Most marketers treat AI as a faster typewriter. Throw a prompt at Claude, get 2x the copy in half the time, declare victory. That's not leverage—that's just outsourcing repetition.
Real leverage lives in systems. Agents that route work based on context. Retrieval pipelines that ground LLMs in your actual product docs so hallucination goes to zero. Evaluations that catch quality drift before it ships. A prompt library your whole team uses consistently. A content pipeline that turns a 10-minute idea into a published post in 45 minutes, not a day.
I built the blog/ai-pipeline at sohaib.gg as a proof of concept—18 essays shipped, each one following a 4-phase workflow. No magic, just systems. That's what I'm doing across customer success ops and marketing now. The tools matter less than the architecture.
Models & Chat
- Claude — Daily driver. Context window, instruction-following, reliability. Primary for everything.
- ChatGPT — Second opinion. Occasionally catches angles Claude misses. Useful for comparison prompts.
- Gemini — Long-context experiments. Testing where 2M-token windows unlock new workflows.
Image & Video
- Midjourney — Hero images, cover art. Consistent style. Fast iteration.
- Sora — Watching. Not actively using yet, but early experiments are promising.
- Canva AI — Quick social graphics, thumbnails. Accessible for non-designers on the team.
Agents & Workflows
- Claude Code — Building this page, automating code review, managing GitHub workflows. Core daily tool.
- n8n — Custom automation. Webhooks, integrations, cross-system orchestration.
- Custom GPTs — Team workflows. Prompt library capture. Evangelized adoption across APAC peers.
Analytics & Data
- Exploring Hex — Natural-language analytics. Democratizing data access for non-analysts.
- June — Event tracking, cohort analysis. Still in evaluation phase.
- Hyperarc — Collaborative analytics platform. Watching the space.
Marketing-Adjacent (Watching)
- Mutiny — Personalization. Web experience orchestration. Early-stage interest.
- Clay — Enrichment and outreach. Building AI-native GTM workflows.
- Common Room — Community intelligence. Assembling intent signals from fragmented channels.
The blog/ai-pipeline is a working artifact—a 4-phase content creation workflow that I've shipped 18 essays through. Read the README for the quick start, or dive into WORKFLOW.md for the step-by-step.
It's not theoretical. It works. 18 essays at sohaib.gg prove it.
- May 2026 — MCP servers and tool-use patterns. Connecting LLMs to live systems (GitHub, Slack, analytics) to enable agentic flows without hallucination.
- May 2026 — Evaluating Hex for natural-language analytics. Can we make data self-serve for the team?
- May 2026 — RAG at scale. Moving beyond vector databases into hybrid retrieval for long-context system design.
- May 2026 — Prompt engineering rigor. Building eval rubrics that catch quality drift in content pipelines before humans review.