I gave Claude a big job. It hired help.

I have been putting off a piece of research for weeks. Not because it was hard, exactly. Because it was large. There were six things I needed to look into before I could make a decision, and every time I sat down to start, the size of it stopped me before I began.

Last Tuesday I typed the whole thing into Claude. Not just one of the six questions. All of it.

What came back surprised me. Instead of working through the questions one by one, Claude told me it was going to split the job into parts and work on them in parallel. It spun up what it called sub-agents. By the time I had made a cup of tea, it had a structured answer that would have taken me most of an afternoon.

I want to tell you what actually happened, because I think this matters more than most of the AI news you have been half-reading lately.

What is a sub-agent?

You do not need to understand the technical side of this. The useful way to think about it: Claude can now act like a small team.

Previously, when you gave Claude a big job, it did one thing, then the next thing, then the next. Like a very capable person working through a list. Efficient, but sequential.

Now, when a task has parts that do not depend on each other, Claude can work on several at once. One part of it is looking into one thing while another part looks into something else. They report back, it pulls the results together, and you get the whole picture.

For small tasks this does not matter. For large ones, it changes what is worth attempting.

What this means for you in practice

The tasks that become newly reasonable are the ones that felt too big to start.

A few examples of things I have either done or could see doing now:

Compare six mortgage products and tell me which suits our situation and why. Previously: the kind of thing you pay a broker for, or put off indefinitely. Now: worth a try.

Research everything I need to know before a difficult conversation with someone at work. Pull in background on the context, common approaches, things to avoid, and draft me an opening. That is four jobs at once. Ten minutes of waiting instead of two evenings of preparation.

Help me understand this legal letter. Pull out the key obligations, the deadlines, the questions I need to ask a solicitor, and write a short summary I can share with my partner. Not one job. Several.

None of these are tasks where Claude replaces the professional. But they are tasks where Claude gets you from confused to prepared, and prepared is where most of us are stuck.

The honest version

It does not always work cleanly. I have had multi-step tasks where Claude lost track of one of the threads, or where the parallel results did not quite fit together. It got better when I gave it a clearer brief upfront.

The rule that has helped me: be specific about what a good answer looks like before you start. Not "research mortgages for me" but "I want a comparison of fixed vs tracker rates right now, the break-even calculation for a five-year fix, and a list of three questions to ask a broker." That specificity is what lets Claude divide the work sensibly.

The size of task that is worth attempting has gone up

That is the plain version of what has changed.

A year ago, AI was useful for the small stuff. The email, the summary, the first draft. That was already worth having. But the large things, the research projects, the decisions that needed ten things looked up before you could land on an answer, those still felt like yours to do.

They are still yours. You still have to read the results, make the call, act on it. But Claude can now do the legwork at a scale it could not before.

If there is a big job you have been avoiding because it felt too large to start, it is worth a second look.