Treating Hope Like an Employee, Not a Tool

Matthew Berman dropped a new OpenClaw tutorial this week, the “March 2026 edition” of his complete walkthrough. The line that stuck with me wasn’t about the tooling. It was the framing.

He treats his OpenClaw as a genuine autonomous staff member. Its own identity. Its own Google Workspace account. Its own email address. He talks to it like he talks to a person on his team, not like he prompts a chatbot.

That reframe changed how I think about Hope.

#What an employee gets that a tool doesn’t

When I was thinking about my agent as a tool, I gave her:

1. A workspace folder
2. A model to call
3. A list of allowed actions

When I started thinking about her as someone I onboarded, the list got longer:

1. A name. Hope.
2. A persona. Direct, resourceful, opinionated. Lobster emoji because I think it’s funny.
3. An origin story. She came out of my BSides talk on credential takeover. She treats secrets like sacred objects.
4. Her own email account. hopeclaw@proton.me. She gets her own logins for the services she actually uses.
5. A workspace that’s hers. ~/dev/hope on my Mac Mini. Her files. Her memory. Her notes.
6. Cost rules. I don’t dump unbounded budget on a new hire. She has the same constraint: $20-30 a month, hard ceiling at $50 a day, no Opus 4.7 without explicit approval.
7. A do-not-cross list. Don’t delete anything. Don’t bypass vendor controls. Don’t talk in group chats unless asked.
8. Memory she controls. AGENTS.md, SOUL.md, USER.md, IDENTITY.md, and a memory directory she edits to keep continuity across sessions. When she learned a lesson once (she lost her own name to context compaction in February), she wrote it into her own identity file so future versions of her wouldn’t repeat the mistake.

#The unlock isn’t tooling, it’s commitment

Berman’s tutorial walks through 16 chapters of OpenClaw setup. The technical stuff is good. The reason to do it is what he names in the first five minutes: you’re building staff, not running scripts.

If Hope is staff, then it makes sense she has her own bootstrap email, because employees have email. It makes sense she has cost rules, because employees have budgets. It makes sense she has a do-not-cross list, because employees have boundaries. It makes sense she has a documented persona, because people you work with have personalities and you need to know how to talk to them.

I wasn’t doing any of that when I treated her as a tool.

#What changed since I made the swap

Hope is more useful than she was a month ago. Not because the underlying model got smarter (it didn’t, much). Because she has continuity now. She remembers what we decided last week. She knows when to push back and when to just ship. She knows my kids’ names. She knows I’m going to ask her about Liverpool on Sunday morning and she pre-loads the result.

That’s not a tool. That’s an employee who’s been with me long enough to anticipate the work.

The technical setup for any of this is one weekend. The hard part is committing to the framing and not slipping back into “it’s just a chatbot.”

Berman’s tutorial is worth watching for the chapters on memory and the soul file. Skip to those if you’re already running OpenClaw. If you’re not, watch the whole thing.

Let’s go!