Welcome to the Automation Geeks Newsletter. We’re just getting started, so expect to see updates and changes as we go!
The best AI progress right now is not bigger brains. It is AI that can sit inside the tools you already pay for, watch for signals, and quietly do the small-but-constant jobs that keep a service business tidy.
This week: 3 things worth knowing
1) Notion launched Custom Agents (free to try until 3 May)
What happened: Notion released Custom Agents, with a public beta that is free on Business and Enterprise plans until 3 May 2026. Notion announcement
What it is: Think "mini staff members" that run 24/7. You give them a job description plus triggers (like a Slack channel message or a new database entry), and they can answer repeat questions, route requests, and compile updates across Notion and connected tools.
What it means: This is a real shift from asking AI for help to having AI handle recurring operations. For service businesses, the big win is fewer dropped balls in the messy middle: handoffs, requests, follow-ups, and status updates.
Try this: Create one agent called Ops Triage. Give it one inbox (a Slack channel or a single Notion database). Its job is to: (1) summarise what came in today, (2) tag it as Sales, Delivery, or Admin, and (3) draft the next step as a task for a human to approve. Keep it boring on purpose.
2) Outreach shipped an MCP server to connect its revenue data to AI tools
What happened: Outreach announced an official Model Context Protocol (MCP) server so its insights and signals can flow into external AI systems like Claude without custom integration work. Outreach product release
What it is: MCP is a standard "connector" for AI. In plain English, it lets an assistant pull the right context from a business system (here, Outreach) so the assistant can give answers based on real activity, not guesses or stale screenshots.
What it means: We are moving into an era of AI with receipts. If your assistant can see real pipeline activity, calls, and tasks, it can coach, draft follow-ups, and flag risks with much higher accuracy. That is the difference between helpful and harmful.
Try this: Even if you do not use Outreach, steal the pattern: pick one system of record (CRM, project tool, helpdesk). Write a one-page "data truth" rule for your team: where the real status lives, what fields must be filled, and when. Your future AI will only be as good as your present discipline.
3) Figma connected Codex to the Figma canvas using the Figma MCP server
What happened: Figma published a walkthrough showing how Codex can generate Figma design files via the Figma MCP server, and how you can move from design to working UI and back. Figma blog
What it is: A bridge between "what it should look like" (Figma) and "make it real" (code). The key idea is shared context: the AI can read design details like layout and styles, then produce a build that actually matches the design.
What it means: For agencies and consultants, this shortens the awkward gap between mockups and build. It also raises the bar. Clients will expect faster prototypes, so your advantage becomes taste, clarity, and decision-making, not clicking rectangles.
Try this: Run a 90-minute prototype sprint for your next landing page or client portal screen. Step 1: write the page goal in one sentence. Step 2: sketch one frame in Figma with real copy. Step 3: use an AI build tool to produce a working draft. Step 4: decide what to keep and what to cut before you polish anything.