Joepaul,
Your observations are pretty consistent with my own. I've also done a lot of automation where it's not 100% automated, but mostly automated, and the challenge is usually at the handoff points: getting the information in, or sending it to someone else. I call these the first and final miles. The "first mile" is usually solvable with Excel, Power Query, and the usual tools. The "final mile," things like sending an email that integrates the output correctly, is harder, partly because of restrictions on automated sending, and partly because it's a less structured, more non-deterministic problem. AI has real potential in both of those spots.
I really like the idea of using this with your finance teams. It's easier now to start with a draft, run it through a company-controlled or enterprise LLM to anonymize it, then move it into NotebookLM for flash cards, podcasts, whatever fits. The client scenarios are especially interesting to me, since it's about as close as you can get to a role play without actually doing one.
You're right that the cost to try something is very low now. One thing I've been working on is not creating as many open loops. I'm not great at it on the AI side, but I do clear out my chats periodically: things move to a someday/maybe pile, or get deleted if I'm not going to pursue them. It can feel overwhelming to have that many ideas with potential just sitting there.
I also spent time today on Claude Cowork and Claude Code, something I'd been wanting to try for a while. I didn't get far, an attempt to automate a couple of report pulls didn't work in either tool, but it did work through Claude for Chrome directly, which both of those tools turned out to rely on anyway. Rather than fully automating it, I ended up just being able to paste a prompt in, which sounds small but removes the friction of stopping what I'm doing, logging in, and downloading things manually. That friction was the actual reason I wasn't doing it as often as I wanted to. Good enough progress for one afternoon (it took about three hours), but it reinforced something I already knew: even the simple stuff takes real time before you get proficient at it.
------------------------------
Daniel Harley CMA,CSCA
Director/Manager
Newtown PA
https://www.linkedin.com/in/danieljharley/------------------------------
Original Message:
Sent: 09-03-2026 03:41 AM
From: Joepaul Kalathil John
Subject: Hump Day: breaking up the week with something new
This is an interesting experiment, Daniel. The learning angle is probably what caught my attention most.
I manage finance operations teams, and one thing I've been experimenting with is how we can move people from simply knowing accounting to actually understanding the process around it. For example, it is relatively easy to teach someone how to do a bank reconciliation or pass an entry. It is much harder to teach them why something looks wrong, what questions they should ask, and how that eventually flows through to the financial statements.
That's where I think changing the learning modality with AI could become really useful. Take the same dry accounting content and turn it into a podcast, a client scenario, a case discussion, or even have AI play the role of a controller questioning your analysis. Suddenly the learner has to think rather than just remember.
We've also been experimenting with AI and automation in actual finance operations - reconciliations, reporting, workflow improvements, etc. What I'm finding interesting is that the value isn't always in automating the entire process. Sometimes just giving an accountant a different way to investigate or think through a problem makes a meaningful difference.
I also really relate to your point about the path from idea to execution becoming so short. I find myself trying things now that I probably would have just added to an "ideas for later" list a couple of years ago.
But I think that creates another challenge: starting things is becoming incredibly cheap. Deciding which experiments are worth continuing, scaling and actually closing out may become the harder skill.
I haven't played around enough with NotebookLM yet, but this definitely gives me a reason to try it. I might actually start with some dry finance material and see what it can do with it.
------------------------------
Joepaul Kalathil John CMA
Director/Manager
Ernakulam
India
------------------------------