of arguing about which one to write. We can also generate legacy code before lunch: software nobody understands, shaped by decisions that disappeared with an agent’s context window.
This talk explores how to take advantage of cheap implementation without making everyone else absorb its costs. We will examine API and module boundaries as the primary unit of review: the behavior other components rely on, the effects an implementation may have, and the evidence that it keeps its promises. Behind those boundaries, implementations can be generated, delegated, optimized, or discarded while shared commitments remain explicit.
We will also explore the engineering we always wanted to do but rarely had the time or budget for: proving important properties, building stronger testing and simulation tools, and using agents to investigate profiles and evaluate performance improvements.
Along the way, we will encounter humans employed as clipboards, individually productive agents competing over shared state, and a disposable prototype that somehow acquired an on-call rotation.
The goal is to explore more possibilities, make better decisions, and retain software we can confidently own. Make the implementation disposable. Keep the contracts, the evidence, and the responsibility.
Get conference pass
Disposable Code and Other Permanent Decisions
Working Effectively with Agentic Engineering
We already knew code was a liability. Now we can generate it faster.
When producing an implementation becomes dramatically cheaper, what should change about software engineering? We can build tools we will use once, turn more ideas into experiments, and compare eight implementations instead of arguing about which one to write. We can also generate legacy code before lunch: software nobody understands, shaped by decisions that disappeared with an agent’s context window.
This talk explores how to take advantage of cheap implementation without making everyone else absorb its costs. We will examine API and module boundaries as the primary unit of review: the behavior other components rely on, the effects an implementation may have, and the evidence that it keeps its promises. Behind those boundaries, implementations can be generated, delegated, optimized, or discarded while shared commitments remain explicit.
We will also explore the engineering we always wanted to do but rarely had the time or budget for: proving important properties, building stronger testing and simulation tools, and using agents to investigate profiles and evaluate performance improvements.
Along the way, we will encounter humans employed as clipboards, individually productive agents competing over shared state, and a disposable prototype that somehow acquired an on-call rotation.
The goal is to explore more possibilities, make better decisions, and retain software we can confidently own. Make the implementation disposable. Keep the contracts, the evidence, and the responsibility.
Get conference pass
Browse all experts
Here