stefan_lec

Yeah this resonates with me. I think of it more as an “understanding” sandwich:

1) While pointing the agent at the task, poke it with annoying questions until I understand the problem and its context well enough to explain it.

2) Have it write the solution, help it work past problems it encounters during testing, then go through simplification, performance and correctness reviews.

3) Have the agent draft a pr, then poke it with annoying questions about the PR until I understand where we’re at, and why we don’t need to do more work. Also try to rewrite it into readable prose instead of Claud-ish while doing so.

#2 and #3 can repeat quite a bit, depending on how stubborn the problem is.

AA isn’t testing different harnesses there, they’re testing different models on the same common harness:

“We run Terminal-Bench 2.1 with the Terminus 2 agent harness in an e2b sandbox and report pass@1 averaged over 3 repeats per task.”

This is a different agent harness than those Strands was testing with, and there’s no guarantee the pass criteria matches up the same either.

Yes, but a big reason payloads cost so much is the need for extreme mass optimization, which is driven entirely by limitations in launch technology.

When launching large masses to high energies gets cheaper, payloads have a chance to get cheaper, too.

How do these perform for cold-starts? I remember one of the disadvantages of using web assembly for Workers was more spin-up time, but maybe they figured out a way around that.

Have you tried Astro? It lets you do exactly that: server-side rendering with HTML templates in component form, with css scoped to the component. No shadow dom needed, no JS frameworks either.

Very handy for nav bars with expandable drop-downs for more links - can now be done with zero JS. Requires much less CSS boilerplate now, and you don’t have to use z-index at all (which eliminates a whole bunch of tricky issues). They also handle all the tricky aria/keyboard/focus mgmt bits for you, too. It works really, really well.

Supposed to be real handy for tooltips too, but haven’t used that myself yet.

Edit to avoid potential confusion: these are separate from modal dialogs, think more like very small popups anchored to one point on the page that still allow you to interact with everything else. Theoretically you can make them modal, but it would be silly to - you’d be better off using <dialog> and the invoker api

Something about this feels very Tolkien-ish.

Nobody mention it to Peter Thiel, he doesn’t need any more ideas.

Where’s my nanometer-sized violin? Always losing that darn thing …

Sounds pretty cool! I wonder if it closes enough holes that we could finally stop using SPF at all?

Because they couldn’t get the cats to pose obediently for the artist?