February 24, 2026
Runtime roulette, anyone?
Software 3.1? – AI Functions
AI code that runs itself — hype vs panic in dev land
TLDR: Strands Labs’ AI Functions lets AI write and run code inside your app with automatic checks on every call. The community is split: some cheer the always-verified future, while others warn that English-based instructions are too vague for real-world systems. It’s bold, controversial, and potentially game-changing.
The internet just discovered “Software 3.1,” and it’s pure drama. Strands Labs’ new AI Functions promise code that the AI writes and actually runs inside your app, with automatic checks on every call. That’s a leap from “prompt it, ship it” to “prompt it, run it, verify it — forever.” One excited commenter, aspittel, even claimed AWS shipped it, adding to the brand confusion and the buzz.
Then came the spicy takes. waynesonfire slammed the whole idea with a dagger: “English is a terrible programming language.” Translation: vibe-coding your app sounds like chaos. bilekas went full doomsday: “I pray I never have to have this in a production environment,” sparking memes about “runtime Russian roulette.” Meanwhile, xiphias2 flexed with a nerdy detour, comparing it to Symbolica and dropping a link to alleged record-breaking tests on an AI reasoning benchmark ARC, which only made the thread hotter.
In the peanut gallery, moffers asked if this could be done with Erlang (the old school phone company language), because of course the Erlang crew showed up. Fans say this is the future: AI hands you real objects, checks itself, and retries when it messes up. Skeptics say you’re inviting ambiguity into mission-critical systems. Verdict: thrilling, terrifying, and extremely clickable.
Key Points
- •AI Functions is a Strands Labs project built on the Strands Agents SDK to run LLM-generated code at runtime.
- •The approach returns native Python objects and uses post-conditions to verify outputs on every call with automatic retries.
- •This model shifts AI’s role from development-time text generation to runtime execution and continuous verification.
- •It uses an @ai_function decorator for natural-language specifications instead of implementation code.
- •The article frames this paradigm as “Software 3.1,” evolving from Karpathy’s Software 3.0 generation–verification loop.