Welcome to Day 1 of the challenge.

Yesterday was set up. Today we use the AI you installed. About seven minutes. The single most foundational shift in how you brief AI for anything from here on.
Today’s task is maybe something you’re already doing. However, being specific in briefing, and doing it the correct way, is at the heart of building a great, personalised AI system.
​

You can watch along here as I demo the challenge - or feel free to read on below.

A word before we start

This challenge isn’t about getting AI to write things for you. Everything we do over the next 30 days assumes you stay the author. AI is the colleague. The researcher. The editor. The thinking partner who pushes back when you’re being lazy. Never the voice that replaces yours.

The move: brief AI like a colleague (TCCA)

Most of us, when we want feedback on something we’ve written, send it to a colleague with a casual “what do you think?”. They come back with “yeah, looks great.” Polite. Useless. They didn’t want to hurt our feelings. They probably haven’t really read it.

When we ask AI the same way, we get the same answer. “Great post! Consider adding more emotional resonance.” Useless in a different voice.

A brilliant editor wouldn’t do that. They’d quote the weakest line back to you and tell you why it’s the weakest. They’d say “paragraph three is hedging” and they’d name the hedge.

The cleanest mental model for AI: treat it like a new hire on Day One. They’re capable, but they don’t know your business, your audience, what you tried last quarter, or what NOT to do. They need context. They need constraints. And the good ones ask questions before they start.
​

TCCA is the structure that gets AI to do that.
Four pillars, plus one line that does most of the heavy lifting.

This framework was popularised by Sabrina Ramonov, who’s been one of the sharpest voices teaching AI on Instagram and YouTube. I’m sharing it because it’s the cleanest four-line brief I’ve come across, and the “Ask” line is the magic in it.

• T. Task What you want done. Specifically.

• C. Context What’s true about the situation that matters.

• C. Constraints Length. Tone. Format. What NOT to do.

• A. Ask “Before you say anything, what do you think the weakest part is? Tell me first, then we’ll discuss.”

The “A” is the line most people skip. It’s the most important one. It flips AI from bluffing praise to honest diagnosis. From producing to thinking with you.

​

Step by step

You’re going to use TCCA on a real draft you’ve written.
Something that’s been sitting unfixed because you can’t quite say why it isn’t working.

Step 1.
​Find one piece of work that we can improve. Here are a few examples of what works:

- A LinkedIn post you drafted but never published
- A paragraph from a board update or strategy doc
- An opening for a keynote you’re shaping
- A reply to a tricky email you’ve been avoiding sending

Real beats hypothetical. Pick the one that’s been bothering you longest.

Step 2.
​Open claude.ai in your browser (for this you can use Chat GPT too).
Open two browser tabs, both showing claude.ai.
We’ll feed both tabs the same draft and ask for feedback two different ways.

Step 3. ​
Make sure Claude Opus is selected in both tabs (model selector at the top of the chat window).
​

Step 4. ​
Paste this into Tab 1: (the way most people ask for feedback)

Here’s a draft I’m working on. What do you think?

[Paste your draft below]

Hit Enter.
Don’t read what comes back yet. Switch to Tab 2.
​

Step 5. ​
In Tab 2, we will brief AI using TCCA
Paste this, with the bracketed bits filled in for your draft:

I’ve got a draft of [a LinkedIn post / board paper / email / keynote opening] and I know it’s not landing yet, but I can’t put my finger on why. I want your honest read, not your polite one.

Task: Look at this

draft and tell me what’s NOT working. Don’t rewrite it. Tell me what’s weak, what’s missing, where it loses energy, what to cut.

Context: This is for [your audience - be specific: not “marketers”, but “CMOs at mid-size B2B companies”]. The point I’m trying to make is [the single sentence underneath the whole thing]. The feeling I want them to walk away with is [defensive / called out / agreeing / curious]. The thing I’m worried about is [what you suspect isn’t landing].

Constraints: Don’t offer me a new draft. Don’t be polite. Don’t say “great post.” If you don’t think the opener is doing its job, say so and quote the specific line. Identify what to cut, what to sharpen, what’s missing.

Ask: Before anything else, tell me what the single weakest part is, quote it back to me. Then list the other 4 or 5 things that aren’t working, also quoting specific lines. Don’t wait for me to ask.

[Paste your draft below]

Hit Enter.
​

Step 6. ​
Claude will quote a specific line back to you and tell you why it’s the weakest. Sit with that for a second. Then read the rest of what it says.

Step 7.
​Push back where you disagree. “Why is that line worse than this one?” “What about the ending - does it work?” That’s the conversation. You stay the author. AI is your honest reader.

Step 8. ​
Switch back to Tab 1. Read what AI gave you cold.
​

What you should notice

Tab 1: AI was polite. It told you the post was “thought-provoking.” It suggested “adding more concrete examples.” The feedback was vague enough that you could agree without changing a thing. This is what a colleague who didn’t want to hurt your feelings would say.

Tab 2: AI was honest. It quoted a specific line. It named the weakness. It told you what to cut. It didn’t write you a new draft. It made you a sharper reader of your own work.

