A recent post from "Secret CFO" hits on the big questions underneath the theme of this post...
"Here’s something finance teams are going to have to figure out fast with AI…
Where the goal Is the outcome itself vs where the process is part of the goal. I.e. you wouldn’t send a robot to the gym on your behalf because the goal is not for the weight to be lifted, it’s to lift the weight.
Some tasks in finance… the objective is simply the outcome (transaction processing prime example) others the process is just important.
Ie budgeting is more valuable than the budget itself (imo).
Working out where this dividing line sits is going to determine where AI is most valuable and most dangerous.
This gets abstracted one level further when you start to think about the influence on entry level roles and developing base line skills."
AI is the transformer version of Microsofts “Clippy” 📎. The rate at which it’s revolutionizing and the way it’s changing jobs, makes you wonder if hiring going to shift towards “theory-based” for some roles, instead of experience. I think once AI starts to get regulated it won’t be as fast as you think. Bolt is another good vibe coding platform that’s able to build databases
Lawyers and IBankers and PE folks and the similar white collar workers would all say that junior/interns workers are all not profitable for the business but absolutely crucial for recruiting pipeline for the firm. Would they like to have people internally funneled vertically through the firm who know the culture/work ethic/system or hired externally where they have to adapt? I'm not sure where AI will leave this dynamic - where does the learning come in? Or is it a race to get to better than human error and that's acceptable? Models should improve where we underwrite or accept those problems? But how quickly does a pipeline erode the human-in-the-loop over the course of 5-10 years....
It's a challenging question for the dynamic of some of these larger firms that relish the opportunity of curating (aptly - human capital management) for the futures of their businesses. We've had the ability to automate this work by any self-taught Python/R enthusiast but you're shunned in banking/traditional because they couldn't change things in the same format as before, even if it were easier - eventually it goes to the investment committee GP or the credit officer who's been on Excel since it first came out and can't be bothered to switch.
I, for one, relish the opportunity for Excel to enter the 21st century as code/dynamic, though probably less so if I were at a Causal or these other startups.
This is such an insightful take on LLMs; you're aboslutely right about the career advantages of getting up to speed quickly. Thanks for sharing your exploration, it's vital for everyone to understand how these tools are reshaping fundamental skills, even beyond coding.
Hands-on example of how AI is reshaping financial modelling - not just for cap tables, but for any repetitive, logic-driven analysis. This shift matters just as much in B2B finance, where cash flow forecasting, credit scoring, and payment-term modelling are ripe for similar automation. TCLM often explores how tech changes the operational layer of trade credit and liquidity management. A tangible look at the tooling that’s already here.
I've found Excel/modeling AI tools most helpful for people with a solid level of modeling proficiency (I'm biased, started my career in banking). If you already know how to swim in Excel waters, you can handle disasters (i.e., auditing the model) a bit more readily when you know how the thing has been built across each plain-language request.
Otherwise, you may be AI modeling your way into a workbook that'll fall apart. But I can also imagine a scenario where the tools become bulletproof in the future. That proficiency and need to audit fall to the wayside.
Hey Jack, glad you're on this topic. I've been testing Shortcut AI for the last couple of weeks, and I'm really impressed, but it's really expensive. Do you want to try Shortcut AI and compare it to Claude? Curious how these two stack up.
Yes, AI for using Excel more effectively is possible, especially if you use a limited vocabulary and only functions built inside Excel. As for asking Claude to write another Excel, that would still be a while.
A recent post from "Secret CFO" hits on the big questions underneath the theme of this post...
"Here’s something finance teams are going to have to figure out fast with AI…
Where the goal Is the outcome itself vs where the process is part of the goal. I.e. you wouldn’t send a robot to the gym on your behalf because the goal is not for the weight to be lifted, it’s to lift the weight.
Some tasks in finance… the objective is simply the outcome (transaction processing prime example) others the process is just important.
Ie budgeting is more valuable than the budget itself (imo).
Working out where this dividing line sits is going to determine where AI is most valuable and most dangerous.
This gets abstracted one level further when you start to think about the influence on entry level roles and developing base line skills."
Play around with Shortcut.ai or Tracelight.ai
AI is the transformer version of Microsofts “Clippy” 📎. The rate at which it’s revolutionizing and the way it’s changing jobs, makes you wonder if hiring going to shift towards “theory-based” for some roles, instead of experience. I think once AI starts to get regulated it won’t be as fast as you think. Bolt is another good vibe coding platform that’s able to build databases
Lawyers and IBankers and PE folks and the similar white collar workers would all say that junior/interns workers are all not profitable for the business but absolutely crucial for recruiting pipeline for the firm. Would they like to have people internally funneled vertically through the firm who know the culture/work ethic/system or hired externally where they have to adapt? I'm not sure where AI will leave this dynamic - where does the learning come in? Or is it a race to get to better than human error and that's acceptable? Models should improve where we underwrite or accept those problems? But how quickly does a pipeline erode the human-in-the-loop over the course of 5-10 years....
It's a challenging question for the dynamic of some of these larger firms that relish the opportunity of curating (aptly - human capital management) for the futures of their businesses. We've had the ability to automate this work by any self-taught Python/R enthusiast but you're shunned in banking/traditional because they couldn't change things in the same format as before, even if it were easier - eventually it goes to the investment committee GP or the credit officer who's been on Excel since it first came out and can't be bothered to switch.
I, for one, relish the opportunity for Excel to enter the 21st century as code/dynamic, though probably less so if I were at a Causal or these other startups.
This is such an insightful take on LLMs; you're aboslutely right about the career advantages of getting up to speed quickly. Thanks for sharing your exploration, it's vital for everyone to understand how these tools are reshaping fundamental skills, even beyond coding.
Hands-on example of how AI is reshaping financial modelling - not just for cap tables, but for any repetitive, logic-driven analysis. This shift matters just as much in B2B finance, where cash flow forecasting, credit scoring, and payment-term modelling are ripe for similar automation. TCLM often explores how tech changes the operational layer of trade credit and liquidity management. A tangible look at the tooling that’s already here.
(It’s free)- https://tradecredit.substack.com/
I've found Excel/modeling AI tools most helpful for people with a solid level of modeling proficiency (I'm biased, started my career in banking). If you already know how to swim in Excel waters, you can handle disasters (i.e., auditing the model) a bit more readily when you know how the thing has been built across each plain-language request.
Otherwise, you may be AI modeling your way into a workbook that'll fall apart. But I can also imagine a scenario where the tools become bulletproof in the future. That proficiency and need to audit fall to the wayside.
Net-net: agree with you.
Hey Jack, glad you're on this topic. I've been testing Shortcut AI for the last couple of weeks, and I'm really impressed, but it's really expensive. Do you want to try Shortcut AI and compare it to Claude? Curious how these two stack up.
Curious what this flow looks like w/clawd bot, if you do it.
Yes, AI for using Excel more effectively is possible, especially if you use a limited vocabulary and only functions built inside Excel. As for asking Claude to write another Excel, that would still be a while.