the guide I wish someone sent my colleague (AI edition)
welcome, back friend.
Short on time? Read the first two sections and the last one. Bookmark it, come back to it.
A friend of mine just came back from mat leave and messaged me in a bit of a spiral. Someone in her first meeting back mentioned “MCP servers”. She’s in marketing, not eng... her whole team is using Cursor (sames!) and she doesn’t know what Cursor is.
I’ve been here working away the whole time (with an AI newsletter, too!) and I still feel overwhelmed and lost some of the time.
So I wrote her a very long message on tools to check out, people to follow. And then I thought… this should probably just be a newsletter. Because she’s not the only one coming back to a workplace that looks completely different than when she left. And the panic of “I’m so behind, I’ll never catch up” is mostly wrong. So let me walk you through what I think actually changed, especially if you are in ops, marketing, or any other non-technical role.
Thanks for reading! This post is public so feel free to share it.
most of what matters? you already have it.
You and AI can get a lot done together now. The second draft of a brief, a competitive landscape, a campaign outline, a deck structure — things that used to take days can take an hour. You bring the thinking, the context, the initial instinct. AI helps you build on it, pressure-test it with data, shape it faster than you could solo.
How much AI can do vs. how much needs to be you depends entirely on the task. Rough percentages from my own + friends’ experiences (as of March 2026 … this can change very quickly..):
Research summaries, meeting recaps — AI can do ~80% of the lifting. You edit and steer.
Email drafts, internal comms — Maybe ~60-70%. AI gets you a solid start, but tone and context are yours to finesse.
Campaign briefs, strategy decks, competitive positioning — Closer to 50/50. AI can help you structure thinking and fill in data, but the actual strategic point of view is you.
Brand voice, creative concepts, writing that needs to feel human — AI is maybe ~20-30% of the work. It can brainstorm with you, but the taste, the instinct, the “this doesn’t sound like us” — that’s all you.
Reading the room, managing stakeholders: 0%? I trust us as humans on this more. :)
That said — I’m working hard to not start with AI. I wrote a post about this if you’re interested. I was running to ChatGPT/Claude (more on these guys, below) as my first stop for and skipping the most important step: my own brain.
Everyone has access to the same AI. Your taste, your creativity, your context — that’s the differentiator now.
it’s not a secret tab anymore
When I first started using AI at work, I’d minimize ChatGPT when someone walked by my desk. I felt almost embarrassed — like I was cheating on a test, like using it to help me edit a brief meant I couldn’t actually do my job.
That era is over.
Walk around any office right now. Cursor is open. Claude Code is running in terminals. People have custom GPTs pinned. Teams are building internal tools over lunch. It’s not a hack or a shortcut — it’s how work works now.
The awkward phase is done. It’s now about how to know when to best use it, what tool is right for what problem, and to also now know when to step away.
the mental shift
Specific tools may come and go. I don’t think these below frameworks will.
It’s an exoskeleton, not a coworker. I love this framing from Ben Gregory. We keep trying to think of AI as a colleague or an assistant — some separate entity that “does work” for us. But the better mental model is an exoskeleton. Ford uses actual exoskeletons on their assembly lines — the worker still does the overhead lifting 4,600 times a day, but with mechanical support that reduced injuries by 83%. The human is still doing the work. They can just do dramatically more of it, more sustainably.
That’s AI. It doesn’t replace your judgment or your taste. It amplifies your capacity to use them.
The question isn’t “what can AI do for me?” It’s “where do I experience the most strain, and how can AI support that specific point?”
Decompose tasks, not roles. Also from the same piece, don’t ask “can AI do a marketer’s job?” Ask: what are the 30 things I do in a given week, and which ones could be amplified? Summarizing meeting notes, researching competitors, drafting first versions of briefs — those are amplifiable. Deciding what message matters, reading a room, understanding why a campaign didn’t land — those are you. When you break it down this way, AI stops feeling like a threat and starts feeling like a relief.
The jagged frontier. This is Ethan Mollick’s concept (Wharton professor). AI capabilities aren’t a clean line where it’s good at “easy” stuff and bad at “hard” stuff. It’s jagged. AI might pass the bar exam but fail at tic-tac-toe. It can write a decent first draft of a strategy doc but completely botch a simple math problem. The only way to learn where the frontier is for your work is to throw AI at everything and see where it shines and where it face-plants. That’s not a sign of incompetence — it’s the learning process.
Centaur or cyborg — pick your style. Also Mollick. A centaur draws a clear line: “I’ll do the strategy, AI does the summarizing.” A cyborg weaves it in constantly — starting a sentence and letting AI finish it, drafting together in real time. Most people start as centaurs and gradually become more cyborg-like. Neither is better. Both are better than not using it at all.