That’s the difference between AI as a polite typing tool and AI as an honest editor.
The shift takes one line: “tell me what the weakest part is, quote it back to me.”

​

Same framework, three jobs

The critique exercise you just did is one use of TCCA.
The same brief structure can work for other jobs you do all the time too, and keeps you as the author.
Here’s two examples..

​

Job 2: Find an angle worth writing from.

You’ve got a topic but no point of view yet. Use TCCA the same way, with one shift in the Task line.

Task: I want to write about [topic]. Help me find some interesting angles worth writing from.

Context: [your perspective on it, your audience, what’s already been said about this in the obvious places].

Constraints: Don’t write the piece. Suggest 5 angles I could take. Each one should be sharp enough that only I could write it, because of [my specific background, access, or perspective].

Ask: Before suggesting angles, ask me what I’m actually trying to say underneath the topic.

Use this when you know you want to write about something but haven’t found the angle yet.

​

Job 3: Get a sharper opening for something you’re about to write.

You haven’t drafted yet. You’re stuck on how to start. TCCA the same way, Task line shifted again.

Task: Help me find the sharpest opening for [piece]. Not a full draft. The angle I open from.

Context: [who it’s for, what they already know, what they’re sceptical about, what they actually need to walk away believing].

Constraints: Don’t write the whole thing. One paragraph at a time. No marketing language.

Ask: Before you suggest anything, ask me three to five clarifying questions to make sure the opening you suggest is the right one. Then offer me two or three angles I could open with, and let me pick.

Use this when the blank page is the problem.

​

One framework. Multiple uses. TCCA.

The Task line shifts. The Ask line stays. AI moves between honest editor, thinking partner, and brainstorming colleague. You stay the author across all three.

Try TCCA on something else today.
Notice how often AI lands on a specific, named weakness you’d been working around.

The most important thing about briefing AI properly: the skill outlasts any one tool. ​
TCCA works in Claude, ChatGPT, Gemini, whatever ships next. You’re not learning a Claude trick. You’re learning how to brief AI properly. The brief is yours forever, regardless of which tool you end up using.

​

Where this is heading

Today was the easy step. Crucial, but easy. Briefing AI properly is the foundation everything else stands on, and getting specific is what makes any of this work.

We’ll keep coming back to that principle. Especially when we write Skill files together - reusable instructions Claude remembers forever, personalised to you, your work, the way you sound. That’s when this stops being prompts and starts becoming a system. One day at a time. We’re building toward it.

​
Tomorrow on Switched On

Day 2: Tell Claude who you are - the easy step that quietly powers the rest of the challenge.

Five to ten minutes. No install. No new tools. One short paragraph in Claude’s Settings, your work, your audience, how you want answers, what you never want Claude to say.

Save it. Open a new chat. Notice Claude already knows.

From tomorrow, every TCCA brief you write lands sharper because Claude knows you.

We come back to a deeper version of this later. But Day 2 is the foundation, and it takes ten minutes.

Got a friend or a colleague who this might be useful for? Send her to nishma.co/30daychallenge

#SwitchedOn #SwitchedOnAIChallenge

​

Welcome.

Thank you for signing up to the AI challenge. I’m so glad you are here and I’m looking forward to going on this journey with you!

This challenge is to get you from AI anxious to AI confident in thirty days. Ten to twenty minutes most days. By Day 30 you’ll have built your own AI agent that runs on a schedule while you sleep. Built in your voice. About your work. Useful to you. Saving you time and stretching your brain!

Today is setup. This email is longer than the ones you’d usually receive as I need to explain a few steps to get you set up and ready. ​
Two tools to install, about twenty minutes in total. Then we start properly tomorrow.​
​

Why Claude (and not ChatGPT or Gemini)

Quick note before we go further.

Claude, ChatGPT, and Gemini are the three major AI tools you’ll hear about this year. All three can help you. None of them is objectively “the best” for every task.

I’m building this challenge around Claude for three reasons:

1. It writes more like a person out of the box. Less of the “moreover, in conclusion, it is important to note” slop that the other tools default to. For women whose work depends on sounding like themselves, this matters.

2. The Cowork feature is unique to it. From Day 7 onwards we install the Claude desktop app and use a feature called Cowork that reads your files, edits your files, and takes actions on your computer. ChatGPT and Gemini can’t yet do this in quite the same way. It’s what makes the agent build in Week 4 actually possible.

3. It’s what I use every day. Across the 30 days I’ll be sharing the exact prompts, screenshots, and shortcuts that work for me. Following along with the same tool means you can copy and paste without translation.

Can you use ChatGPT or Gemini instead? Yes. Some of the prompting techniques translate directly. However, when we get into some of the sexier stuff we will use Claude Cowork, which is on the paid subscription.
Also, worth noting that crucially I will show you how to set up a system that is fully portable - so you can move to a new AI if you wanted to, and the race is still wide open and changing fast. That’s why good foundations really matter.

My recommendation: just use Claude for the 30 days. If you love it, stay. If you don’t, switch back to your tool of choice after Day 30 and you’ll have the same skills, sharper. The systems and setup can be ported to any tool - as to be honest the AI race remains wide open.

