(Hot?) Takes on AI
A series of takes on where we are at with AI
Introduction
Let me say this up front: I’m not here to debate AGI. I don’t think it’s around the corner, and honestly, I wouldn’t even know how to define it if it were. (Spoiler: it seems Sam Altman doesn’t either.)
Instead, I want to talk about what I am seeing — in tools I use every day, in teams I work with, and in the flood of hot-take podcasts, product updates, and new AI startups appearing weekly. These are my working notes. Unpolished, slightly sarcastic, hopefully useful.
My Takes on AI (So Far)
1. This Technology Is Transformational
Yes, it sounds obvious now. But it’s worth repeating. This feels like the early days of the internet, mobile, or cloud — but moving way faster. What took years to develop and distribute back then now seems to happen in months. The main difference is that previous shifts gave companies room to adapt. This one doesn’t.
“The AI Marketing Playbook” by Reforge discusses how AI is reshaping marketing strategies and the rapid pace of change. Read more here.
2. Product and Software Development Are the Most ‘Advanced’ Areas
Although AI is available to everyone, I’ve seen an explosion of new tools (especially in GenAI) targeting product managers, engineers, and designers. The distribution and adoption curve is steepest here because:
The users are already technical
The pain points are clear
The ROI is measurable (hours saved, bugs fixed, experiments shipped)
This is also a huge signal that there are still opportunities where this technology can be applied and make a significant impact.
3. Accessibility Is a Game Changer
Unlike blockchain (remember that?), AI tools actually work and are easy to use. You don’t need a PhD. You don’t need to download a whitepaper. You just open a browser tab and start typing. The interface is quite human-friendly, and that, in terms of distribution and usability, makes a huge difference.
4. Using vs. Building GenAI Are Completely Different Skills
Using AI is about enhancing your workflows: summarizing notes, drafting emails, iterating on code, debugging logic.
Building AI tools? That’s a different beast. You’ll need to understand transformers, fine-tuning, prompt evals, CoT (Chain-of-Thought), and all the tiny acronyms your LLM forgot.
5. Change Is Non-Linear
New models don’t just improve things slightly. They leapfrog. GPT-4 over 3.5. Claude 3 over 2. Devin over everything else. The gap isn’t incremental — it’s exponential.
6. You’re Always Using the Worst Version of AI
Whatever tool you’re using today? It’s the worst it will ever be. That’s wild. And motivating.
OpenAI’s CPO, Kevin Weil, discusses this concept in detail on Lenny’s Podcast. Listen here.
7. Incumbents Have an ‘Advantage’
Compared to other waves and trends, incumbents are very well positioned to surf the AI trend. They have the people, the money, the users, to avoid the innovator’s dilemma situation. Nonetheless, I think they will have a hard time from the management and cultural change, to get their employees to navigate this new ‘change’ wave.
8. What to Build Becomes Crucial
With the idea of AI Agents going around and already here (like Devin), it becomes highly relevant to understand what problem to solve, how to break down the problem into many pieces, create the solution, and provide the concrete actions to the agents. I’m experiencing this myself with personalization algorithms, where we’ve spent a considerable amount of time building the framework and concepts, and it took us a couple of days to get a prototype up-and-running in our dev environment.
9. Team Changes, Not So Far
Although I believe teams will be reconfigured in the mid-future, and the typical trio might become a duo, I don’t see that happening right now. We are seeing team members using a bunch of AI tools (the same or different) and each of them bringing their insights. We don’t see today that team members are being replaced.
Claire Vo shares insights on this topic in her conversation with OpenAI. Read more here.
10. AI Agents + Workflows
Besides OpenAI, which seems to be addressing more horizontal use cases, we are now seeing a bunch of different companies optimizing specific workflows for many different ICPs. For instance, Anthropic is focusing on coding skills, Devin / Cursor / Windsurf on software development, and others like Enterpret, Reforge Insights Analytics on gathering and summarizing data insights (qualitative and quantitative). The main advantage is automating workflows that previously were not possible due to the lack of AI capabilities.
11. Stickiness Over Locked-In Resources
With AI being available to many people and a lot of products being created, the main differentiator will be great product experiences that keep users engaged, not the regular seat-based pricing model. Of course, incumbents won’t change that overnight, but pure AI-based companies are monetizing by usage and the things people are able to produce rather than by mere access to the tool.
The concept of “selling work, not software” is explored in this article. Read more here.
12. Not a Silver-Bullet Workflow
One of the exciting things about using these tools is that there is no single best flow. I tend to go with Claude / ChatGPT > v0 > Cursor AI when going from idea > ideation > building. But others have different ways of doing the same.
13. Show Me Something That Works, Not Mock-Ups
One of the things we’ve faced recently is the change in pace on how these tools can help you move faster. Very soon, having a meeting with mock-ups will be replaced by showing a ‘real’ prototype that is ready to be implemented, and later directly integrated into our code base. Exciting times ahead for pragmatic companies who want to move fast.
The new Lenny’s podcast of How I AI is a magnificent proof of that. Listen here.
Challenges Ahead
The Novelty Effect Will Wear Off
Right now, launching “yet another AI tool” still turns heads. That won’t last. The bar is rising fast. Only real value will survive.
Memory and Context Are Still Broken
Even the best models forget what you told them five minutes ago. Context windows are getting better, but memory is still the UX bottleneck.
The Real Battle Is Context Ownership
Whoever becomes the Okta of AI (the unified identity & context manager) will win big.
Brian Balfour and Fareed Mosavat discuss this in their podcast, Unsolicited Feedback. Listen here.
💬 We Want Tools with Opinions
Users don’t just want impartial assistants. We want AI that has a take. That says, “based on your style, I recommend this.” The more confident the agent, the more useful the interaction.
Final Thoughts
One thing I’ve noticed about the people building the most impressive tools today: none of them know where the boundaries are. And that’s not a bug. That’s the point.
We need users pushing the limits. Testing edge cases. Giving feedback that makes these tools better.
And while much of the buzz is around productivity, I’m particularly excited about how AI could improve access and affordability in critical fields like healthcare and education. That’s where the real wins lie.
To the question, “Is AI going to take over my job?” Well, maybe yes, maybe not. As an optimist, I prefer to ask, “How can AI help me do my job better, faster, simpler?”