🤖 "If you don't learn to how to orchestrate agents now, you'll spend 2027 catching up to people who started today.”
- Sundar Pichai, CEO of Google

Hello Reader,

Xero recently published a report comparing how New Zealanders are using AI. They surveyed more than 1000 professionals, and they found that most people are still using AI like a random toolbox. They open ChatGPT, or Copilot, or Claude, or Gemini. They ask a question, get an answer, copy the useful bits, and move on.

That’s fine. A toolbox is useful, but a toolbox does not do the work for you. It's just available at your workbench, and it can add complexity and chaos to an already messy process. The worst thing is, it requires your presence to get anything done.

Some people are using AI as an assistant. They give it a task, it drafts the thing, summarises the thing, critiques the thing, or helps to think through the thing.

That’s better, but it still waits for you.

An AI agent is different.

An agent can run on its own computer, use its own memory, follow instructions, check its own tasks, write files, use tools, manage workflows, send reports, and keep working after you close the chat window.

That’s where things get interesting (and messy).

Because now we are not just talking about prompting anymore. We are talking about managing digital workers.

The AI Maturity Ladder

When I work with teams, I often see them move through five stages:

1. Toolbox — AI is a random collection of apps.
2. Assistant — AI helps when prompted.
3. Agent — AI completes workflows on your behalf.
4. Agent team — multiple agents coordinate around different roles.
5. AI-native — the business is designed around agentic workflows.

Most people in New Zealand right now are somewhere between Toolbox and Assistant. I do a lot of my work here, teaching AI training workshops to help growing professionals learn to use AI more proficiently.

But I also spend a lot of time at a really interesting frontier, between the AI Agent and the AI Agent Team.

One of the biggest decisions to make at this transition is: what agent harness do you use?

You could use Claude Cowork, but that's like taking the bus. Somebody else decides where you go, and you hitch a ride on their route. But with an open-source agent, like Hermes or OpenClaw, you get to drive.

Last week, I taught a webinar comparing two of the most popular open-source agent harnesses. Here's my assessment:

OpenClaw Is Like a Ferrari

OpenClaw is fast, flashy, powerful, and fun.

It also breaks down every two blocks.

That’s not an insult. Ferraris are great. But you do not buy a Ferrari because you want boring dependability. You buy it because you want performance, speed, and the thrill of handling something slightly dangerous.

OpenClaw has a large community, lots of skills, and huge momentum. If you are comfortable with terminal commands, GitHub repos, pull requests, cron jobs, and troubleshooting things at 11:47pm, you may love it.

I have two OpenClaw agents, and they are powerful. But they are also high-maintenance little lobsters.

When OpenClaw works, it feels like having your own personal Jarvis. When it fails, you become the unpaid sysadmin for a digital crustacean. That’s the tradeoff.

Hermes Is Like a Toyota Corolla

Using Hermes is different. Hermes is less flashy, but more dependable.

It feels calmer. Milder. Less volatile.

Where OpenClaw is exciting for technical tinkerers, Hermes feels more suitable for business implementation. Especially now that it has a desktop app, it's easy to manage the dashboards, kanban boards, and memory, and I find it has better orchestration with sub-agents.

A Hermes agent is not just installed - it grows. It learns from workflows, develops skills, remembers useful context, and can become more specific to your business over time.

That makes it more interesting for the client work that I do - installing custom AI agents into businesses, to take over real workflows. I think Hermes is much better than OpenClaw for most small-to-medium sized businesses - because if you don't have an IT department that knows how to troubleshoot in Terminal, you don't want to deal with becoming a mechanic for your fast little Ferrari. Most SMBs want something that works, reliably - and that's what Hermes can do.

Because businesses do not need a toy that impresses developers.

They need something reliable enough to become part of operations.

Agent Teams Need Management

Managing agents is like spinning plates. If you ignore them for too long, some of them wobble. Some fall over. To keep them spinning, they need constant attention. They need roles, boundaries, review cycles, memory pruning, skills, workflows, and oversight.

The job of the human operator is not to do all the work. Our job is to remove bottlenecks.

When you are managing your AI agents, ask yourself:

• What keeps needing my approval?
• What can be turned into a skill?
• What should be delegated to a sub-agent?
• What should never be delegated?
• What needs a dashboard, not another chat thread?

That is the work of the AI operator.

Beginner AI Agent Experiment

If you want to understand agents, do not start with a massive business workflow. Start small, and automate a Morning Briefing. If you're taking the bus, you can do this with Claude Cowork, or OpenAI's Codex, or Copilot Agents.

Use this prompt:

Then use this prompt:

Finally, use this one:

This is how you move from a prompt to a workflow, from the AI Toolbox to an AI Agent. As the creator of Claude Code, Boris Cherny, recently said, "I don't prompt Claude anymore. I have loops running that prompt Claude and figure out what to do. My job is to write loops."

My SAGE Framework for Building Agents

When I design an agent, I use the SAGE framework:

Scope - Keep the agent’s job narrow. Wide agents get vague. Narrow agents get useful.

Automate - Find the repeatable pieces. If you do it more than twice, it probably wants a workflow.

Generate - Create prompts, skills, templates, and systems that create more output later.

Evaluate - Review what works, what wastes tokens, what needs human judgment, and what should be improved.

The evaluation loop is where the magic compounds. Because the first version of an agent is rarely excellent. But your tenth version can become so useful, you set it and forget it. That frees you up to go work on your next loop.

The Big Shift

The shift from AI assistant to AI agent is not just a technical upgrade. It is an operational shift.

You are moving from asking AI to answer questions, to asking AI to run workflows.

That means you need new skills:

• Delegation
• Oversight
• Workflow design
• Security judgment
• Dashboard management
• Skill development
• Agent coaching

The people who learn those skills early will have an unfair advantage. Because the future does not belong to people who merely “use AI.” It belongs to people who can manage digital workers.

If you want to accelerate your own learning curve, watch this month's webinar.

📰 New AI News This Week

• Google announced the Agent Resource Discovery Specification, to standardise AI Agent verification and discoverability
• Coinbase launched Coinbase for Agents, a connector that allows ChatGPT and Claude to make financial transactions on your behalf
• Microsoft released Copilot Cowork to all users

👓 What I’m Reading

• ​Privacy, trust, time: Main barriers preventing Kiwi SME AI adoption by Xero
• ​Build Your AI-Powered Morning Brief by Dave Kline
• ​The Hermes Sensei Loop by 0xJeff

Free Community Event

Upcoming Webinar for members of the AI Coaching Academy

Friday, July 17 at 1:00 PM GMT+12 AI and the Exponential Age of Abundance What if we're underestimating what could go right? As intelligence becomes cheaper, faster, and more accessible, we're entering an era unlike anything in human history. Just as electricity transformed industry, and the internet transformed communication, AI is transforming intelligence itself. In this webinar, we'll explore the optimistic case... RSVP

Keep on growing!

- Caelan Huntress ​[email protected]​ ​https://ai-coaching.academy/​

​​​​​