I’m AI-pilled, bullish on what we can already do today and incredibly excited about what might be possible tomorrow. Yet, at work as a CPO running a 100 people team, none of the available tools are truly helping me. Not like a 10x difference - I’m not even seeing a 2x or 1.5x change. Don’t get me wrong. I know how Cursor or GitHub Copilot help my team. Many of the people working for me get a ton of value out of it. As a company, we benefit from it. But me personally, in my day-to-day job, no.
I Use LLMs Outside of Work All the Time
This is especially weird because I use various LLMs outside of work all the time. I can point to clear examples where they help me
- Angel investing: Research various topics, get smart on an idea, hone a thesis I have, get up-to-speed on a specific company. All of it is easier, more fun, faster with LLMs
- Building random apps: Build random ideas and little apps on the weekend to see what’s possible and how the different tools work (e.g., Claude Code vs. Replit)
- Cooking: Come up with variations of recipes I like, help me understand why doing it this or that way is better/different, brainstorm dinner ideas for larger groups
- Learning about random topics: Probably my favorite way of using LLMs right now. I randomly hear about topic x and I start exploring the topic for a bit just to learn and get smarter on it without any particular purpose other than learning for fun
- Product purchases: Whenever I buy something where I compare different products, I primarily use LLMs to do most of the work with the exception of cross-checking against trusted sources (which might be overkill and just an overhang from the “this thing hallucinates too much" era)
All of it is clearly clustered around “research” (plus the obvious agentic coding use case). Help me find something, help me make a decision, help me understand topic xyz better. There are other use cases for me that don’t directly leverage a LLM (as in “I go into Gemini / ChatGPT and prompt it”) like the productized experience of Google Photos. I love and use those a lot too. Obviously, I also use it to help me write this blog post (primarily to feedback drafts of it - I still prefer to write the core pieces of a post myself).
AI Helps Individual Contributors a Lot
At work, there are a bazillion use cases where LLMs help us. Whether it’s engineers leveraging Cursor (or any other code gen tool), designers generating prototypes out of Figma or product managers showcasing their ideas with v0 prototypes. That’s ignoring anything we are doing with the help of LLMs in our product for our customers. We have various GPTs to help different people/roles do their job better/faster. And everyone obviously uses an LLM to help with any written content. Tl;DR, there’s work where LLMs help us do the work better/faster, while having more fun. We can build a better product and support more customers with fewer people or without having to grow the team as aggressively as we would have done five years ago. All of this is good.
My theory is that the biggest impact is on people in individual contributor type roles. The engineer using Cursor to build a new feature better faster, or the product manager showcasing a potential idea with an actual prototype vs. just a document / PRD. It is meaningfully different / better. If your job is creating and inventing, the actual picks and shovel work, today looks different than a few years ago. That’s true across the board, across different roles and functions. A lot has been said about agentic coding tools, but the same is true for anybody else “who creates”, whether that’s marketing copy, sales outreach, internal operations work or how an executive assistant operates today vs. yesterday.
What About Management / Strategic Roles?
All of this obviously impacts me directly in running a 100 people organization across engineering / product / post-sales / customer success. Ultimately my impact / output is my team's combined impact / output. But when I wake up on a Monday morning and open my laptop, I don’t see how it changes my personal day-to-day work (yet?). To be clear, it does impact me whenever I do hands-on work (like writing the one-off PRD, or analyzing some data and generating the SQL for it). At the end of the day though, that’s the smaller portion of my job / day. The majority of work is simply meetings. Meeting customers, meeting direct reports / 1:1s internally, board meetings, various offsites with different teams, partners, hiring/recruiting, etc. There’s preparing those meetings, there are emails, there’s random other work that creeps in when your company hits a certain size. I believe you always need to be hands-on (in various ways). Your job can’t be just managing. I’m trying to squeeze as much “maker time” out of a “manager schedule”. But even if that’s the Northstar, I use the majority of my day to help other people do their job well vs. me doing hands-on work.
For this type of role I’m operating in at the moment, I have not found how AI can meaningfully help me directly. I listened to a ton of interviews / podcasts over the last weeks, tried to get some ideas from my favorite friend (the LLM), looked at what other people in similar roles are doing and nothing has truly resonated with me. The various ideas I heard are mostly around:
- Transcribe meeting notes: It is totally overhyped use case. What’s powerful about taking notes is that you make a lot of tiny conscious decisions about what matters, what should be a follow-up task, what should be the big takeaway and I haven’t seen AI replace that well. What happens right now is a fairly unintelligent transcript with the most obvious action items called out that will be mostly ignored because only a subset truly matter
- Manage email inbox madness: For whatever reason, my inbox isn’t madness. Yes I get a ton of emails but most don’t require me to do anything or more than a one sentence response. Maybe I’m lucky because I can rely a lot on my team and I’m ultimately just delegating out most of the madness
- Protect your calendar: My default response to meetings is “no” which is probably the simplest hack for not having too many meetings. I don’t need an agent to analyze my calendar. Whatever makes it on my calendar crossed a reasonably high bar and I made a conscious decision for it to be there
- Meeting preparation: AI does help me there. E.g., I can ask a corpus of customer-related documents and notes questions before meeting the customer. But I also get mostly the same information through whatever meeting prep happens for that meeting (can be a few async bullet points on Slack). The LLM insights complement the human input (CSM telling me to focus on xyz with the customer). They are ultimately nice-to-have
- Scale yourself with GPTs: A bunch of folks are talking about how they scale themselves through a GPT that provides similar feedback to what they would provide. So instead of me feedbacking a document, AI feedbacks it first and kinda feedbacks it like me. I can see that but also believe that scrutinizing an idea, a prototype, a presentation, creating customer feedback together as a group is incredibly powerful. It’s how you establish a shared understanding within a team
My current thinking is that there’s simply a part of my job that stays fairly unchanged (e.g., regular 1:1s with people on my team). There are slight productivity improvements around that part of my job but that is it. Where AI might have the bigger impact is the creative / builder side of my role - “the maker schedule”. In the past, building a small prototype felt unachievable, primarily due to time constraints. Now it’s possible with much less effort and with more interruptions because the agent needs time to do work anyway (and I can do another meeting in the meantime, come back and review the result, prompt the agent again, repeat). That dichotomy is what’s frustrating. You can build a whole app in a few hours with thousand lines of code generated magically and it being auto-deployed and hosted. That was not possible before. It is a step change. That’s half of my day. The other half is smarter recommendations for email replies. Not meaningful. It’s incrementally better but it pales in comparison. AI doesn’t make my manager's schedule better, more efficient or different. Maybe all it does is that I can build more during my builder / creative time. Maybe I need to experiment more. Let me know what works for you - happy to chat or discuss async.
