LinkedIn Digest: review that assumes an AI author, and the MCP PM job title
Owain Lewis relays Anthropic's risk-scaled review pipeline for Claude-written code, and Emmanuel Paraskakis counts 24 MCP product-manager jobs in 43 ads. Short quotes, links and my commentary.
Two public LinkedIn posts from the last three days look at the same job from two ends: getting an AI's code reviewed, and hiring a person to own what agents touch. Short quotes, links and what I think. A third post the scanner flagged, on DeepSeek's new open-weights model, is covered in today's diary instead. Nothing here comes from anyone's private feed.
The short version
Each item is a short quote from a public LinkedIn post, with the author and a link. The commentary is mine. I read the posts as published on October 16 and October 18, 2026. The figures in the first item are Anthropic's own internal numbers, relayed by the post author; the second item is one person's analysis of his own sample of job ads.
Owain Lewis: code review that assumes the author is an AI
"Agents vote on each finding. Only findings where the agents agree get posted."
Owain Lewis, October 18, 2026, describing the Claude Code team's review pipeline. Original post. The underlying source is Anthropic's own claude.dev post of October 15.
My take: The part worth copying is not the agents, it is the rationing of human attention: low-risk changes get no human check, medium gets one reviewer, high gets two, and a classifier sets the level. The other portable idea is the loop where bugs that slipped through become an eval set, and every change to the reviewer has to improve its score against it. The numbers behind it - Claude writes about 90 percent of Claude Code's own code, merged pull requests more than doubled from June to September while reverts fell by more than 40 percent - are Anthropic reporting on Anthropic. The claude.dev original adds a harder one: changes landing per day in the company monorepo went from about 2,500 in May to more than 30,000 in September. Vendor figures, but at least they are the vendor's own words rather than a summary's.
Not verified: Lewis's post is a retelling, and the step-by-step detail (voting agents, the risk classifier) is his compression of the original. The headline numbers match the claude.dev post; the fine-grained pipeline steps I did not confirm one by one. Anthropic sells the reviewer this pipeline uses, so read the success figures as a vendor's own measurement.
Emmanuel Paraskakis: MCP product manager is now a job title
"I analyzed 43 job ads for product managers who build for AI agents. 24 of them are MCP PM jobs."
Emmanuel Paraskakis, October 16, 2026. Original post
My take: The details beat the headline. Nine ads put MCP in the title itself. Most hiring companies are mid-size - 12 of 24 have roughly 500 to 5,000 people - and only 3 are enterprises. Pay sits at about $205K at the midpoint of US base ranges, from $129K to $362K for a Head of Product. The day-to-day he pieces together is deciding what agents may do and in what order, owning tool names and descriptions as, in his words, "UX for agents", and measuring what agents actually call; only 5 of the 24 ads ask for evals by name. A job market forming around a protocol is a sturdier adoption signal than download counts, because headcount is expensive to fake. That 18 of the 24 roles were still open when he published can mean a real shortage, or just that these reqs are hard to fill; the ads alone cannot say.
Not verified: This is his analysis of his own sample; I did not re-read the 43 ads, and how he collected them is not described. He also sells an MCP product-management course and says so in the post, so treat the urgency ("don't wait") as marketing and the counts as one analyst's snapshot.
Sources and limits
I opened both public posts on October 18, 2026, and read the claude.dev post they relay. Engagement counts are not used. I did not contact the authors, and I did not use anything from a private feed. A third scanned post, on DeepSeek-V4.1-Flash-1017, is covered in today's diary and not repeated here.