A few weeks ago, our CEO, Esteban Castaño, gave every team leader at TRM 14 days to ship something real with Claude.

Going “full send” on Claude was a new muscle for most of these folks. They could have watched hours of prompt-engineering videos or caught the latest AI tips on LinkedIn — but that kind of passive learning only gets you so far. So instead, we encouraged them to learn by doing: Open a session, pick a problem, use Claude as your thinking partner, and ship the smallest version that works. And that's exactly what they did.

“Am I using this right?”

I run AI enablement at TRM. Once the challenge from Esteban went out, I expected a flood of Slack messages — setup problems, prompt tips, questions about which tool to use. In reality, the most common question I kept getting was:

Am I using this right?

Whether it came up during office hours, one-on-ones, or via Slack, people were more curious about whether they were using AI the right way than about anything else. To get a better picture of AI utilization (and utilization gaps), our AI automation team built dashboards to track usage across the company — which tools were being used, by which teams, on which days. Together we identified hotspots, traced what our power users were doing differently, and started to understand which teams had the highest Claude Code adoption (and which had the lowest).

The dashboards worked. They told us where AI was landing and where it wasn't. But the "Am I using it right?" question kept coming up anyway — interestingly, often from the same people who showed up as power users in the data. So I started watching what the champions actually did, not just what the data logged.

The people whose work had visibly transformed had something in common the dashboards didn't capture: they had fewer things open on their screen.

Why using fewer tools is creating tighter focus

I used to think AI effectiveness was a tool-count problem. More Claude Code, more Claude Cowork, more MCPs, more skills. But watching these champions changed my mind. The shift was actually in how many other tools AI let them stop using.

Cal Newport has a line in Deep Work that I think about a lot: the two skills that matter most in modern knowledge work are learning fast and executing at depth. Both require long, uninterrupted blocks of attention. Both die the second you start tab-hopping between seven SaaS platforms to "just quickly update" something.

For 20 years we built our jobs around tab-hopping, with the SaaS bargain being “one specialized tool per problem.” Project management here, docs there, customer relationship management (CRM) over there, analytics somewhere else. By the end, most knowledge workers were running through nine to 12 platforms a day. Every switch cost attention, and every pull-and-paste between tools taxed deep work.

We learned that our people who were “using AI right" were actually using AI to make most of the tools they once bounced between disappear from their attention completely.

From toolbox to Swiss Army knife

In the age of AI, the toolbox era is over. We don't need a suite of specialized tools anymore; we just need one surface that reaches all of them.

At TRM, that surface has become Claude Code and Claude Cowork — and the 14-day challenge to our team leaders yielded incredible results that prove why. Both reach into everything else through MCPs, skills, and connectors: Linear, Notion, BigQuery, Slack, GitHub, Jira, and more. All of these tools aren’t going away — they’re just becoming places we don’t need to visit directly.

An example of what this looks like for me: Instead of opening a SaaS dashboard to update a project status, I have an agent in the background that handles it while I'm still in Claude Code. The multi-tab research I used to do across BigQuery, Notion, and Linear to scope a project is now handled by an agent that pulls from all three. Every platform becomes a data source behind an agent instead of a UI demanding my attention.

It’s enabled me to operate at TRM Speed — the unique velocity and urgency that drives our operating tempo in service of our mission — without friction, and has reduced the distance between a TRMer having a great idea and being able to ship it.

If you’re using AI to level up the way you work and are ready to build at TRM Speed, check out our open roles. We’re hiring and would love to hear from you.