Team AI Directives: From Agentic SDLC Methodology to Production Evals (Online)

Date: Monday, May 25th, 2026
Time: 17:00-18:30
Location: Online Webinar
*The webinar will be held in Hebrew

On the Agenda
17:00-17:05 Opening Words
A short introduction and overview of what to expect.

17:05-17:45 Team AI Directives as Part of the Agentic SDLC
Avishay Maor, AI Lead, Fullstack Architect and Group Lead, Tikal

As more teams adopt AI into their software development lifecycle, one challenge becomes hard to ignore: the behavioral instructions behind these systems are often scattered across prompts, tools, and local team habits, with no shared structure for how they are created, maintained, or improved.
In this session, Avishay will present Team AI Directives as part of Tikal’s Agentic SDLC methodology. He will introduce the broader engineering mindset behind the methodology and explain why directives should be treated as a shared, versioned engineering asset rather than an ad hoc layer of prompting.
The talk will also cover the role of Level Up in this flow, and how it supports the lifecycle of building, organizing, and evolving directives in a more structured and scalable way. The goal is to move from scattered local practices toward a repeatable engineering approach for working with AI across teams.

17:45-18:30 From Prompts and Directives to Evals: Keeping AI Behavior Reliable in Production
Keren Finkelstein, Backend Tech Lead & Group Lead, Tikal

Building directives is only the first step. Once they shape production workflows, teams need a reliable way to verify they still behave as intended, especially after updates.
In this session, Keren will demonstrate the evaluation layer that makes directive systems maintainable. She will show why systematic evaluation is essential, not optional, and how teams can validate changes, protect quality, and prevent behavior drift across versions.
The session covers practical approaches to directive evaluation as part of ongoing engineering work: identifying regressions, measuring quality with automated graders, and creating feedback loops that keep systems reliable under real-world conditions.

Please Note
*By registering for this event, you agree to receive emails from Tikal.

See you online