Claude vs Claude

Hey there,

David here.

I wanted to share a cool way we’ve been using Claude to get the best out of it.

And the secret is simple.

We often run two Claude sessions on one task, and the second one only checks the first.

It sounds like extra work, and it is a little. It is also the reason people keep asking how we get this much detail out of Claude.

I recorded a short video on it this week, about five minutes long.

Watch: Why We Run Two Claude Sessions on One Task

For those who prefer to read, here is the short version. It comes down to three things.

011. Context before anything else

Before we ask Claude for anything, we load the session with as much context as we can.

Let's say we are working on a RevOps problem inside a client's CRM. We load in:

• every conversation we had with the client
• the chats and the emails
• the CRM itself through the HubSpot connector, plus Clay

Then we ask Claude to research all of it and give us a summary of what it understood.

It is easy to assume Claude will ask when something is missing. In practice, it takes the easiest and fastest path to an answer. When it cannot find something, it takes a shortcut instead of looking deeper. The more context already sits in the session, the fewer shortcuts it has to take.

We also changed how we treat sessions. We used to open a new one for every task to keep token spend down. Now we keep specific sessions for specific tasks and go back to the same one each time. The context built up there makes Claude much sharper than a fresh session we have to teach from the ground up.

022. Dig deeper

Whenever we get a result, we do not take it for granted. We ask Claude to counter-check it.

This is where the second session comes in. We use it only for quality checks, and we test the first session's output against it. Putting the two against each other is how we uncover issues.

Claude can sound like it already knows everything. An answer can look finished while the task is still unsolved. We only know it is solved once we have dug in and checked the details.

033. Manual checks

The last step is ours. We pick specific points and verify them ourselves.

On the same RevOps problem, that means opening the CRM and checking certain fields by hand. Then we see whether they match what Claude told us.

All of this takes more time. The context has to be prepared, the counter-checks set up, and the manual checks planned. At the end of the day, the result is much better, and for work where the detail matters, that is the part we care about.

I would like to hear how other teams handle this. How do you check what Claude gives you before you use it? Hit reply and tell me.

PS. We build custom go-to-market systems with Claude at the centre. If you want to see what that could look like for your team, book a call with us here.