Navigate with arrow keys or → button
Claude Code launched early 2025. It took the WSJ, The Atlantic, the stock market — everyone — roughly 10 months to grasp the impact.
I estimate Pi will take about 5 months. Half the time. Because people now understand agentic coding — Pi is the next layer up, and it's 10x more powerful because of composability.
Powerful, but you need all of this to get anything done
Many moving parts. Can't extend them with your own primitives.
Pi is composable. That single difference replaces the entire stack.
That's it. Extensions are your MCP servers, your skills, your RAG, your tool integrations, your custom protocols. You go directly to the primitive and compose it yourself. No middlemen.
I wanted to port the Chrome DevTools MCP to Pi. Then I realized: just look at the underlying protocol — the Chrome debugging port — and interface with it directly through a Pi extension.
Cut out the middleman. Opus 4.6 is so strong it figures out complex system integration like this easily — where 4.5 would get stuck.
Things are insanely fast between the latest models and how well-composed Pi is. Claude Code was incredible. Pi as a refactor on top of it is incredible. Each layer compounds.
Approximate capability / velocity over time
Armin Ronacher on Pi — the future is agentic composition. See also his agent-stuff repo.
Tobi Lütke — called Pi "the most interesting agent harness" and the "dawn of the age of malleable software." A tiny core that writes plugins for itself as you use it, RLing itself into the agent you want.
The individual building blocks — models, protocols, tools — are now strong enough and fast enough to recompose on your own via prompting. This is pretty incredible.
With some experience in long-term architecture thinking, you can compose very sophisticated, strong applications using Pi to bootstrap yourself.
You don't need a team of 20. You need vision, architectural thinking, and the right composable pieces. The model figures out the plumbing.
10 months for Claude Code. ~5 months for Pi. Each cycle doesn't just add capability — it halves the time to the next breakthrough.
That's not linear progress. That's the signature curve of approach to a singularity.
Each layer compresses the cycle. The curve bends toward real-time.
There won't be a single dramatic moment where a machine "wakes up" and everything changes overnight.
It will be more mundane and more profound at the same time. It's someone building a sophisticated app in an afternoon that would have taken a team months. It's recognition cycles compressing until new capabilities are absorbed in real-time.
The magnitude of change is real. Individual developers with the right tools and architectural intuition will produce what used to require entire organizations.
The singularity isn't a moment — it's a phase transition we're already inside of. We just 2x'd our speed through it.
This presentation was built and updated with Pi. View the session export →