Some of the work behind this newsletter is done by AI agents. I still decide what goes out.

That is not the interesting part. The interesting part is how much of my AI use has nothing to do with coding.

I use Claude Projects to help family and friends build websites. I have Grok AI agents that manage this newsletter and keep track of AI news for me. I use chief-of-staff style assistants. And every week, AI agents review my LinkedIn analytics.

None of that is a demo. It is the ordinary stuff of a week: a favour for family, a side project, the admin, the weekly numbers. That is where I now think most people reading this have the most to gain.

Most of the AI conversation for career changers is about coding: which tool writes better functions, which model passes the take-home. Fair enough. But you spend far more of your week on recurring admin than on writing code, and that is where an assistant keeps paying back.

Here is what I see a lot with career changers. They will happily let AI refactor a module, then spend a Sunday evening rebuilding the same job tracker by hand, rewriting the same follow-up email, re-reading the same analytics page.

The task that repeats every week is the one worth handing over, precisely because it is boring.

I built ResumeRadar, now at 7,700+ CVs scanned, and I still think the most useful thing any tool can do is take a repeating chore off your plate.

List what repeats. Everything you do weekly: applications logged, notes tidied, news skimmed, numbers checked. If it happens every week, write it down.

Sort each one into draft, track or review. An assistant can draft (a first version of a message or a plan), track (watch a source and tell you what changed) or review (read something you produced and flag what is off). Nearly every recurring task fits one of the three.

Pick one and hand it over fully. Not a prompt you type each time, but a setup you build once: a project with your context in it, a scheduled AI agent, a template with your rules. Handing over half a task saves nothing, because you are still holding the whole thing in your head.

Decide what you still check yourself. This is the part that keeps it honest. AI agents help run this newsletter, and I still read what goes out. Pick the decision you will not delegate and write it down before you build anything.

Repeat the audit once a month and the list of things you do by hand shrinks in the right order: boring first, judgement last.

Claude and OpenAI models come to Google Cloud. An Anthropic-run Claude option is now generally available, and OpenAI models and Codex will arrive in the coming weeks. What it means: prompting skills now travel across models.

Claude stops asking before every step. New sessions in most Claude apps on Pro, Max and Team plans are switching to auto mode this week: an automated check approves safe actions and blocks risky ones. What it means: learn how agent permissions work.

Researchers stress-tested OpenAI's agent monitoring. METR and Redwood Research found several new weaknesses. OpenAI says the production flaws are patched; the researchers have not checked. What it means: logging skills carry into AI governance.

AI helped researchers find critical flaws in PHP, which runs much of the web. Every version from the last decade is affected, and CISA is urging an immediate patch. Attackers are now exploiting it, and CISA added it to its list of known exploited flaws on 26 October. What it means: patching fast is a core cloud skill.

This week's sponsor is Atlassian. Their State of Product report looks at what AI still has not solved for teams, which is a useful reality check before you hand any task over.

80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.

Read the report

💬 What is the one task you do every single week that nobody would notice if an AI agent did it instead?

Reply with EDGE and the task. I will tell you whether I would draft, track or review it, and how I would set it up. I read every reply.

Know someone who only uses AI for code? Share your referral link. One friend subscribes, you get a free LinkedIn Profile Optimisation Checklist.

{{rp_refer_url}}

Shola

P.S. If you want to list your weekly chores and mark which ones an assistant could draft, track or review, here is the Everyday AI Audit (1-page PDF), direct link, no signup. It also works with a team: suggest it to your manager, fill in one each, and compare lists at your next team meeting. A chore that shows up on several sheets is a good one to hand over first. Next Tuesday: another lesson from my own notebook. 6PM UK.