A century in 5 years

Justin and Marty’s conversation covers rapid AI progress, how that progress is changing what it means to build software, and practical implications for freedom tech, Bitcoin-adjacent projects, and individual agency. They mix technical examples (OpenClaw, Marmet/white noise, Nostr, local models) with strategic and cultural observations about open source, decentralization, and personal “will” to act.

Key points (with supporting evidence)

• AI capabilities are accelerating and enabling much faster development.
  • Justin: “a century could happen in the next 5 years.” He contrasts past months/years of incremental work with recent speed: “over the course of 24 hours I built what I couldn't build in a month.”
  • He attributes the jump to “how much smarter the models have got,” plus standing on previous experiments.
• Agents / LLM tooling change how software is built: more prototyping, personal apps, and “vibe coding.”
  • Justin describes doing “two 300 projects” over the last year, mostly to play and learn, then using LLM-driven workflows to quickly produce usable apps (mobile native-feeling Rust app, OpenClaw integration).
  • He recommends a play-first approach: “just be like, I'm just going to f around here a little bit... be a kid in a sandbox.”
• Open-source/model competition reduces runaway-winner risk and enables decentralized innovation.
  • Marty: recent open-weight models (GLM 5, Minimax 2.5) “are damn near as good as” big closed models; the dynamic is “extremely competitive.”
  • They argue this competitive landscape empowers builders outside hyperscalers.
• Freedom tech and Bitcoin communities must bridge to AI or risk losing influence.
  • Justin: “Use this stuff to build the freedom tech. It's never been easier. I've never been more bullish on freedom tech.”
  • Both stress that open-source/self-hosted stacks plus AI agents can be used to decentralize services (e.g., Marmet/white noise for messaging, Nostr relays).
• Real products win by solving user problems and distribution, not purity of design or ideology.
  • Marty: examples like Pitch succeeded because it solved a clear user need (organize when internet is off); “nobody cares how a problem is solved. They care that the problem's solved.”
  • Justin recounts an HRF hackathon where activist + dev pairings produced eight useful projects — starting from distribution and real problems worked.
• Practical architecture trends: apps-as-APIs, media/pub-sub for real time, and mixed client-server models.
  • Justin explains media-over-quick (binary pub-sub) mapped to Nostr’s pub/sub principles as a way to support real-time audio/video while preserving decentralization-ish relay architecture.
  • Peter Steinberger’s thesis: useful applications will expose APIs so agents can interact without brittle UI scraping.
• Local models, hardware, and privacy trade-offs are important but messy.
  • Justin predicts “a lot of us will spend tens of thousands of dollars on hardware” to run local models; but notes local models are less capable without web access and can risk ex-filtration when connected.
  • He summarizes a mixed approach: “use the smartest model regardless of where it came from,” and consider local for privacy when possible.
• Education, mental models, and “will” matter more than raw coding skill for adoption.
  • Justin: building mental models of what an AI “skill” is and treating tools as anthropological subjects helps users be effective.
  • He re-frames the future skill set: “we lived in a world where intelligence was really important... maybe we're moving a little towards a world where like will is what you need.”

Significant or surprising insights

• Velocity shock: tasks that previously required months can now be done in days.
  • Evidence: Justin’s 24-hour rebuild of last summer’s month-long project, plus a near-complete mobile app in days.
• The social unravelling + tech surge mental model: they sense both cultural and technical tipping points.
  • Justin links cultural moments (“the Epstein thing...big conspiracies that are unravelling”) with a tech ground swell, arguing both make “something’s happening here.”
• Agents won’t autonomously adopt niche economic rails (e.g., Bitcoin) without intentional design:

.
- Justin: agents “will do whatever you prompt them to do” and they are trained on internet data dominated by dollars; “the agents aren't primed to do Bitcoin.”

• Coding as capability vs. code-writing: more people will become “programmers” by composing agents and servers even if they never write traditional code.
  • Marty notes his own transformation: “I'm not a programmer... but now you are. What you were just describing is computer programming.”
• The “will” thesis: success in the new era may favour people with agency, curiosity, and persistence over mere credentials.
  • Repeated emphasis from Justin: play, iterate, and apply willpower to ship, teach, and build communities.

Actionable takeaways

• Start small and play: build a tiny, low-risk automation or integration (e.g., distil PDFs, create an ad deck) to learn the tooling.
• Focus on problems + distribution: prioritize simple solutions that solve a clear pain and have a path to users (activist networks, a community).
• Combine local and hosted models thoughtfully: use local models for private tasks, hosted for capability, and design explicit controls (kill-switch, separate drives) where possible.
• Invest in mental models and education: learn what “skills” and APIs mean for agents so you can make informed design/deployment choices.
• Lean into building and teaching: contribute small, usable tools, and document/teach them to lower the barrier for others.

These points come directly from Justin and Marty’s recorded discussion: their demos, hackathon anecdotes, and repeated advice to “play, learn, ship” form the backbone of the argument that AI is both a massive accelerator and a call for active, values-aligned participation.