If AI disappeared tomorrow, how screwed would we be?

I had a slightly morbid thought this week: what if AI just… stopped? Say every AI tool was switched off at the end of the year. No more chat, no more agents, no more "can you just…". What would I actually lose? My first answer was "speed", which doesn't tell you much. So I did the nerdy thing and went through the evidence: about a thousand of my own prompts over one month. They covered client projects, our own marketing, admin, sales prep and a few experiments. I expected most of it to be "write this for me". It wasn't. Most of the value came from how I'd learned to work alongside AI. It made a handful of good habits so cheap that I did them all the time without noticing. The tools could disappear. I'd rather the habits didn't, so I've written them down. These are the ten I'm keeping, whatever happens. ## 1. Lay out the options before committing The prompt I sent most often wasn't a request. It was a reply, something like *"1. yes, 2. hold for now, 3. go ahead"*. That happened about 170 times in a month! That's because AI almost always set out the options before doing anything: here's the plan, here are the two or three decisions you need to make, and here's what I'd suggest. All I had to do was decide, and that's the right job for the person accountable for the outcome. **Keeping it:** before any piece of work that isn't trivial, I'll write a short note with three parts: *proposed*, *options* and *my default*. I'll ask the people I work with for decisions in the same shape too. In my experience, a message that can be answered with "1 yes, 2 no" gets answered much faster. ## 2. Work in small chunks you can ship "Commit, push and deploy" shows up in my history about ninety times. Almost every piece of work ended the same way: a small change, checked, shipped and tidied up. Nothing sat half-finished for a week, and I'll admit there's a small thrill every time something goes live. Small chunks are easier to review, easier to undo and easier to explain. AI didn't invent this. It just made each round so quick that working any other way felt slow. **Keeping it:** one change at a time, finished properly before the next one starts. I'm turning my finishing steps into a checklist so that doing it properly takes no longer than skipping it. ## 3. Capture feedback as it happens Dozens of times a month, I pasted a message from a client or colleague straight into a session. A comment on a draft, a change request, a bug someone had found. Each one became a change, a reply or a decision within minutes. Without AI, that's exactly where requests get lost. Messages pile up in chat threads, and three weeks later someone asks whether you ever did the thing they mentioned. **Keeping it:** a feedback log for each project with five columns: who, what they said, what I decided, status, and where the change landed. Not glamorous, but it means I can answer "did we do that?" with a date and a link. ## 4. Give feedback all at once, not in bits The design feedback that worked best was a single bulleted list covering everything I'd noticed in one pass. The feedback that didn't work arrived in pieces over an hour, each change undoing part of the last. Bundled feedback is kinder to whoever has to act on it, human or not. **Keeping it:** I'll review in one sitting, write a single list and send it once. ## 5. Automate your reporting I asked for some kind of report around forty times: weekly summaries, "what's changed since last week", results after a launch. Over the month, the ones I asked for most stopped being requests. They became scheduled reports that turn up every Monday without anyone asking. Reporting is usually the first thing to slip when things get busy, and it's also a big part of how people judge your work. **Keeping it:** if I've asked for the same report twice, it gets scheduled. The numbers come from the systems, not from memory, and the summary is a template I fill in. ## 6. Take the time to understand how things work A lot of my prompts were *"walk me through this"* or *"is this right?"* with a screenshot attached. They came up whenever I was doing something unfamiliar: a new tool, a process I only do once a year, someone else's system. (Honestly, a year ago I didn't know half of what some of those screens were asking me.) Looking back, the useful part wasn't being told what to click. It was having the reasoning explained before I clicked: why it's done this way, what that setting changes, and what happens if I get it wrong. I understand far more of my own business now because of it. **Keeping it:** for anything I do rarely, I'll write a checklist with a line explaining why each step matters. The second time round, I want to be checking the work, not learning it again. ## 7. Think out loud, and write it down A lot of my prompts started with *"discussion:"* or *"question:"*. Is it worth hiring for this yet? Should we try that idea now or next quarter? What should we even call this thing? AI was useful in those conversations, but half the value came before I pressed enter. To ask a good question, I had to write out the context, the options and what was bothering me. That's just good note-taking, and AI gave me a reason to do it. **Keeping it:** a decision log with four parts: context, options, decision, and why. Then someone to talk it through with, usually another founder. ## 8. Testing still rules My most careful prompts were about anything that would reach real people. *"Do one first and report back before doing the rest."* *"Make it a dry run so I can check it by hand."* *"Pause everything until we know why it failed."* No amount of AI changes the basics. Test with one item first. Make going live a setting, not a code change. Always have a way to stop. **Keeping it:** a test mode, a first batch of one, an off switch and an error alert, built into everything as standard rather than added later. ## 9. Keep the project in order: documents, signposts and FAQs Some of my prompts were simply "where is that?" When's the deadline? Where's the latest report? How did we do that last time? AI could answer because every project had notes it could read. Those same notes work just as well for me, for a client at handover, and for whoever picks the work up next. **Keeping it:** every project gets a "where things are" page, a page of key dates, and a list of frequently asked questions that grows each time someone asks one. ## 10. Use templates for meetings, and diagrams for thinking Before an important call, I'd ask for a short brief: who they are, what we've done together before, what I want from the call, and three questions. Afterwards, the notes became actions. For planning and workshops, I used a diagram rather than a document, such as cards on a board or a simple timeline that everyone could move things around on. I've found that a one-page brief beforehand and written actions afterwards make a meeting far more useful. A diagram usually gets a room to agree sooner than a document does. **Keeping it:** two templates, one for preparing and one for the debrief, and a whiteboard open whenever more than two people need to agree on something. ## The tools go, the habits stay To be clear, I don't think AI is going away any time soon. Nobody is switching it off at the end of the year, and I'll happily keep using it. The thought experiment was never really about a ban. It was a way of separating what AI does for me from what it has taught me. None of these ten habits needs AI. AI made each of them cheap enough to do every time, and doing them every time is what made me faster. If the tools went away tomorrow, I'd barely slow down. Well, maybe at the typing. The lessons are there for anyone who uses AI daily, and not building them into how you work, with or without AI, would be a waste. If you lead AI enablement in your organisation, try this exercise with your team. Don't only ask what AI does for them. Ask what it has taught them to do. If you'd like help with that, [we're happy to talk it through](/contact-us).