”A bad worker blames their tools. Then the manufacturer. Then the supply chain. Then the industry. Then the economy. And, finally, the stars.” Unknown
Previously at the Volcano Base, I was tazzing about, doing keynotes about AI and enjoying experimental film-making techniques. This week has been dominated by hands-on AI agent creation.
Mission Briefing
Don’t let the tools choose you
I’ve spent the past few days wrestling with Microsoft’s Copilot Studio. I was building an AI agent prototype for a client. It was like trying to build a house with a toothbrush. Copilot Studio has lots of clever features, and is readily available for most businesses with a MS 365 subscription, but using it to build what was actually needed was like DIY dentistry.
What drove this choice of tool? It was available, unlike dentists in the UK.
Software availability bias
Often, a new tool is picked up by our radar, promising to revolutionise everything from our email management to cold water swimming. The tool is available, it’s supposedly cutting edge, and it’s been pre-approved by IT or procurement. Before you know it, you’re three hours deep into the settings and you’ve almost bitten off your own tongue.
An inconvenient convenience
I’ve wasted countless hours trying to bend unsuitable tools to my will, simply because they were mandated. It’s like that CRM system you’re using not because it’s perfect for your business, but because it was included in some package deal three years ago. Now your entire operation dances to its peculiar rhythm.
The AI experimentation tax
With artificial intelligence tools, this problem becomes even more pronounced. They’re evolving constantly, which means today’s toothbrush might indeed become tomorrow’s house-building marvel. But there’s no guarantee. The cost of constant experimentation is now effectively a tax on doing business in the Synthetic Age.
A better approach
Instead of letting available tools dictate your processes, try this:
- Define your actual needs first (duh)
- Set clear criteria for what success looks like (ideally including a financial metric)
- Test tools against these criteria
- Be prepared to walk away, even if you’ve invested time learning the tool
- Crucially, re-evaluate the tools 6 months later
Just because you have access to something doesn’t mean it deserves access to you.
Classified Intel
Some interesting stuff I discovered on my adventures.
OpenAI has launched new tools to help developers create reliable AI agents, including the Responses API and the Agents SDK. These tools simplify the development process by integrating features like web search and file search, making it easier to build complex applications.
Conflicting opinions on Manus, but it’s another tool from China. On the surface it looks like a very capable agent, able to complete complex tasks, especially related to web-based research. Some people suggested it’s just a wrapper for Claude, but either way, it’s worth knowing about.
This project won the ElevenLabs 2025 London Hackathon. In this video, one agent calls another to book hotel rooms for a wedding. The call is received by the hotel’s agent. Both agents realise they’re agents and switch from English to a more efficient tonal “language”, a bit like the old modems used to use. I wrote about this eventuality back in (checks notes) 2015 and features in the keynote I’m doing for Publicis Media today.
Get unlimited 1:1 advisory calls with yours truly
