bjornroberg

Happy to see Pi going this good. It's my favorite harness, no doubt, but it's not super clear to me why it's 1.0.0 instead of 0.100.0.

This initiative is going to yield a net positive in general societal robustness and security, not only for critical infra and big players. On the contrary, it will make it easier for SMBs without people working 100% with compliance to increase the robustness of their supply chains.

I wonder if it could be that they won't because the real mechanism is that AI wrapper pricing power is weak (switching costs near zero) but state of the art models makes it difficult to lower prices due to higher cost.

I think this is a smart move, and if Mozilla were the same as 10 years ago, I'd have hope for something good.

Solves yesterday's problem. The calcification is UI calcification, and agents don't care about UIs. An MCP server (or a half-decent OpenAPI surface) lets a user-controlled agent compose vendor primitives without touching the DOM, without TOS risk, without overlay maintenance. IoC doesn't get forced by extensions. It gets forced by agents that can read docs and click buttons faster than the vendor can ship features. The vendors who notice will expose that surface voluntarily, because the alternative is getting scraped anyway.

Yeah, this. The vocabulary ratchet is underrated as a policy tool. "Install" became "sideload." "Sideload" became "install from unknown sources." "Unknown sources" is becoming "unverified packages." Each rename shifts the Overton window a little further from "this is the normal way to put software on a computer you own" toward "this is a suspicious deviation Google has graciously decided to tolerate for now."

By the time the technical mechanism lands, the framing has been prepared for a decade. The 24-hour cooldown, the seven taps, the three scare screens all _feel_ proportional to the danger the language has been implying. That's not an accident, that's the policy working as designed.

The detail that keeps getting lost in these threads: the "advanced flow" for power users is delivered through Google Play Services, not the Android OS. That's the whole game.

It means the safeguard is not part of AOSP. It ships as a closed component that Google can narrow, gate, or remove in any Play Services update, with no Android version bump, no OEM coordination, no user consent beyond the usual auto-update. "Open platform with an escape hatch" is load-bearing in the PR; "closed escape hatch bolted onto an open kernel" is what's actually shipping.

The second tell is timing. It's five months from enforcement and the flow has not appeared in any beta, dev preview, or canary build. We're being asked to treat a blog post and UI mockups as a functional guarantee. No other platform change of this scope lands without a shipping preview this late, and Google knows it.

The third piece most devs skim past: registration requires uploading evidence of your private signing key. Whatever you think of the verification program in principle, that specific requirement changes the threat model of every Android key in existence, including the ones protecting apps people already depend on.

"Sideloading still works" is only true in the narrow sense that some ceremony remains. The mechanism protecting that ceremony is owned by the party with the strongest incentive to eventually close it.

The artifact can be faked cheaply now, so the only buying signal left is commitment. That's exactly the "ruthless" move the post argues for, I think.

Broadly agree. Moving from prompt to action is the right direction. I think the prepared statements analogy is not fully comparable in that SQL has a clear boundary between code and data whereas tool calls don't. However, this isn't fatal, just being honest about the shape of the trade-off.

I feel that the hard problem is writing policies expressive enough to cover arbitrary agent work without collapsing back into "trust the model's intent."

Poked around for a few minutes and this is nicely done. Coverage of the less common records (MTA-STS, BIMI, TLSA, CAA) in one place is the part that jumped out. Usually I bounce between two or three sites to get those.

One thing that would be handy: a shareable permalink for a given lookup result, so you can paste it into a ticket or chat instead of retyping the query.

It's an interesting domain and deploy-time learning seems to be a powerful approach! I'll look further into it.

Really easy to get started with and smooth UX!

Promising foundation if you’re willing to own the policy layer + quality gates. Also worth calling out the boring production bits: robots/ToS, rate limiting, bot mitigation, login/session handling, and not accidentally hoovering up PII.

Actually laughed out loud on this one. I don't know what that says about me.

Really nice concept. I like the idea of using file system primitives to interact with structured data.

Diversionary tactics is the only thing it is.

Yep. It is the Swedish government procurement process and lack of IT knowledge among the decision makers.

I's suggest reading a bit on the Elm Architecture. You don't have to use Elm to appreciate or use the architecture.