I wanted to understand where the capabilities of the various agent protocols and implementations actually overlap, so I had Claude map them out onto a capability map, now sharing it in case it's any use for others.

Warning: Yesterday an "investor" booked an intro call with me, and the call turned out to be a delivery mechanism for fake meeting links, I assume to take control over my device.

. @Google's @GeminiApp AI summaries not getting extreme basics like the correct time really does not inspire confidence in AI... Thanks to @EHotta for flagging this!

Three weeks, three data points on how much we trust the companies building personal agents:

I got around 20 reactions on a post in under ten minutes on @LinkedIn this week, almost all from leadership coaches, and I couldn’t tell whether a person, a pod or a script was behind any of them.

"Sorry, it wasn't me, guv" doesn't work as a legal defence in California anymore, and @AnthropicAI’s own IPO filing shows just how critical this is.

First heard of independent verification organisations, or IVOs, from @BriTreece_, (co-founder and president of @fathom_org and executive director of @pactai_org, on a @linuxfoundation webinar launching “Proof of Control” (congrats @triciawang and @aai_society)

Great to see agent trust on @cheqd_io getting started. This follows a stretch of assessing a lot of fast-moving repos👇

I wasn't expecting to be using product shipping as a metaphor for how AI is reshaping software agencies, but here we are.

@Meta's @Muse agent has had two very different weeks in one… Launched September 8, it passed 2.5 million downloads inside two weeks and overtook @ChatGPT on the US iOS App Store. Meta's stock followed: up roughly 30% this month and an 11%+ single-day move on Monday.

Have been down a small Jev rabbit hole since @typesafeai released it, this time on the evals and governance side rather than the angle I posted about last week.

@typesafeai opened access to Jev this week, and one of their own engineers described it, half joking, as "a really smart switch statement": 2016-era classifiers wearing 2026 levels of capability, which is a more honest description of what's new here than most model launches

I spent Sunday morning cycling round Regent's Park with a friend, talking about trust. He's a senior engineer at a scale-up in an industry where trust isn't a nice-to-have. Insurance, finance, the kind of industry where "it mostly works" gets you fired...

Two repos, donated to two different industry bodies, roughly three months apart. Two completely different pictures of what "donated" means in practice.

I'm mostly against humanising AI agents, but I've started using the language of contractors to talk about them and finding it helpful. For two reasons: First: the AI agent market appears to be mirroring the worker onboarding journey, you: - Hire them (pick a model).

Hit a small milestone on @Authrty : we just crossed 500 installs across @GitHub and @PyPI for a free tool we released a few weeks ago (link in the comments).

The two leading AI companies had a bad week but I seem to be seeing and hearing a lot more about @OpenAI's breach rather than @Anthropic's data leak.

Noticed two things in the reaction to @OpenAI ’s sandbox escape👇

In the early stages, most product feedback doesn't arrive nicely formatted in a form you can act on.

Had a few days off and used them to play around / built on top of Tarek, CTO of @Authrty 's Claude Code analytics tool, obviously with Claude doing most of the actual coding. (link to Tarek's tool below)

Nick Lambert from @dock_labs's post on @LinkedIn on "derived credentials" put into words something I've been chewing on for a while: EUDI is critical infrastructure, but I don't think it's where the value accrues.

Spent the last two afternoons at @AICodingSummit , in person Monday and virtually Tuesday, alongside developers from @OpenAI, @Microsoft @amazon, @LSEGplc and many more. A few takeaways 👇

Two different races are running inside EUDI Wallet rollout right now, on two different clocks.

Posted about this yesterday, then went down a bit of a rabbit hole last night, a productive one at the very least.

Over 50 customer interviews now and one topic keeps coming up before anything else. Not risk, governance or compliance. Cost... Everyone seems to be assuming governance starts with “is this safe” and “are we compliant.”

Anyone else finding themselves migrating spreadsheets off Google over to @notion.com purely so @anthropic.com's Claude can enable a workflow? Pretty sure I've had Claude replicate a spreadsheet into notion at least 5 times in the last weeks

I had a hunch there were plenty of Agent ID and credential specifications and standards emerging but maybe hadn't appreciated exactly how many, I came across >10 just on a first pass, I'm sure there are many more.

eIDAS 2.0 and EUDI conversations appear to be shifting gear. cheqd built a tool to model exactly the question everyone's now asking, link in the comments.

An engineer friend at a scale up shared their company hit their company-wide token limit last night… so it’s not just Uber Anthropic dropped Fable 5, free on existing plans, so seems naturally everyone just switched it on.

Anthropic just published a Zero Trust framework for AI agents. It’s the most thorough treatment of the problem I’ve seen from a model lab. Cryptographic agent IDs at Foundation. X.509 certs at Enterprise. Hardware-backed attestation at Advanced. All correct. All necessary.