Like many of you, I have spent four years trying to keep up with AI fast enough to use it. This is my first issue as its new author.

Last Friday you read that this newsletter was changing hands. That is me, and I am glad to be here. I am a former corporate lawyer and management consultant who turned himself into a full-stack applied AI consultant and entrepreneur. Nobody handed me that path. I read, I tried things, I built small things that broke, and I kept the parts that worked. You can do this too.

The weekly brief stays; I read the release notes so you do not have to. Every issue now adds one thing to do on real work this week and one small thing you get to keep. If there is something you want me to write about, reply to this email. I read every one.

Last issue said trust will decide which brands win with AI. How well you wire it into your work will decide whether you do.

For two years, using AI at work meant copy and paste. Email in, answer out. That is changing. AI now reads your calendar, your inbox, your threads, and your files, and then it does the task. The industry word for this is agentic: the assistant does the task instead of answering the question. The professionals who wire it in cut a forty-five-minute Friday report to ten, see more of their week at once, and get handed the bigger calls.

This week’s loop

Learn: the three layers that let AI into your week: connectors, MCP, and skills, the one you own. Use: the Friday status report, forty-five minutes to ten, with a source on every line. Build: turn the prompt into a saved skill, then give it your team’s template.

Learn: what changed that matters

Every major assistant now ships the same idea for people who do not write code: connect it to where your work lives, then ask for the task instead of the answer. Three layers make that work.

Connectors. What the assistant can reach. A switch in settings for Gmail or Outlook, calendar, Drive or SharePoint, Slack or Teams. Click, sign in, and it can read what you can read.

MCP. How tools plug in. Model Context Protocol is an open standard that lets any tool connect to any assistant, the USB port for AI. You will never build one; you only need to know that “we have an MCP server for that” means “your assistant can reach it.”

Skills. The layer you own. A skill is a small plain-language file that tells the assistant how you want a task done: which sources, which rules, which format. Save it once and it runs the same way every time, and you can hand it to a colleague. Anthropic published the format as an open standard last year; Claude, ChatGPT and Codex, Microsoft Copilot Cowork, Gemini, Perplexity Computer, and Grok Bot all have skills today.

Most of them will write the skill for you from a conversation. Microsoft’s own documentation example is a skill called Weekly Report: “Generates a weekly status report from my recent emails and calendar.” Each platform wires the three layers a little differently, so this issue stays platform-neutral and the coming weeks take each one apart.

Keep this: connectors, MCP, skills. The first two belong to the platform. The third belongs to you.

Use: do this on real work this week

The Friday status report. You rebuild it from a calendar, a sent folder, two or three chat threads, a tracker, some texts, and whatever you remember. Forty-five minutes on a good Friday, and most of it goes to re-reading your own week. That re-reading is the part the machine takes.

Before you start. Turn on calendar, mail, and chat connectors in the assistant your company allows; five minutes, once. Check your company’s policy on what can go in, and strip client names if you are unsure.

Do it

1. Gather the week. If you are connected, skip ahead. If IT has not flipped the switch, export the week’s calendar, copy the subject lines from your sent folder, and paste the team thread and the changed tracker rows into one message. Ten minutes. This is still copying into the chatbot; it is where you start, and what you graduate from the day the connectors turn on.

2. Run the prompt. Copy this and fill the brackets. It follows the CRAFT shape, context, role, action, format, tone, plus two guardrails, non-negotiables and sources. Prompting matters less than it did a year ago, but the shape is what becomes the skill.

Context: It is Friday and I owe my manager a status report for [Monday’s date] to [Friday’s date]. My team channel is [team channel]. My tracker is [tracker]. Role: You are my chief of staff. Action: Draft my weekly status report from the sources below. Format: Five sections, in this order: Done this week. In progress. Blocked or needs a decision. Next week. Needs confirmation. Every line names its source in parentheses (meeting title and day, email subject and day, or channel and day). After the report, list what you could not find. Tone: Plain and short, written for a busy manager. No adjectives. Non-negotiables: A line with no source does not go in the report. If something looks like a decision but was not confirmed in writing, put it under “Needs confirmation,” not under “Done.” If a meeting was cancelled or moved, do not report it as held. If you have to infer a status, say so in brackets. Never invent progress. Sources: My calendar for the week, my sent email from the same week, the [team channel] thread, and the [tracker] rows that changed. Use only these.

On the paste path, change the last line to “Sources: pasted below. Use only these.”

3. Read the parentheses. Every line comes back with its source attached:

Done this week: Q3 media plan approved by client (email, “Re: Q3 plan v4,” Tuesday). Needs confirmation: vendor cut described as “probably fine” (Slack, #acme-team, Wednesday). Not confirmed in writing.

Check it. The machine read for about a minute. You spend nine on the lines, and the source tells you where to look. Expect it to be wrong somewhere, and expect the wrong places to repeat:

  • It will report a cancelled meeting as held.
  • It will promote a Slack aside into a decision.
  • It will write “on track” about something you never said was on track.

Every one of those is a place you now know to look. You learn that by running the thing and watching it break.

The clock: ten minutes, down from forty-five, with a source on every line your boss can click.

Keep this: a line with no source does not go in the report.

Build: something you can keep

Three rungs on one ladder. You only need the first this week.

Rung 1, this week. Save the prompt as a skill named “Friday report.” On Copilot Cowork, Claude, Perplexity Computer, or Grok Bot, ask the assistant to save it as a skill; on ChatGPT Work or Codex, call the skill-creator. If skills have not reached your platform, save it as a named prompt; the text is identical.

Rung 2, next week. Add your one-page status template to the skill as a companion file. Now the skill says “use the attached format,” a colleague can run it, and your manager changes the headings once for everyone.

Rung 3, when you are ready. Schedule it. Copilot Cowork already takes “create a weekly status report every Friday” as an instruction, Grok Bot calls it a routine, and Claude runs scheduled Cowork sessions. That is a later issue.

You do not need to become a technologist for this to pay off. The small thing you built taught you where the machine goes wrong, and that still applies when next quarter’s tools arrive.

Keep this: the prompt becomes a skill, and the template rides along in its folder.

This week, in order

  • Turn on connectors, or set aside ten minutes Friday to paste.
  • Run the Friday report prompt once. Read every parenthesis.
  • Save it as a skill, or a named prompt, called Friday report.
  • Reply and tell me which platform you are on.

Reply with your platform: Copilot, Cowork, ChatGPT, Gemini, Grok, Perplexity, or something else. That decides the order of the deep dives, and the deep dives are how you wire it in. Same time next Friday.

All writing Builds