In type-theoretical terms, build the very core of stuff *intensionally/constructively*, only rely on small dose of extensionality from the agents, then all the obvious mistakes can be tracked down extensionally by them too. DO NOT start solely with it, or you lose the whole view.
Pathetically wrong. “Delivering the correct contexts” is like “write more specs and test cases for strengthening” and every dev knows the specs and tests are often larger than the code. The real problem is, why more specs instead of a senior that writes it correctly, JUST ONCE.
Don't be too hard-working to work.
The classic "top talent" often is. I've been on more than one "elite" team, the kind you measure in olympiad/ACM gold and headcount from top schools, with some of the highest-paid employees you'll ever meet. In the agent era, the very confidence and diligence that made them are what hold them back. They don't believe real AGI is already here, so they won't hand the critical code, the algorithms, the research itself over to an agent. The moment an agent underperforms, the reflex is "this thing is useless," not "did I give it the right context?"
Since the start of this year, I've fully believed agents are already smarter than most people, with no ceiling I can see. Every dumb mistake traces back to context we failed to design for them. That's the whole reason I started talking about AX, Agent Experience: stop assuming there's a class of work an agent can't do, and start asking what you forgot to tell it.
The talent of this new era wants to do everything the agentic way
Reussir is a programming language project centered on RC-based memory reuse analysis. It brings MLIR into the functional programming world through a customized bufferization pipeline designed to efficiently optimize RC-managed objects.
Check https://t.co/IMbpViOOvH
Your periodic reminder FFTW, the “Fastest Fourier Transform in the West,” is metaprogrammed in OCaml, and generates highly optimized C:
<https://t.co/3y8yt1Q4yp>
With 4 years industry experience. Specializing in traditional compiler mid-end optimizations using MLIR, with hands-on LLVM framework experience. Seeking opportunities to pursue a suitable PhD position in program analysis or "classical" compiler .🙏
A Review of 2025: The 'Little Miracles' in Two Years After Resignation
Overcoming every 'little miracle' in life, the journey of building confidence for the timid.
https://t.co/KtV9GI4V3u
P.S. The boss behind this language emphasized their “careful and rigorous” work ethic on this one with at least 3 follow-up posts in his social account.
I am more than proud of my PhD students, colleagues and collaborators!
Together we managed to have 5 papers at this year's OOPSLA'25 (R1 and R2) and 1 at the colocated ICFP'25.
🥳🎉
The papers cover a wide range of PL and compiler topics 🧵