Automation Geeks Weekly: AI is getting a job description

What happened: Microsoft introduced Copilot Cowork, which moves Copilot from answering questions to planning and carrying out multi-step work across Outlook, Teams, Excel, and files.

What it is: Think of it as a background operator. You give it an outcome, like clean up my calendar, prepare a client briefing, or research this company, and it builds a plan, pulls the right context from your Microsoft tools, and works through the task with checkpoints and approvals.

What it means: This is the clearest sign yet that office AI is becoming operational, not just conversational. For consultants, agencies, and coaches, that means admin-heavy work like prep packs, follow-ups, research, and scheduling is getting much closer to delegated work rather than assisted work.

Try this: Pick one repeating task that eats 30 to 60 minutes a week. Client meeting prep is a good candidate. Write a simple playbook with the inputs, the output, and the approval point. Even if you are not on Microsoft, that exercise shows you exactly what an AI worker should own and what should still stay with you.

What happened: Google’s February Workspace Drop added Gemini-powered time suggestions in Calendar, proactive proofreading in Gmail, audio summaries in Docs, and near real-time speech translation in Meet.

What it is: These are small, practical AI helpers inside everyday work. Instead of opening a separate AI tool, the AI shows up where the work already happens, helping you schedule, write, catch up on long documents, and collaborate across languages.

What it means: This is how AI actually spreads in small businesses. Not through one huge transformation project, but through dozens of tiny friction removers stacked together. The businesses that get value first will be the ones that notice and adopt these boring little wins before their competitors do.

Try this: Run a one-week experiment with your team. Choose one built-in AI feature in the tools you already pay for and make everyone use it for seven days. Then ask one question: did this save real time, reduce mental load, or improve quality? If the answer is yes, make it standard. If not, ignore the hype and move on.

What happened: After last week’s Custom Agents launch, Notion followed up with a look at how its own team uses them and how it built security into them. The examples are more revealing than the launch post.

What it is: Notion’s agents are being used to answer repeat questions in Slack, route product feedback, triage issues, and handle office or style-guide questions. The important part is the permission model. Agents start with almost no access, then get narrow permissions added so they can help without causing chaos.

What it means: This is the new pattern to copy. The winning AI workflows are not magical. They are specific, repetitive, and permissioned. If your AI idea needs broad access to everything on day one, it is probably not ready. Good automation starts narrow, proves trust, then expands.

Try this: Make a list of questions your team answers over and over in Slack, email, or WhatsApp. Pick one category, like pricing, onboarding, or meeting logistics. Build a tiny answer system first with approved source material and clear boundaries. That is often the fastest route to your first genuinely useful AI workflow.