Umair

Anthropic just shipped Opus 4.7 today and i had to write about it somewhere because the jump is weird.

I ran the same backlog task on 4.6 and 4.7 back to back. same repo, same prompt, same tools. 4.6 looped on a bug for 25 minutes and was not going to solve it. 4.7 closed it in eleven, and the part that freaked me out is that it paused in the middle to sanity-check an assumption i had not asked it to check. literally wrote "before i write this migration, let me verify the actual shape of the response object, because my assumption here might be wrong" and then went and verified it. unprompted.

That self-verification behavior is the thing. Vercel is reporting it does proofs on systems code before starting work. Hex says it flags missing data instead of making up plausible-but-wrong fallbacks. Genspark measured loop rates on hard queries and 4.7 basically stopped looping. different teams, different harnesses, same pattern.

the numbers are nuts too:

YC-backed @Mintlify (YC W22) just announced a $45M Series B round, bringing their total funding to $67M, to "accelerate [their] mission of building the knowledge infrastructure for AI."

Read in their blog announcement:

Mintlify now powers documentation for over 20,000 companies, with content reaching more than 100 million people every year. This round accelerates our mission to become the knowledge layer that makes products understandable, usable and discoverable by AI agents.

TL;DR: .agent is the most strategically important TLD still without an owner. ICANN's application window opens in weeks. A company is going to bid for it - unless a community claims it first. Here's the story, and two questions I'd actually like pushback on.
---
Agents already need real addresses. I've been working on this for over a year, and the pitch just keeps getting simpler. Right now agents mostly live at a subdomain of whoever built the framework platform.com/yourname-7. Borrowed identity, borrowed trust.
If you want the concrete version, look at OpenClaw. An autonomous agent running on your machine with its own email (real OTP codes, real password resets), OAuth credentials across hundreds of thousands of SaaS apps, sub-agents spawning and talking to each other, webhooks, the whole thing. A worker, not a demo.
And in its first three days live, OpenClaw had to rename itself twice. The name it launched under - gone. That's the whole argument in one news cycle: agents running real lives from borrowed names are one legal letter away from disappearing.
.agent is the most strategically important TLD still without an owner. ICANN's next gTLD round opens in weeks. And once .agent is claimed, it's claimed - the internet's naming system doesn't hand these out twice.
If one company wins it, .agent becomes their product. They'd set pricing. They'd set policies. They'd decide who gets yourname.agent and who gets blocked. One company choosing shareholder interests over an open internet - because that's literally what it would be.
The community bid is a formal ICANN community application its own specific path with its own process. The goal, if approved: keep .agent open infrastructure. Open standards for agent discovery. No gatekeeper. Governance by the people actually building agents, within ICANN's rules.
23,000+ members have joined. Esther Dyson, who used to chair ICANN, and Illia Polosukhin, who co-wrote "Attention Is All You Need," are advising. 
It's not done. ICANN scores community applications on size, governance, nexus, and endorsement depth. You need 12/16 points to beat the corporate applicants who are absolutely going to file. Every signal matters.
Two questions I'd genuinely like pushback on:
1. Is the naming layer for AI agents something the community should own, or is it fine if it goes corporate? I have a strong view, but makers building agents every day see things I don't.
2. If you think it should stay open - what governance rules would you want locked in from day one? What would make you still trust the TLD in 5 years?
If this resonates, the non-binding endorsement is here (30 seconds). The one-pager has the deeper version.
Either way, would love to hear what you think. Especially the pushback.