Someone told me recently that they'd tried using AI to summarise a meeting. It pulled out the actions, captured the discussion points, produced a neat summary. But it gave everything equal weight. The meeting had been about one specific topic. Eighty percent of the conversation, the decisions, the tension, the follow-ups, all centred on that one thing. The AI treated it the same as the five minutes spent on office logistics.
Their conclusion: AI is okay, but it doesn't really get it.
My diagnosis was different. The AI didn't fail. It was never set up to succeed. All it had was a transcript. No context about the project. No sense of what mattered. No understanding of why the meeting was called in the first place. It did exactly what you'd expect a brilliant stranger to do if you handed them a recording and said "summarise this." It summarised everything equally, because it had no reason to do otherwise.
This is the single biggest mistake people make with AI. They start every conversation cold.
Think about what happens when you bring a new person onto a project. You don't hand them a document and say "go." You brief them. You explain what the project is about, what's been decided, what's contentious, who the stakeholders are, what good looks like. You give them context. Without it, they're going to spend weeks figuring out things you could have told them in ten minutes.
AI is the same, except the amnesia resets every single conversation. Every time you open a new chat, you're talking to a brilliant colleague who has no memory of yesterday. No memory of last week. No memory of the project you've been working on together for a month. Unless you give it that memory yourself.
Most people don't. They open a chat, type a question, get a generic answer, and wonder why AI feels underwhelming. The answer is sitting right there. You gave it nothing to work with.
Here's what I do instead.
Every project I work on has a set of context documents that live inside my AI environment. Most AI tools now have some kind of project or workspace feature that lets you store files alongside your conversations. Claude has Projects. Other tools have similar setups. The specific platform matters less than the principle: give the AI something to read before you start working together.
At minimum, I keep a project summary. What the project is. What we're trying to achieve. Key decisions that have been made. What's still open. I'll often have formatting guides so the AI knows how I want documents and presentations structured. And I keep a running project log that captures what happened in each major work session.
At the end of a significant session, I ask the AI to update that log. Summarise what was decided, what was built, what changed. When old information starts piling up, I ask it to archive the older entries into a separate file and keep the current log lean. This matters because AI has a finite amount it can hold in its head at once. If you dump everything in without curation, you're back to the same problem. Too much noise, not enough signal.
The format matters too. When the AI creates these documents, it naturally writes them in clean, structured text rather than the kind of formatted documents humans prefer. Think of it as the AI writing notes for itself, in its own shorthand. It's simple, readable, and the AI can work with it efficiently. That's the point. You're creating materials that live in the AI world, not the human world.
Then, when I start a new session, I open with something like "have a read of the project document to get up to speed with where we are." If I'm going to be writing presentations, I'll add "also look at the formatting guide." That's it. Two sentences, and the AI already knows more about my project than most people's AI ever will.
Even if your AI tool doesn't have a built-in project feature, the principle still works. At the end of a session, ask the AI to produce a summary document capturing where things stand. Save it. At the start of the next session, paste it back in. It's manual, but it works. The AI picks up where you left off instead of starting from zero.
The important thing is that the summary is concise and structured. Don't give it a stack of PDFs and hope for the best. Ask it to create a clean, condensed version that captures what matters. The AI is remarkably good at this. It knows what's important if you ask it to distil things down.
The difference this makes is not incremental. It's transformational.
An AI with context doesn't just avoid bad outputs. It produces work that feels like it came from someone who's been on the project for weeks. It understands priorities. It knows what's been tried. It remembers the decisions and the reasoning behind them. Each session builds on the last, and the quality compounds.
An AI without context is a parlour trick. Impressive in isolation, useless for real work. The meeting summary that missed the point wasn't a limitation of the technology. It was a limitation of the setup.
The people who are getting 10x value from AI aren't using smarter prompts. They're giving their AI something to remember.
Your move, human.
Damien Healy is the founder of Qanara, an Australian AI consultancy helping businesses accelerate from strategy to impact. He writes about AI-native workflows, frontier AI capabilities, and practical transformation.