Tool of the Week: Connectors (a.k.a. MCP)
For two years, "using AI in your business" mostly meant copy-paste. You'd grab an email, paste it into ChatGPT, get a reply, paste it back. The AI could advise, but it couldn't do anything.
That's the part that just changed, and most owners haven't noticed yet.
What it is: Connectors let ChatGPT or Claude plug directly into the tools where your work actually lives — Gmail, your calendar, Google Drive or Notion, your CRM, even a job board — so the AI can read and act, not just talk about it.
Why this is real and not a fad: The plumbing underneath (a standard called MCP) is no longer one company's experiment. It was handed to the Linux Foundation and is now backed by Anthropic, OpenAI, Google, Microsoft, and Amazon, with roughly 97 million downloads a month and built-in support across ChatGPT, Claude, Microsoft Copilot, Gemini, and Cursor. Translation: it's infrastructure now, the way "log in with Google" became infrastructure.
What it changes for you: Instead of "summarize this email I pasted," you say: "Check my inbox for clients waiting on a reply and draft responses." Instead of "what should I say to this lead," it's "pull this lead's history from my CRM and write the follow-up." The AI stops being a smart intern who can only give advice and starts being one who can actually touch your tools.
How to start (no code): In Claude, go to Settings then Connectors. In ChatGPT (Plus and up), it's also under Settings then Connectors. Add one tool you touch every day — Gmail or your calendar is the easiest first win — and authorize it once. That's the whole setup.
The honest caveats:
You usually have to switch a connector ON per conversation. It's not silently listening to everything — you decide when it's in the room.
Only connect tools you trust. A connector can act on your data, so treat it like handing someone a key.
Start read-only. Let it search and summarize for a week before you let it write or send anything.
Who this is for: Any owner who lives in email, a calendar, and a CRM. The first time your AI books the meeting instead of telling you how to book it, the whole thing clicks.
Quick Hits
Microsoft put "agentic coworkers" on sale. Copilot Cowork reached general availability, and Microsoft showed off "Scout" — an always-on agent with its own identity wired into Teams, Outlook, and OneDrive. Why it matters: the exact "let AI handle the busywork" pitch is now coming from the biggest incumbent in business software. That's proof the model is real — and a preview of what your customers will soon expect you to have.
OpenAI's 2025 numbers leaked. Roughly $13B in revenue against about $34B in costs — a ~$21B operating loss (and a $38.5B net loss after one-time charges) — surfaced ahead of a possible IPO. Why it matters: the AI tools you rely on are being sold to you below what they cost to run. Don't be surprised when prices climb or specific models get retired. Build on tools you can swap out, not ones you're locked into.
A general AI beat the specialists. A study in Nature found general-purpose models outperformed several purpose-built clinical AI tools. Why it matters: before you pay a monthly fee for some niche "AI for [your industry]" product, test whether a good general model and a sharp prompt do the same job for a fraction of the price. Often they do.
A useful reality check on "agents." New benchmarks show AI agents still fall short of professional-quality work on complex, multi-step tasks. Why it matters: automate the repetitive 80% — drafting, sorting, summarizing, scheduling — and keep a human on the judgment calls. Anyone selling you "fully autonomous, set it and forget it" is overselling.
Prompt of the Week: The Automation Audit
Before you buy or build anything, find out what's actually worth automating in your business. Paste this:
Help me find the best automation opportunities in my business.My inputs:What my business does: [one line]
My weekly recurring tasks: [list 5-10 things you or your team do every week]
Tools I use daily: [email, calendar, CRM, spreadsheets, etc.]
For each task, tell me:How repetitive and rule-based it is (low / medium / high)
Whether it's safe to automate now, or needs a human in the loop
What it would take — a connector, a simple workflow, or custom work
The 3 I should tackle first for the biggest time savings at the lowest risk
Be conservative. Flag anything where a mistake would cost me a customer or
money — those stay human-reviewed.The last line is doing the work. The goal isn't to automate everything — it's to find the handful of tasks that are repetitive, low-risk, and quietly eating your week.
Like what you're reading? Forward it to someone who'd get value from it. And if you're curious what AI could actually do inside your business, book a free 15-minute audit - no pitch, just a look at where you're leaving time on the table.