The tools to build AI apps in 2025 are where web dev was in 1998: raw, primitive, and absurdly manual. Just heard Sam Partee break this down perfectly on a podcast: We're witnessing the natural evolution of abstraction in real-time. From machine code → assembly → C → Python → and now to natural language interfaces. But there's a critical difference this time. As Sam pointed out, we're not just abstracting logic — we're abstracting meaning itself. That's why prompt engineering feels more like philosophy than programming. The next frontier isn't better prompts. It's standardizing how AI connects to real-world systems and actually takes action. At Arcade.dev, we've been obsessed with this problem: building the infrastructure layer that turns "book me a flight" into actual bytes hitting airline APIs with proper authentication. The winners of the next decade won't be who builds the best chatbot. It'll be who builds the infrastructure that lets AI apps securely do things that matter. The rest is just talk. Link to full video in the comments.
This is a really fascinating take. Thanks Demetrios Brinkmann for the great convo!
Great conv Demetrios!
👎 Keep AI out of the creative arts!
Acube is ready to collaborate and build the AI infrastructure that powers real-world action. let’s create the future together.
Brian O'Grady
Interesting angle on abstraction. But how do you see the role of domain-specific interfaces evolving? At some point, won’t we need a middle layer that balances flexibility with guardrails, especially as AI touches more sensitive systems?
Yep. Everyone's obsessed with model quality, but the ability to take real action is the actual bottleneck.
Co-Founder/CEO Arcade.dev, Helping AI Agents Securely Connect to APIs and Data
1wLink to full video: https://youtu.be/5_Ii1JkD0wY?si=hlIj6KHblcFtacVo