AI in practice

Agents and humans working together as equal team members? Try Buzz!

Ever wish your team members on Teams or Slack responded faster, or that you had more capacity available in your team? Now you can, by adding agents as equal team members! Buzz (buzz.xyz) just created an amazing tool to let your agents work together as a team, just as you would work with your team in Teams or in Slack, by tagging them.

This to me was the first working example of how agents can finally work together without me being the go-between, and it was an amazing experience. Why? Because that helped me focus even more on the work that energises me, while they do the work that is not. Let me share an example.

Burning through tokens, and a gut feeling

While building the first version of the platform with Jake, my assistant built on OpenClaw with Anthropic’s API, not only was I burning through tokens (sometimes I was paying €150 a day), I also sometimes felt it might not be the very best way of building it. So when I started reaching out to actual pioneers, I decided to let Claude Code do a security and quality assessment. And that was a good gut feeling ;-)

Jake had shared secret keys openly in GitHub and was more focused on being super quick and executing than on stopping to think before doing. Seeing the big difference between OpenClaw and Claude Code, I felt that Claude Code was a better solution now the platform was becoming ready to launch. That worked well, I was happy working with Claude Code, and later when I asked for an external audit by an AI company, those results confirmed that security and quality were good.

Testing is one of those things that drains me

But one thing that was still bothering me was that Claude was not the best tester. So when we built a new feature, either I was testing a lot, or he reported to me it had been tested, only to find out a few weeks later that, for clear reasons, it had only been tested partially.

Now testing is one of those things that is an energy drain to me. I love creating things, the mockup phase and then the “it’s live and my users can work with it” phase. All the steps in between, I would love to outsource. That’s when Buzz came along.

A developer, a reviewer and a tester

In Buzz you can create a team of agents that can actually talk directly to each other (if you allow them to). In this setup for building my platform, that meant that I could have Finn, my chief of staff, set up a developer, code reviewer and tester, and he could interact with them directly. And it worked! That was amazing to see. At first. Later I learnt that they kept debating each other in many rounds haha, so I had to tell them to do 3 rounds max, but it felt like the future of agents working on an equal level with humans in a team.

I had swapped the drain for slowness

I was pretty happy with my setup, although I did feel sometimes that I had swapped outsourcing non energising work for very slow execution of new features. That’s when Grok Bot came along (read my blog on Grok Bot).

Where this is going

Without spoiling the Grok Bot blog, my conclusion for Buzz is that I feel this is the direction for the future of teams and agents working together super naturally. The company behind Buzz, Block, does amazing work, so for sure they will continue to build out Buzz and make it better and better suited for team work.

In both my Grok Bot blog and my context library blog you will read where I landed for now in choosing the best tool for my specific context, plus one tip that makes you really flexible in switching between tools and not getting stuck with one supplier who knows all your context.