​
​
​The three sides of Claude

There are three ways to use Claude. You don’t need to remember this now. You’ll meet each one when we need it. But for orientation:

Chat lives at claude.ai in any web browser. It’s a conversation. You ask, Claude answers, you go back and forth. Each conversation resets when you close the tab unless you save it as a Project. We use Chat across all of Week 1, in the browser, no install required. Best for: quick questions, drafts, anything you’d ask a smart colleague over the desk.

Cowork is the desktop app, which we install on Day 7. Big difference: Cowork can read your files, edit your files, and take actions on your computer. You hand over a goal (“read this 60-page board pack and tell me what matters”) and it does the work. This is where the agent gets built, the one that lands on your screen at 7am while you’re still in bed. Best for: multi-step tasks, work that involves your real inbox or calendar, anything that would otherwise mean copying things between tabs.

Code is the more powerful sibling of Cowork. It was originally built for software engineers but it’s now used by anyone running serious AI workflows. We touch this lightly in Week 4 when we schedule the agent to run on its own. Don’t be put off by the name. You won’t write code. You’ll write instructions in plain English.

​

A word on free vs. paid

The free tier caps how many messages you can send each day, and it doesn’t give you reliable access to the smartest model (Opus). So you’ll burn through your daily allowance fast on the side-by-side comparison exercises, and the comparisons themselves won’t be as sharp as they could be.

When we move to the desktop app and start building the agent, free really won’t cut it. Cowork and the scheduled tasks need Pro.

My recommendation. If your budget says no right now, start free and see how you feel by Day 4 or 5. If you want the meaningful version of this challenge, upgrade to Pro. I think it will pay you back ten fold in time and more success. It is the cost of getting build an agent personalised to you.

I’m not pushing you into it. I just want to be honest about what you’ll get for free vs. what you’ll get for paid. Pro is what builds the agent. That’s the whole point.

​

Install Wispr Flow too

There’s one more tool worth installing today. Wispr Flow.

It’s a voice-to-text tool. You hold a key, talk into your laptop, and whatever you say appears as text wherever your cursor is. Any app. Any text field. Including Claude.

Why this matters across the 30 days. We’re going to ask you to brief Claude properly. Give it the real context. Tell it who you are. Tell it about your audience. Tell it about your business. Doing all that by typing is slow. Doing it by talking is fast, more natural, and almost always sharper. You say things differently than you write them.

By Day 15, when Claude interviews you to build your About Me brief, you’ll be answering in full sentences for ten minutes straight. You do NOT want to type that. You want to talk.

Wispr Flow’s pricing: there’s a free tier with a generous word limit that gets you through the whole challenge comfortably. There’s a paid tier that unlocks more, but you don’t need it for this. Start free.

To download: search “Wispr Flow” and go to their official site. Mac and Windows both supported. Works on your mobile too. Install, grant the microphone and accessibility permissions, set the hotkey. Three minutes.

​

A note on the other tools we’ll use

A handful of other tools come up across the 30 days. None of them are required. Most have a free tier or a seven-day trial. My standing recommendation is the same for all of them:

Try the free tier or trial first See if the tool earns its place in your work Only invest in the ones that actually help There are real opportunities you unlock when you pay for a subscription. That’s true of most things in life. But I’m not asking you to pay for anything you don’t need.

Be mindful of your budget. Skip what doesn’t fit. The challenge works without any of the optional tools.

​

Your only job today (two tools, twenty minutes)

1. Get Claude ​
Go to claude.ai/upgrade in your browser. Sign up or log in (use a personal email if you can. Easier later when we connect calendar and Gmail). Pick Pro at £18/mo if you can. The Free tier works as a starting point if you’d rather. Confirm payment. Open claude.ai in your browser. You should see a chat window.

2. Get Wispr Flow ​
Search “Wispr Flow” and go to their official site. Download the app for Mac or Windows. Install, then grant the microphone and accessibility permissions when prompted. Both are needed. Set the hotkey. The default is usually Fn or right-Option. Pick whatever feels natural to press while you’re talking. Test it: open any text field, hold the hotkey, say “hello, this is working.” Let go. The text should appear. That’s it.

​

If anything goes weird at any step in either install, take a screenshot, paste it into a new ChatGPT tab (yes, ChatGPT is fine for this since you don’t have Claude working yet), and ask: “What am I looking at and what do I do next?” We’ll teach this same trick properly on Day 6.

​

How to share and chat as you go

Two things that make this challenge more fun.

Tag #SwitchedOn when you post.

Whether you’re celebrating something AI helped you build, asking for help, or showing off the output that landed in your inbox, use the hashtag on LinkedIn, Instagram, anywhere. I’ll see it. The cohort will see it.
The sharpest examples get featured in the round-up newsletter at the end of each week.

You can also use #SwitchedOnAIChallenge on platforms where that’s more searchable. Both work. Use whichever feels right.

​

Know someone who might benefit from going on this AI warm-up to buld confidence? Send her to nishma.co/30daychallenge

​

SwitchedOn #SwitchedOnAIChallenge