The 50x reframe. From Azeem Azhar’s Exponential View. Most people ask AI: “How do I speed up what I’m already doing?” Instead ask: “What would I do if I had 50 people working on this?” Then work backwards. Which parts of that dream team’s work can be simulated with AI? This stops you from making your existing process 10% faster and starts you thinking about what’s actually possible now.
If you keep having the same conversation with AI, turn it into a tool. Also Azhar — If you find yourself explaining your brand voice, your audience, your goals to AI over and over, that’s a signal. Build it into a Claude Project or Custom GPT with all your context pre-loaded. The first conversation is exploration. The fifth conversation could be a system.
the buzzword decoder
You may hear new terms thrown around in meetings like everyone’s been saying them forever. They haven’t. Most people learned these words like two months ago. Here are a few that may pop up:
Vibe coding — Building software by describing what you want in plain English and letting AI write the code. You don’t look at the code. You just… vibe with it. It’s why non-technical people are suddenly building apps.
Agents / agentic AI — AI that can do multiple steps on its own — not just answer a question, but go research something, pull data, take actions, and come back with results. We’re early on this but it’s the direction everything is moving.
MCP (Model Context Protocol) — A way to connect AI to your other apps — Calendar, Slack, CRM — so it can actually do things in your real workflow, not just in a chat window.
Prompt engineering — Writing good instructions for AI. Think of it as good creative briefing — which you already know how to do.
Hallucination — When AI confidently makes something up. Always check its work, especially for data and quotes. It’s not a bug that got fixed — it’s how these models work.
RAG — When AI pulls in your company docs before answering, instead of just going off what it was trained on. When someone says “we’re using RAG,” they mean “we taught the AI about our stuff.”
my toolkit (as of right now)
These will change. New tools will launch every week. Some will shift your entire workflow; most will not. Think of this section as a snapshot.
Start with Claude. If you only set up one thing, make it this. Claude (by Anthropic) has been pulling ahead in a way that’s hard to ignore — the writing is more natural, it holds context better, and the ecosystem around it is where the most interesting stuff is happening. The features worth knowing about:
Projects — Upload your brand docs, style guides, past briefs, whatever defines your work. Write custom instructions. Every chat inside that Project has all that context. Claude doesn’t forget. It doesn’t need to be re-briefed. I have separate Projects for different workstreams and it’s like having a specialized teammate for each one. This is the single highest-leverage thing you can set up in your first week.
Deep research — Ask Claude to go deep on a topic and it spends several minutes actually researching across dozens of sources, synthesizing findings, and delivering a report with citations. Five-hour research assignment back in five minutes.
MCP integrations — Claude connects directly to Google Calendar, Gmail, Slack, Granola (best meeting transcribe tool IMO), and other tools. Ask it to check your calendar, summarize recent emails, search Slack channels — without leaving the conversation.
Memory, web search, artifacts — It remembers you across conversations, searches the internet in real time, and generates documents and code in a separate panel you can edit and iterate on.
If your company only has ChatGPT Enterprise or Gemini: Not every company has Claude. If your org is on ChatGPT or Google’s ecosystem, you’re still in great shape — just know where to focus.
ChatGPT Enterprise / Team:
Custom GPTs — Same idea as Claude Projects: upload your brand docs, write instructions, create a specialized ChatGPT for a specific workflow. One for brand voice, one for data analysis, one for social copy. Your team can share them.
Deep research — Works the same as Claude’s. Great for competitive analysis and catching up on what you missed.
Canvas — Opens a side panel for longer documents where you can highlight a section and ask it to rewrite just that part without regenerating the whole thing.
Voice mode — Still unmatched. Surprisingly useful for brainstorming on your commute or talking through a problem away from your desk.
Gemini (Google):
Google Workspace integration — Gemini’s superpower. It drafts in Docs, builds formulas in Sheets, creates presentations in Slides, and summarizes email threads in Gmail — all without leaving the apps you already use.
NotebookLM — A standout. Upload PDFs, docs, or links and it creates an interactive research notebook you can chat with. It even generates podcast-style audio summaries, which is kind of wild.
Deep research — Available here too. Same concept.
Enterprise search — Gemini can search across your company’s Google Drive, Slack, Confluence, and other connected tools. “What did our team decide about X last quarter” actually works if your org’s data lives in Google’s ecosystem.
One tip for all AI tools: give them context before asking for output. Don’t say “write me a blog post.” Say “here’s our brand voice guide, here’s our audience, here’s an example of a post we loved — now write me a blog post.” The more context, the less editing.
Now the big one: you can build things. This is the part that I am the most excited about. Non-technical people like us — marketers, ops leads, content strategists — are building our own tools. Not learning to code. Describing what they want and watching AI build it.
