Questions from the Automate the Boring Stuff workshop at the Next Venture Summit in Greenville, South Carolina.
Look for repetitive work that drains the energy you want to spend on more interesting, meaningful work. Taking information, reformatting it, and putting it somewhere else is a good example of what I call mental assembly-line work.
Pick one small, annoying piece you understand well. Get that working, then build from there. Some boring things, like exercise, budgeting, and maintenance, are still worth doing. Automation can help make those more tolerable, too.
You can start learning the method with a free chat. Direct access to folders and more involved workflows may require a paid plan. I think that's worth paying for when it saves time and makes your work more enjoyable.
For agents that work directly with your files, look at Claude Cowork in the desktop app or OpenAI's desktop app. Look for the option to select a folder on your computer; a project that only organizes chats won't do the same job.
Start with one folder for the workflow and give the agent access to that folder. Keep the setting that asks before making changes while you get comfortable. You can grant access to more folders as you need them, within whatever your company's systems allow.
I suggest starting with text instructions and an agent to see how far you can get. Once the workflow works, look for steps that always do exactly the same thing. Those are good candidates for Make, Zapier, or a script. You don't need to spend AI usage on work that doesn't require interpretation.
I keep a tools folder where the agent can write scripts, then have the workflow say which tool to use at which step. One set of instructions can combine agent judgment, your input, and software. You can improve or replace individual parts as you go.
Start with a small part of the workflow and inspect the result. When a calculation is wrong, ask the agent to help trace what happened, update the instructions, and run it again. Check that the change fixes the result.
For repeatable calculations, have the agent write a script that the workflow calls. The instructions stay in the workflow file; the calculation lives in the script. My daily brief took about 15 iterations to work the way I wanted. Expect to spend time on this.
I use GitHub to keep a history of changes, return to an earlier version, and share files between devices. If you've never used it, ask your agent to help set it up and save versions as you work.
Also, resist the temptation to automate everything after your first success. I've fallen into that trap plenty of times. Work on one piece, review every line of the instructions, and make sure it works before adding more. Agents are very good at adding too much.
Yes. Point them at the same markdown files. I use a stronger, more expensive model to help with planning and difficult problems, then have it prepare a plan a cheaper model can execute.
The files are yours to keep. You can build on them over time and use them with different agents as those agents improve.
Create a file for each person or audience describing their role and what they care about. Have the workflow select the person, read that file, and gather relevant information from the sources you give it.
I'd start with something small, like preparing an agenda. Get it reliably gathering the right material in one place. That frees you to find the interesting insight or make the briefing more engaging.
That's a harder problem. My hunch is that you may be able to build a workflow that understands your company's systems and limitations and helps you work within them. What it can do will depend on the access your company allows. I wouldn't assume the same setup works everywhere.
At Pathwright, we have security documentation and data-processing agreements that govern how we handle educational data. Our agents have instructions pointing to those documents so they can check the requirements while the team works. We can also ask them to run a security review using the same documentation.
An agent can make information public by mistake, just as an employee can, but much faster. The rules need to be available to both your team and their agents. Pointing an agent at a policy is one part of that.
Many consultants offer that help. My experience is that they often have good procedures and prompts, but don't always have a good system for putting them to work.
For the workflow itself, I'd start with a folder, clear instructions, and one small task. Working that way gives you a better sense of what you need before buying more software or a larger program.