How we think about AI memory, agentic work, and what it actually takes to make your AI assistant useful.
Your AI can already connect to Gmail, Slack, GitHub, and dozens of other tools. The question isn't which ones to add — it's what ties them together. MCPs are pipes, not memory. Here's how stacking works and why a context layer changes everything.
Anthropic uses Clay, LeanData, Salesforce, Gong, Ironclad, and Slack to run its sales motion — with Claude carrying context across the workflow. Small teams need the same jobs done, but not the same six-figure stack.
You spend more time re-explaining your work to AI than actually working with it. The problem isn't your conversations. It's that your context has no home.
People aren't waiting for their company to pick an AI tool. They're choosing Claude, ChatGPT, or Gemini on their own and using it for real work. The missing piece isn't the AI. It's the infrastructure underneath it.
Notion, Obsidian, and Roam taught a generation to capture knowledge. But personal notes are optimized for a human returning to their own thoughts. AI needs something different: operational context structured around people, relationships, and commitments.
AI platforms are getting better at remembering you. But remembering your preferences is not the same as knowing your clients, commitments, relationships, and follow-ups. That difference is where business memory begins.
You spent an hour teaching your AI about your business. Then you closed the tab. Tomorrow it won't remember any of it. The problem isn't memory. It's that sessions are the wrong container for real work.