Cursor — A code editor with AI built in, but don’t let the word “code” scare you. You open a file (a landing page, a doc, a spreadsheet), describe what you want to change in plain English, and it does it. Marketers are using it to tweak website copy without filing an engineering ticket. Ops people are building internal dashboards. It collapses the distance between “I have an idea” and “it exists now” in a way that used to require a developer. Check out “The non-technical PM’s guide to building with Cursor” by Zevi Arnovitz (Meta) via Lenny’s Newsletter to kick start your learnings.
Claude Code — This is the one I’m most excited about. You can build entire tools from scratch just by describing what you want. I’ve built a push notification copy checker for my team, an icebreaker randomizer app, and even my personal website into a 2000s rainbow explosion under and hour. All without writing code. The setup can be a little tedious — but once you’re in, you can build truly anything you used to need an engineer for. This “Intro to Claude Code (for non-engineers)” Maven Lightening Session, hosted by Ryan Smith, helped make setup much easier for me.
And if both of those feel like too much right now? Lovable, Bolt, and Replit let you describe what you want in a chat interface and it builds a working app in your browser. No terminal. No setup.
Tools that round out the stack: Perplexity (search that actually gives you the answer), Gamma (gorgeous decks in under a minute from a messy outline), Granola (my favorite meeting notes tool), Descript (edit video by editing text), Slack AI (summarize channels you missed), Zapier AI (build automations in plain English).
Please do not try all of these in one week. Pick one. Maybe two.
If this is helpful so far, share it with someone who’s about to come back from leave. Or someone who’s been back for a month and is still quietly spiraling. We all know that person. Maybe we are that person.
how to keep up without it consuming you
This isn’t going to slow down. Every week there’s a new tool launch, a new model drop, a “this changes everything” post on LinkedIn. Some of it will be real. Most will not be relevant to you. I say this as someone who has a notes app full of tools I swore I’d try three weeks ago. The FOMO is real. But it’s manageable.
Most launches don’t matter for your work. Every few months, something comes along that actually shifts how you do your job. Cursor was one. Claude Code was one. But in between, there are hundreds that are interesting and fine to ignore. The skill isn’t keeping up with everything — it’s recognizing the signal when it shows up.
Pick two or three sources and let them filter for you. You don’t need to be on AI Twitter. Subscribe to a couple newsletters, check them when you have time, let the rest go. If something is actually important, it’ll find you through multiple channels. If you only saw it once on a random LinkedIn post, it probably isn’t urgent.
Set a “learning window.” I spend about an hour a week using new tools — not reading about them. Sometimes it’s 20 minutes on a Tuesday. That’s enough. The people who look like they’re keeping up just have a rhythm.
Remember what you already know. Every generation of tools has had this moment. Social media had it. Remember when everyone was panicking about Clubhouse? The chaos always settles into a smaller set of things that actually matter. We’re in the chaos phase. It will settle.
The goal isn’t to keep up with AI. It’s to keep up with what AI means for your work. Those are very different things.
resources I’d send a friend
Free courses: Anthropic Academy · OpenAI Academy · DeepLearning.AI “AI for Everyone” · Google AI Essentials
Reads: Everyone Should Be Using Claude Code (Lenny Rachitsky) · Claude Code for GTM Teams · AI as Exoskeleton · Six Mental Models for Working with AI (Azeem Azhar)
Newsletters: Carley Lake (that’s me :) ) · One Useful Thing (Ethan Mollick) · The AI Exchange (Rachel Woods) · Exponential View (Azeem Azhar) · The Neuron · Superhuman (Zain Kahn) ·MKT10 (Emily Kramer)
Podcasts: Hard Fork (NYT — best weekly AI catch-up and the hosts are entertaining/funny too) · Lenny’s Podcast (interviews with Cursor founders, Perplexity CEO, etc.) · Late Checkout (Greg Isenberg)
one more thing
You took time to take care of yourself and/or your family. And now you’re back with a perspective that nobody who stayed at their desk has.
Time away from the noise does something I believe that no AI tool can replicate. It gives you fresh eyes. A different lens on what actually matters, what problems are worth solving, what’s urgent vs. what’s just loud. Your ideas, your time away from the constant scroll, your new role as a parent or caretaker — I’m excited for you. And the fresh thinking you’re going to bring to how you use these tools, what problems you’ll solve, what you’ll build for yourself and others.
So I hope you’re leaving this with a little more excitement than dread. And if you want to jam, bounce ideas, do a walkthrough of any tool, feel free to write comment, DM me, reply to this email.
Glad you’re back!
Know someone coming back from leave soon? Send them this. It’s the guide I wish someone had sent my colleague.
Thanks for reading Carley Lake! If you’re new here, this newsletter is for people who want to level up with AI and wind down without it. Subscribe for free.