You're Not Behind on AI

Most people aim too big.

You have a ChatGPT account. Maybe a Claude one too. You've used it for things. It was fine. Sometimes useful. But you don't feel like you're "using AI" the way everyone around you seems to be. You watch someone describe what sounds like an entirely AI-assisted workflow (research, drafts, code, summaries, decisions) and you think: I've tried that, but it's not as good as they sell it to be.

What am I missing? You're putting the wrong size problems in.

Define the context window in literal terms.

Lets talk about the context window. It's the amount of information an AI model can hold and work with at any one time. It's finite. Everything you put in (your question, the background, the documents, the conversation history) takes up space. And when it's full, the model starts losing track of what was said earlier.

This matters because it changes what kind of problem AI is actually good at.

Watch what happens when a chat gets long. You started by asking about a strategy doc, then pivoted to a code question, then back, then sideways into an email draft. Around the third or fourth turn, things start to slip. The model confuses your projects. It forgets a constraint you set in your first message and references something you didn't say. The context window has filled up.

When you put a big, messy, open-ended problem into a chat window ("help me figure out my career," "what should we do about this project," "how do I deal with this difficult stakeholder") the model has to fill enormous gaps. It doesn't know your history, your constraints, the specific personality of that stakeholder, the multitude of decisions that led to this situation. So it gives you something technically reasonable and almost entirely useless. And you think: this isn't the revolution everyone's been selling.

But put a small, specific, well-contained problem in ("I need to write a follow-up email to someone who went quiet after a promising first meeting, here's what we discussed") and suddenly it's sharp. Actually useful. Better than what you'd have written yourself at 9am on a Wednesday.

Here what we changed is the size of the problem you handed it and this is what AI is fantastic at.

Where to actually start

Most people can't just "AI-enable" their work overnight. Your organisation might have handed you a basic chat interface and called it a day. Or IT locked everything down. Or nobody's given the green light. So let's work with what you actually have.

Start with your personal life. This is where you have full control and zero politics.

1. Plan your week by dumping everything in your head into a chat and asking it to help you prioritise.
2. Summarise something long you need to understand but don't have time to read properly.
3. Draft a message you've been putting off because you don't know how to say it.
4. Work through a decision you've been going round in circles on by writing out both sides and asking for what you're not seeing.

None of this requires your organisation's permission. None of it needs a special tool. They're problems small enough, specific enough, contained enough, that you can hand them off today.

From there, personal projects. Side projects, freelance work, learning something new, building something outside of work. Here you have both control and stakes. Real problems you care about solving. This is where AI starts to feel like it has a gear you haven't used yet, because you're bringing context you actually own. You know the full picture, so you can give the model what it needs.

Work comes last. And even then, start with tasks. The transformation framing ("how do I AI-enable my team's planning process") is the same trap as we talked about earlier, but much bigger. Try something concrete instead: "I have to write a project update for stakeholders and I hate writing project updates. Let me see what happens with AI." One task. See what happens. Review it honestly. Move on to the next one.

If your organisation has given you something more powerful (Copilot, proper integrations, access to APIs), you have more surface area to work with. But the approach is the same. Keep the problems narrow.

Lets define a small problem

This is the part that trips people up. "Start small" sounds like advice to use AI for trivial things, which lands wrong because it makes the whole approach feel beneath the work that actually fills your week. Plenty of small problems are high-stakes. The person you've been putting off a hard message to is small in scope and heavy in importance. The decision you're going round in circles on about whether to leave a job is small in shape (you're working through one move, you're not redesigning your career) and weighty in consequence.

The size of a problem is about how much context it needs to be answerable. Importance is a different axis. If everything the model needs to give you something useful can fit in a few paragraphs of your own writing, it's small. If it needs your last three quarters of email, the history of a relationship, the politics of two teams, and an unspoken constraint that lives in your head, it isn't.

A useful test: can you write down everything that matters about this problem in the chat window in under ten minutes? If yes, AI can probably help. If you'd need to spend an hour briefing it before you even got to the question, you'd be better off doing the question yourself the old school way.

Most of the work-shaped problems people try to "AI-enable" are the second kind. They look small from the outside (one meeting, one decision) but they're load-bearing on a year of context. That's why they fail. AI can think well. It just can't do that while you're still briefing it on the situation.

The trap everyone falls into

The feeling of being behind usually comes from comparing your early experiments against someone else's mature workflow. You're seeing the output of months of iteration, someone who has figured out which problems fit the tool, how to write a prompt that gets something useful, what to do with the output. You tried it twice and it didn't wow you.

What you're seeing in those demos is usually the back end of someone's iteration cycle, the part where it finally works. The thirty failed attempts to get a research summary that didn't hallucinate dates didn't make the screen recording. Neither did the wrong prompts or the rewrites where they realised they were asking the wrong question. The clean output you're watching is the survivor of a process you didn't see.

That's just not having done the reps yet.

The reps are small. Annoyingly small. Solve the email you've been procrastinating on. Summarise the report you don't have time to read. Draft the agenda for a meeting that keeps going off track. Do it ten times. You'll start to feel the shape of the tool, what it's good at, where it goes vague, when to push it and when to just do it yourself.

That's it. That's the whole thing. A habit of reaching for it on problems small enough for it to actually help.

The people who feel most behind are usually the ones who started with the biggest ambitions. They wanted AI to change everything. It didn't. So they concluded it was overhyped.

It is overhyped, for big, messy, organisationally complicated problems. It's underused for the small, specific, boring ones. And there are a lot more of those.

Start there.