Nick Loui

1/ I've spent my career building technology and leading digital marketing teams. The techniques we used to reach audiences and grow businesses are being used to manipulate people. That realization led me to co-found @peakmetricsapp.

GitHub says its new storage handled 5,000 agents on one repo in internal tests. Most of each write runs independently. Only reference updates need agreement.

For agent workflows, where do you actually need coordination?

github.blog/engineering/ar…

If an AI answer calls your product “hard to use,” what do you fix?

First, inspect the answer’s sources. Then check whether the criticism holds up.

We checked the claims that the Yandex strikes silenced Russian influence operations.

Our team at @peakmetricsapp looked at suspected bot activity across 20 topic areas on October 11, comparing it with nine previous Sundays. That matters because weekend activity is often lower, even without an outage.

18 of the 20 stayed within their usual Sunday range. Two were below it. We didn’t see the broad collapse the headlines suggested in our sample.

The Cloudflare chart behind the story measures automated web requests from Yandex’s network, which can include search crawlers. It doesn’t tell us which social accounts are run by Russian operators, or whether those accounts stopped posting.

Before declaring an influence operation offline, we need evidence from the accounts and conversations it was driving. A sharp drop in web traffic alone doesn’t get us there.

radar.cloudflare.com/bots/as13238?d…

At @peakmetricsapp, @typesafeai’s Jev showed ~82% lower cost per doc and ~47% lower median call latency vs our baseline. A blind LLM judge preferred Jev in 69% of 200 disagreements. Workloads and batch sizes differed, so these are directional, but very promising results.