I recently set up three scheduled tasks in Anthropic's Claude Co-Work that run autonomously against every system I use to manage Gadoci Consulting. Gmail across three accounts, Slack, Notion, Google Calendar, and Fireflies for meeting transcripts. Each task hits all of those systems, synthesizes what it finds, and delivers the result to me via Slack or directly into Notion.
One gives me a morning briefing every weekday. One writes a journal entry about my day every night. And one reviews the entire week every Sunday and tells me what I dropped.
They've been running for a few weeks now, and the results have been genuinely useful in ways I didn't fully anticipate when I set them up. Here's what each one does and why the details matter.
The Morning Brief
Every weekday at 7am, an agent checks my inboxes, Slack threads, Fireflies transcripts from yesterday's meetings, my Notion task database for overdue items and what's due in the next three days, and today's calendar. Then it sends me a Slack DM.
I made a specific choice with this one: no bullet points. The brief is written as a short narrative, a few paragraphs of prose that connect items together. If I have a client meeting at 1pm and an overdue task for that same client, it ties them together. If yesterday's meeting transcript captured a commitment with no corresponding task, it calls that out. If a Slack thread went quiet but had an open question directed at me, it surfaces it.
The narrative format turned out to matter more than I expected. Reading a few connected paragraphs is just more engaging than scanning a list. I actually remember what was in the brief later in the day because the format makes a case for why things are important rather than just presenting them. It even works in some motivational framing, which sounds like it could be cheesy, but it's specific to what's actually on my plate, so it lands. I find myself referencing things from the brief in conversations hours later, not because I went back and looked at it, but because the narrative made it stick.
When you force the output into prose, the AI has to reason about connections and priority instead of just extracting and listing. A list of seven overdue tasks presents them as equally weighted. A narrative tells you which two matter today and why.
The Journal
The journal entry runs at 4am, capturing the previous day. It reviews the same systems but writes something completely different: a first-person reflection, written as if I'm looking back on building this company.
Each entry gets a title that captures the day's essence. Not "March 18 Journal" but "The Day the Plugin Landed." It tags a mood. It writes in my voice, as a memoir entry. If a client meeting had real energy, it describes that. If something was frustrating, it names it. If there was a quiet win, it's in there.
The thing I didn't expect is how much I enjoy just reading these. My days are packed. Calls, Slack threads, emails, tasks, context-switching from one client to another. By the end of the day, I've already half-forgotten the morning. The journal captures all of it, and reading yesterday's entry reminds me how much actually happened. There's something grounding about seeing your own day reflected back to you with the texture intact, not just what you did, but what it felt like to do it.
These go directly into a Notion database with structured fields for achievements, challenges, and learnings. Over time, that's becoming a searchable archive of building Gadoci Consulting. When I need to write a case study or remember how a client engagement actually unfolded, the entries are there with real names, real conversations, and real turning points.
The Weekly Review
The Sunday review pulls every journal entry from the past seven days, every completed and overdue task from Notion, emails where I committed to follow-ups, Slack conversations where I made promises, and Fireflies transcripts with action items and decisions.
Then it cross-references all of it: what I said I would do versus what actually got done.
The output opens with a short theme for the week, then gets straight to the gaps. Commitments made in meetings that don't have a completed task. Prospects discussed in calls that never got a follow-up. Ideas mentioned in journal entries that never became tasks. Emails that needed responses but went quiet.
A recent review flagged five specific things I'd missed. A prospect follow-up that was six days stale. Deliverables I'd acknowledged as overdue in my own journal but hadn't addressed. Three deals from a partner call that existed only in a Fireflies transcript, with no tasks created. A contract extension that my team reviewed in Slack but never confirmed externally. And two commitments from a Friday call that hadn't turned into action items.
None of those would have shown up in any single tool. The overdue follow-up wasn't a Notion task because it never got created. The unconfirmed contract looked handled in Slack but hadn't actually been sent. The partner deals lived in a meeting transcript and nowhere else. The only way to find those gaps was to read across every system and compare what was discussed against what was done.
Why Narrative Matters for All Three
I chose narrative output for every one of these tasks, and it's the decision that makes the whole setup work as well as it does.
Most AI productivity tools default to bullet points. It makes sense as a starting point, but lists strip context. They present items as discrete and equally weighted. You still have to figure out which ones matter, how they relate, and what to do first.
Narrative is more fun to read. That's a simple point, but it matters. I actually read these outputs every day because they're written in a way that holds my attention. A bullet list of six emails and four overdue tasks is something I'd skim and close. A short story about my day, or a briefing that connects my calendar to my task list to something that came up in yesterday's meeting, is something I engage with. The information sticks because the format carries it.
How I Set This Up
All three of these run on Claude Co-Work's Scheduled Tasks feature. You write a detailed prompt, connect it to your tools via MCP (Model Context Protocol) integrations, set a cron schedule, and it runs autonomously. I have MCPs connected for Gmail, Slack, Notion, Google Calendar, and Fireflies. Each task is a prompt of around 500 words that tells the agent what to check, how to reason about what it finds, and how to format and deliver the output.
The specificity of the prompts is what makes the output useful. "Summarize my day" produces generic results. "Check these six systems, connect items that relate to the same client, skip categories with nothing notable, write it as a short narrative in this tone, and send it as a Slack DM" produces something I actually use.
The setup took one session. The ongoing investment is zero. These run every morning, every night, and every Sunday without any input from me.
What I'd Pass Along
If you have access to Claude Co-Work and MCP integrations, this is buildable today. The things worth knowing going in: connect as many systems as you can, because the value comes from cross-referencing sources that don't normally talk to each other. Be specific in your prompts about what you want and what you don't want. And try narrative output instead of lists. You'll probably be surprised by how much better it is to read and how much more you retain from it.
I wrote an article a while back called "Claude Is Almost an Automation Platform. One Feature Away." The missing piece I identified was scheduled jobs: the ability for Claude to run workflows on a timer without a human starting the conversation. Scheduled Tasks is that feature. What I'm describing in this article isn't a chatbot interaction. It's autonomous business automation, running on a schedule, synthesizing across systems, and delivering actionable output. Claude Co-Work is quietly becoming the workflow platform I argued it was one feature away from being.