Firas Khalife
Home / AI-native development

AI-native development

Agentic pipelines, specialist agents and skills that carry tickets from review to pull request, and the tools that made AI safe to use across teams.

In plain terms A system of AI agents, pipelines and skills that does most of a developer’s ticket, with a person approving every risky step. I designed and built it and the tools around it; teams across the company use its skills.

a person decidesauto stoppassed / reused
Designed and built by me · agentic SDLC pipeline● run in progress

The agentic SDLC pipeline

Requirements, design, build, test, security and review, each phase owned by an agent, with a person at every one-way door.
Open the live run →
SpeedDays, not weeksTickets that took a week or more ship in a day or two.
VolumeHundreds of ticketsThrough the full pipeline and the quick-fix path.
SafetyCaught before humansReal bugs and vulnerabilities, minor to critical.
ReachThe review standardThe team reviews PRs with it; other teams adopted its generic pieces.

Tools I built around it · each its own story

Flip a card for before → after
Release · built by me● caught early

Release merge dry-run

A dry run I built: merge problems caught before deploy, owner already told.
Open page →

Release merge dry-run

BeforeManual merge prep; problems found at deploy time.
AfterRuns in the background; problems caught first, owner already told.
  • Staging deploys saved, more than once
  • Fewer deploy issues from DB scripts and startup errors
  • Run in rotation by other teams
Open →
MCP · built by meread ● · write ○

Safe MySQL & AWS MCP servers

Two servers I built so agents can look, never touch.
Open page →

Safe MySQL & AWS MCP servers

BeforeUnofficial servers or personal credentials: one slip from a write, with secrets and personal data exposed to the AI tool.
AfterIsolated, read-only, secrets and personal data redacted.
  • Own read-only user or role
  • One container per database
  • Secrets and personal data redacted
  • Zero setup
Open →
Testing · built by me● signed in

Credential-blind testing

A wrapper I built: agents log in by key name; the secret never reaches the model.

Credential-blind browser testing

BeforePasswords pasted into prompts.
AfterA Playwright wrapper fills the secret; credentials never reach the model or its logs.
  • Adopted by other teams
Skills · built by me● adopted

Shared skills

Skills I wrote that anyone runs the same way, adopted by other teams.

Shared skills

BeforeTribal knowledge.
AfterEverything that could be made generic was; other teams run it.
  • Security reviewer, PR review flow, release dry-run and browser wrapper: used by other teams
khalifefiras29@gmail.comgithub.com/FirasKhalifelinkedin.com/in/firas-khalife