Skip to content
All experiences

Get Magic (getmagic.com)

2023 – 2024 (consulting) Remote · USA

Senior AI Engineer · Magic Assistant

Building the AI co-pilot that scales Magic’s 24/7 executive-assistant service.

Timeline

2023 – 2024 (consulting)

Location

Remote · USA

Stack

4 areas · 17 tools

Scope

5 areas · 5 highlights

Overview

Magic is an on-demand executive-assistant company serving thousands of executives, founders, and operators in the US. Assistants handle research, scheduling, travel, vendor outreach, document work, and more — Magic operates as a marketplace and a managed service.

  • I worked on the AI layer that augments human assistants: a "Magic Assistant" agent that drafts, summarises, and executes on behalf of the human, plus an open-floor team-wide assistant that helps any assistant get unstuck on any task.
  • Beyond core assistant agents, I supported embedded product surfaces for Magic clients: sidepanel tool-selection UX, dynamic tools APIs, and rollout controls for enterprise-safe feature delivery.

My role

  1. Designed and shipped an enterprise Chrome extension on Plasmo with in-memory RAG (vectors in IndexedDB), semantic chunking, and aggressive pruning so the AI runs in the user’s browser context.
  2. Built MCP-style tool adapters connecting CRM, calendar, email, and internal Magic tooling so agents can take real actions, not just chat.
  3. Implemented hybrid retrieval over assistant playbooks, prior tickets, and client preferences for high-precision recall.
  4. Worked across AWS Lambda and Cloudflare Workers for low-latency edge inference and tool execution.
  5. Contributed to competitive-intelligence style assistant workflows and dynamic tool loading in sidepanel experiences.

Stack

Extension & UI

PlasmoReactTypeScriptIndexedDB

AI

OpenAIAnthropicCustom RAGMCP servers

Infra

AWS LambdaCloudflare WorkersS3DynamoDB

Integrations

CRMGoogle WorkspaceCalendarEmailInternal Magic APIs

Highlights

  • Shipped the Magic Assistant browser extension as a true co-pilot — context-aware, memory-aware, and able to take action via tool calls.
  • Built the open-floor team-wide assistant that lets any assistant ask a question and get an answer grounded in playbooks and historical tickets.
  • Engineered an in-browser RAG store with IndexedDB so sensitive context never leaves the user’s device when possible.
  • Implemented privacy-first patterns: data minimisation, scoped tool access, and full audit trails on agent actions.
  • Hardened prompt/tool safety with scoped execution and processor-level controls for high-risk assistant contexts.

Outcomes

  • Reduced research and drafting time per assistant ticket through always-on AI suggestions.
  • Standardised how Magic assistants discover and reuse playbooks at scale.
  • Enabled safer feature rollout through flag-gated tool exposure and sidepanel-level controls.
  • Created the foundation for Magic’s broader AI roadmap.

FAQ

What is Get Magic?

Magic (getmagic.com) is an executive-assistant service combining vetted human assistants with AI tooling. Clients delegate research, scheduling, travel, vendor outreach, and document work 24/7.

Why a Chrome extension instead of a web app?

Assistants already live in Gmail, Calendar, CRMs, and client tooling. An extension brings the AI to where the work happens — with page context, selection context, and direct tool actions.

Put this experience to work

A free 30-minute call to map your constraints, risks, and a practical first milestone for your team.

Also explore

Have something to build?

Tell me what you're building — start with a free call

Send a message

Founder, SolutionPlus · AI Product Engineer

SQ
Saif Qureshi
  • Berlin, Germany · Production AI agents and systems for companies and enterprises
  • Outcomes-focused delivery: measurable impact, not demos.

Contact

Available for new projects
© 2026 Made withby Saif Qureshi
React · TypeScript · Tailwind CSS