As we end 2024, I wanted to do something a bit like my last update in 2022 and note down my thoughts on one or two big trends that we’ll see in 2025. I can boil all of this down to this basic idea:
AI model innovation will slow down but it’s impact will accelerate.
Here are three things I’m observing right now:
- LLM Models seem to have plateaued a bit:
- The top models are very close in general performance.
- The next-gen model from OpenAI is delayed (supposedly), hinting at some possible scaling issues that could affect other models.
- Improvements seem to be coming from functionality built into the API layers of the major platforms - for example, more round trips to the LLM, intermediate analysis layers, etc.
- There’s a lot of work to do on compute cost and speed at inference time, which will take up a lot of attention at the bigger AI companies.
- Enterprises have lots of low-hanging fruit that are great fodder for experimentation, which will allow them to work through enterprise adoption friction (e.g. compliance, infosec, privacy, etc.). This will accelerate their second/third/next use case (and many enterprises are already at this step).
- Coding tools are good enough to push good developers to greatness and great developers to super developers. (This may also be true for the other product development roles, too)
Together, they point to a good year for AI adoption, even if the models don’t improve dramatically in the short term. They also suggest that any team developing software MUST look at how they adopt these new tools into their work and processes. There’s no excuse at this point even for larger teams - the cost curve will move downward as you figure out your strategy and work through purchasing/compliance, and the current tools are good enough to make a difference.
I’d like to highlight a real example from Claire Vo, who is the CPO at LaunchDarkly. A former colleague’s share about the concept of “Super ICs” pulled me down a rabbit hole that led to Vo’s approach to integrating AI into her big team at her day job as well as building a product as a solo founder. Lots of good little ideas woven into the stories linked below, which makes me excited about what the next year will look like for developers.
- The Super IC: Good, short high level overview of the Super IC concept and how this CPO puts AI toolsinto practice in their personal toolbox and in how they interview product managers. This is a good place to start down the idea of Super ICs.
- Dan Mason on Raj’s post: Dan is the former colleague I mentioned. This LinkedIn post actually put Raj’s article on my radar. Dan adds some more ideas to Raj’s post, all worthwhile, but the money quote in there was “behind every successful person is an anxiety disorder channeled into productivity” which is an amazing statement.
- Config 2024: Hard truths about the future of product management: The Super-IC concept cited in the posts above was introduced by Claire Vo in this talk at Figma Config 2024. Pretty broad discussion, but you can hear how she’s weaving these ideas into her day-to-day as the CPO at LaunchDarkly while also experimenting with a side hustle called ChatPRD.
- She Built an AI Product Manager Bringing in Six Figures—As A Side Hustle - Ep. 24 with Claire Vo: Speaking of ChatPRD, Claire Vo has talked about this a lot. This episode was a nice overview of the way she approached building it, including the insight that the AI was helping her catch features she missed. Put another way, these models are better at generating than analyzing. Exploring where that generative capability provides the key value will lead to more useful products in today’s models. I also appreciated the super-IC concept even more - her hypothesis that these solo creators will be able to maintain clarify of vision by eliminating the committee approach that most product dev turns into. There’s a more recent detailed update on Every if you happen to subscribe to it (I don’t, so if you have the full article, would love a copy 🙂).
I also have linked to some of the tools that Claire Vo used to accelerate the build down below.
I hope everyone has a great New Years and an amazing 2025! To my friends, if you’ve got ideas for a side hustle and want to bounce them off of someone, please reach out!
Reads #
-
Server-Sent Events (SSE) Are Underrated: I need to go back and play with this again. I always found the bidirectional nature of websockets annoying more than anything else for most of my use cases.
-
Cognitive load is what matters: Really good way to think about software architecture, even in the early days of a product.
-
Humanity vs Humans: Good quote about Elon Musk:
Musk cares about humanity, but not much about his fellow human.
Could probably change humanity/human to America/American to explain our current political moment, too.
- Exploring LoRA — Part 1: The Idea Behind Parameter Efficient Fine-Tuning and LoRA: I’ve been going back to basics on some of my ML knowledge, re-reading books and going for articles like this that explain the basics of common techniques. This series has been helpful.
Code & Tools #
-
Cursor: I know you know about this tool, and I’ve talked about it before, but I’m giving the Pro version a go to speed up my side project. Given the topic above, thought it was worth highlighting. Will be comparing this to Github’s CoPilot as I go.
-
Clerks User Management Platform: I’ve used Cognito, home grown solutions, and Auth0 and I generally hate all of them. Found Clerk from the Claire Vo interview I posted above, will be trying it on my current side project.
-
Tiptap: Collaborative editor platform & components (think Notion-style editing) at a reasonable price. Also found via the Claire Vo deep dive earlier - she used this for ChatPRD.
-
List of A16Z Apps Unwrapped : Good list of tools in a couple of different domains.
Watch #
- Deskpi lite NVMe case for Raspberry Pi 5: Neat little case for the Pi 5. Nice review and assembly video.
Listens #
- ATP Insider: Making the Show: The ATP Podcast team did an insider special (aka a subscriber only episode) on how they make the show. Lots of nice tidbits in here if you’ve considered podcasting.