Working with AI
5 ways I actually use Claude at work
Practical habits from using Claude as a small-business owner - built around the three limitations that actually cause problems, not the ones most people assume.
I run Hypernova Transform on my own. No team, no department to lean on - so the tools I use have to genuinely earn their place. Claude is the one I use daily, and most of what makes it useful isn't the tool itself. It's three specific limitations I hit early on, and one habit that answers all three: I keep a single, living Context Card - my business positioning, pricing, standing rules, open questions, even a record of my own mistakes - and it's the first thing loaded into any serious piece of work. It's now on its 33rd revision.
If you're using AI for real work, not just quick questions, these are worth knowing before you hit them yourself.
AI-assisted, human-led: I briefed this, reviewed every word, and stand behind it as if I'd typed it myself. Full detail ↓
Why this matters
The three problems underneath all of this
It doesn't remember your last conversation
Every new chat starts from zero. Nothing you explained yesterday carries forward.
It can't see your other chats
A decision made in one thread doesn't exist in a different one, even five minutes later.
Long threads quietly lose detail
Once a conversation runs long, older parts get compressed down to summaries. Fine detail can drop out along the way - and when it does, the output can sound perfectly confident while being subtly wrong. That's not the AI lying to you. It's a side effect of how long conversations get handled, and it's worth knowing about before you trust a late-thread answer at face value.
One document, kept honestly up to date, is the answer to all three. Here's how.
Give it a proper brief, not just a prompt
A new chat knows nothing about you until you tell it - that's the memory problem, directly. I treat every new thread the way I'd treat a new hire's first day, not a search bar to type a question into. The Context Card is the brief: my positioning, my pricing, the decisions I've already made and why. Loading it at the start of a thread means I'm not re-explaining my own business every time I open a new chat.
Keep one living document - I call mine my Context Card - with positioning, rules, decisions already made. Load it at the start of serious work. Mine's on its 33rd revision.
Make it log mistakes, not just quietly fix them
This is the one that matters most, and it's a direct answer to the third problem - long threads losing fidelity and producing confident-sounding errors. Early on, I asked Claude to check whether a name I wanted to use for my flagship service was genuinely available. It ran a few searches, found nothing, and told me the term was clear white space. I checked myself anyway - and found an established UK consultancy already using it as their core offering, with real client case studies attached. Had I trusted that first answer and posted publicly, it could have caused real reputational damage before the business had even started.
I had the failure written up and dated in the Context Card in full, not smoothed over - what happened, why it happened, what I checked myself. That record doesn't stop a future answer from being wrong on its own - what it does is give me something to check against next time a similar question comes up, instead of taking a fresh confident answer at face value.
Every near-miss gets written into the Context Card, dated, with what was actually checked. Next time a similar question comes up, that record - not a fresh confident guess - is the starting point. The real fix sits alongside it as a standing instruction: state confidence levels and what was actually checked, not a clean answer dressed up as certainty.
Set your standing rules once, not every time
House style, pricing, terminology, tone - I lock it into the Context Card so it's applied automatically, not retyped into every conversation. It sounds small. It isn't. As one example: I have a standing rule to never use em dashes in anything client-facing. It's been missed and caught before - because a rule stated once, in conversation, can get lost the same way any other detail can.
Mine live in the Context Card instead of just being said out loud. That's what actually makes them stick.
Keep an open-items list at the top
Anything unresolved gets flagged at the very top of the Context Card, before real work starts, so nothing quietly drops between sessions - the same compression problem, solved by making sure the important loose ends never rely on being remembered in the first place. Right now that list includes an unresolved website hosting issue, an outstanding policy question, and a couple of open definitions I haven't confirmed yet.
Anything unresolved gets flagged at the top of the Context Card, before real work starts. Nothing gets lost between sessions.
You sign off everything before it goes out
Not a fix for a Claude limitation - the rule that makes the other four trustworthy. If a tool can lose fidelity over a long conversation, or sound certain about something it hasn't actually verified, then checking the output isn't optional extra care. It's the whole point.
The AI drafts. It does the donkey work. The judgement call - and the final word - stay entirely yours. Evidence proportionate to the claim, always.
How AI was used
This page was AI-assisted - I briefed it, it wrote.
I lead AI with my wisdom, not the other way around. Nothing reaches you unread: I review, verify and take responsibility for every word, the same as if I'd typed it myself. Every judgement call here is mine - AI can draft and test an idea, but it never decides what goes in front of you. Read the AI Policy →
That last point is where this connects to the work I do now. A Web Experience Transformation diagnostic runs on exactly this discipline - cross-validated evidence, confidence stated honestly, nothing claimed the data can't support. The habits above aren't just about using AI well; they're the same standard applied to how I work, full stop.
If any of this sounds like where your own business is, I'd be glad to talk it through - no pitch, no obligation.
Get in touch
Let's talk about what's stalling.
Whether it's a full diagnostic or a fifteen-minute conversation to see if there's a fit - reach out directly.
Hypernova Transform
5 Ways I Actually Use Claude
Practical habits for using AI for real work, built around the three limitations that actually cause problems.
