← back
The New Primitives: Building AI Native Software — Kwindla Kramer, Daily
Takeaway
AI-native software will require new interaction and computation abstractions beyond today’s agent loops.
Summary
- Daily’s Pipecat provides open-source, vendor-neutral infrastructure for voice agents.
- Kramer frames current agents and multimodal harnesses as early steps toward a broader category of AI-native software.
- Vannevar Bush’s As We May Think provides a historical lens on new interfaces and information tools.
- Programming languages, interactive computing, databases, and VisiCalc illustrate how new abstractions turn computing capabilities into useful products.
ai-nativeinterfacespipecat
Original description
In 1945 Vannevar Bush described document scanning, OCR, speech to text, hypertext, search engines, a head mounted camera, and voice interfaces, in a single essay, before any of it existed. Kwindla Hultman Kramer uses As We May Think to set up the uncomfortable part: he was a baby programmer in 1995 writing HTML by hand and web servers in C, as excited about web pages then as he is about agents now. Web pages turned out to be a primitive, not a destination. His argument is that agents are the web page of this era, and the thing worth building is the AI native software that comes after. The tour through the decades is the evidence. Each one had a job: programming languages to carry human intent into the machine, interactivity in the 1960s, abstractions for scale in the 1970s, then the personal computer. VisiCalc is the example he keeps returning to, because it did not put accountants out of work, it made vastly more accounting possible and invented categories of work nobody could have described when a screen of calculations took a room full of people. He offers that as the reply to the mass unemployment worry. The talk closes on Gradient Bang, a massively multiplayer game with an LLM at the core of every interaction and hundreds of inference calls in flight, built specifically to exercise the primitives he thinks come next: asynchronous non blocking context compression, long running subagents that share context, progressive skills loading, dynamic interface generation, and conversational voice. Speaker info: https://x.com/kwindla / kwkramer https://machine-theory.com/ https://github.com/pipecat-ai/pipecat Timestamps: 0:00 - Daily, Pipecat, and what comes after agents 1:31 - What Vannevar Bush predicted in 1945 2:13 - Nadella on multimodel harnesses 4:12 - Agents are the web pages of 1995 5:25 - From the abacus to the stored program computer 6:46 - The 1950s: getting human intent into the machine 7:30 - The 1960s: interactivity, Sketchpad, and Star Trek 8:51 - The 1970s: abstractions that scale 9:32 - VisiCalc and why it did not replace accountants 10:49 - Knowledge Navigator, Apple's 1987 prediction 12:46 - The web was multimodal from the start 13:24 - Minority Report, Iron Man, and building it for real 17:10 - Cloud, then agents, then what 18:24 - Knowledge Navigator rebuilt on real technology 19:38 - Gradient Bang and the primitives it exercises 20:52 - Closing