I use workflow builders like Gumloop for a lot of my AI agents, but lately I’ve also been using scheduled Claude Cowork tasks more for a handful of things. Backwards, I know, but I like the control the workflow builders give me over the text instructions of scheduled Claude or ChatGPT actions. These are the seven that are actually running right now and doing useful work for me.
1. My daily scroll
This has so many purposes and is probably the single most useful way I use AI. The original purpose was to get me to stop scrolling Reddit for 45 minutes in the morning before I got out of bed. It eventually grew, as I added more and more instructions, into a single gigantic Notion page that preps me for the day.
It:
- Summarizes every upcoming meeting.
- Tells me the email follow-ups I need to do that morning.
- Checks my web traffic and Google Search Console.
- Does a bunch of SEO and AEO checks, making sure new pages I’m adding are well optimized and getting indexed by Google.
- Scans through all of my YouTube subscriptions and suggests videos, so I don’t have to get caught up in the algorithm.
- Gives me a quick brief on world, U.S., Georgia, and local news.
- Gives me one useful thing to learn, one tool to try for the day, and a bunch of potential content about my interests.
The idea is to give me very specific, tailored stuff to look into, all in one output, so I don’t scroll endlessly. It runs at 7:00 in the morning and is ready as soon as I’m getting started. That way if I’m stuck scrolling in bed, at least it’s semi-productive.
2. My daily follow-up
The flip side is my daily follow-up. Again, the output is a Notion page. I find it useful for these outputs to go into Notion so they’re available for AI to look at in the future, both within Notion and Claude, and anywhere else I’m pointing at Notion.
This one goes through all of my transcripts, Google Calendar, email, HubSpot, etc. for the day and:
- Pulls out any follow-up tasks I need to do and adds them to HubSpot.
- Updates deals and contact records in HubSpot when necessary.
- Mines my transcripts for potential LinkedIn posts, blog posts, and other content ideas.
This is honestly where I get the vast majority of my content: my own quotes from meetings I have with other people. If you look at my newsletter, the theme is literally one hot take, two tools (and one to cancel), and three best practices, all from quotes from my calls that week.
3. Better meeting notes
None of the transcription tools I’m using (right now a combination of Notion and Wispr Flow) does a particularly good job generating summaries out of the box.
I have this one running at the end of the day. It goes through all of my transcripts and generates more custom notes based on what I’m looking for, then tags them with metadata so it’s easier for me to find the transcripts I want later.
As always these are dropped into Notion. If you’re noticing a theme that all of my AI output goes into Notion, that’s exactly the idea. My second brain in Notion is gigantic and extremely relevant to all of the tools that I’m using. I keep my Notion organized with Tiago Forte’s PARA method.
4. Cold email from start to finish
I just started this one relatively recently. If you know me at all, you know I’m not a big fan of cold email. I don’t think it works all that well, but this is working for me to an extent.
I’m basically letting Claude run cold email for me from start to finish. The only thing I do is jump in when I get a reply. I have it connected to tools like Woodpecker, Prospeo, Clay, and a number of others to pull data and fill in more information. Ultimately, it enrolls new potential leads in my Woodpecker campaigns every day.
We’ll see how it goes. I’m not sure it’s going to be worth the token cost, tool cost, data stack, and email infrastructure costs. But as long as it’s booking a couple of meetings a month, it probably pulls its weight, so I’m going to let it keep going.
5. Finding warm LinkedIn leads
This is the LinkedIn version of the same thing. It uses the Chrome browser extension for Claude to go into LinkedIn and find anyone who is engaging with my posts or viewing my profile, then does an ICP check on them. It gives me a Notion database to review so I can decide whether I want to enroll them in their own cold email campaign.
I’m a lot more selective with this one because I don’t want to enroll just anyone interacting with me on LinkedIn in cold email. I’m making sure only the particularly relevant people get enrolled and I’m writing the emails myself.
The tool stack is similar to the other Claude email campaign, with one step where I’m in the loop to choose who gets added and who doesn’t.
6. Finding LinkedIn conversations worth joining
This one is another LinkedIn task, but it’s about finding people making posts on topics that are relevant to me so I can jump in and comment when it makes sense. I’m trying to expand my LinkedIn reach a bit, and it can potentially feed new leads into the prior campaign if the engagement is reciprocal.
It’s fairly similar to the Reddit process I built in Gumloop. The Reddit scraper in Gumloop seems to work better than what I can get Claude to do. It scans a bunch of subreddits and tells me when people are talking about problems I may have solved in the past, then points me to the ones where I should potentially comment or send a DM. This scheduled task is effectively the LinkedIn version of that.
On the Gumloop Reddit workflow, I have that template, and I’ve shared it with a number of people. Feel free to reach out to me and I’ll share it with you and walk you through how to use it.
7. Weekly HubSpot contact cleanup
This one only runs weekly. It scans the people who were added as contacts in HubSpot that week and identifies any who should be deleted or marked as non-marketing contacts (doesn’t count towards my email limits).
Theoretically, I can do this with HubSpot workflows, but I find it more useful to have AI scan the contact and do some digging into whether I’ve had previous engagement with them. That gives it more context for deciding whether I should keep them in HubSpot or remove them.
A few of these are still experiments, especially the cold email ones, so we’ll see which ones stick. What usually happens with these Cowork tasks is that they reach a level of specificity that makes me want to rebuild them in a workflow builder where I can control the actions a bit more (check out my post on how I pick which AI tool to use to learn more). But for now they’re pulling their weight as is.