Post

What do agent security & the future of programming have in common? I’ve had the pleasure to visit #ScalaDays in Berlin and #FuncProgConf in Stockholm day-after-day. The central topics at both was the same: what's the future of programming?

And I agree with Martin Odersky (despite his keynote "End of programming", see link below) & Nada Amin that there *is* a future; programming languages are far from done. For one, if anything, the agents make it *easier* to work with advanced type systems (see Rust's surging popularity), not harder.

Short of developing full formal specifications, there's so many interesting properties that we can prove at compilation time - above memory safety, and efficiently - and that's what Scala actively explores.

Secondly, security security security … how to make sure that agents don't do what they're not supposed to do? We already know they can go rogue, even if out of incompetence - OpenClaw and unrestricted Claude/Codex have demonstrated this a number of times.

Sandboxes provide some protection, but they should only be the first layer. Another is what Łukasz Biały and Jakub Warchoł demoed - verifying agent actions at the level of the compiler, before anything is executed. Thanks to information-flow-tracking, this can be made both secure *and* useful!