AI has totally transformed the way I work over the last month or so, but not in a way that I was expecting. In my previous post about “the coming AI agent tsunami” I talked about my experience with Hermes, and how I expected agents to become an intrinsic part of every job. This latest epiphany wasn’t quite like that, though in retrospect, maybe it’s the next logical step. AI isn’t just changing how we do our jobs, it’s changing how we interact with everything.

I'm a terminal nerd, I live in the command line. Since Windows 10 went EOL, I have switched to Linux as my daily driver, and been so much happier for it. When I'm working on a problem, analysing data, doing assessments, or some other thing that isn't just writing the final report, the first tools I reach for are python or bash. This is how I get shit done.

Back in September I started a new gig which required going back to Windows. It's a small shop with outsourced IT, so there's no capacity to support a non-standard operating environment, or even non-standard applications. A Linux VM still needs patching, which would fall outside the MSP's capabilities. Even WSL was seen as too risky.

I spent a day or so wondering how the hell I was going to get anything useful done without grep or sed. I managed to get VS code and git-bash approved for installation, but it was still only half a solution. I realised it was’t just the shell I missed, I needed everything that goes with it: vim; ssh; programming languages and libraries; the whole plethora of CLI tools in any modern Linux distro, and the ability to install more if the one I needed wasn’t available.

While I was trying to negotiate some sort of middle ground with the boss, they pointed out to me that they have a company Claude account and to give Cowork a try. Sure, I thought, it’s just another kind of command line, if you think about it a certain way. I uploaded a dataset I was working on, and started writing a prompt. Ironically, in my head I was translating what the shell commands I wanted to run would do back into natural language to ask Claude to produce the same result.

This is when something I hadn’t expected happened. Instead of spitting out a processed CSV for me to go over and analyse manually, Claude built me a dashboard to visualise and manipulate the data. Without me asking. I realised I hadn’t been very specific about how I wanted the data presented, just what I wanted to get out of it. It inferred what the best way to do that was itself, and just went ahead and did it.

I felt like I’d finally had that light bulb moment that “less technical” people have when they start playing with AI and it performs magic for them. I’ve tried to maintain a healthy skepticism about autonomous AI up til now. I understand enough about how LLMs and agents work to not be amazed by the parlour tricks most people used them for… but here I was, suddenly amazed. Not because of the output that Claude had produced, but because it solved a problem for me before I asked for it to be solved.

Weeks later, Claude has become not just the first, but more or less the only tool I reach for when I need to do almost anything in this environment. Claude, parse this config file. Claude, generate me a structured JSON report. Claude, build me a dashboard. Claude, turn that into an exportable Excel spreadsheet. Claude, go through this DNS log file, research every unique domain and work out what SaaS products they are associate with… then build me a dashboard about usage patterns and anomalies.

I don’t mean to imply that Claude is the greatest, or the only tool capable of doing this. You can get the same results with OpenAI, or a local Hermes agent wired into OpenRouter (though that takes considerably more work on your part). My point is, I’ve stopped doing a lot of the things that were my bread and butter as a technical consultant. Scripting. Organising data. Gap analysis. Reporting. That’s all commoditised now. AI can do it way faster than I can, and present it much prettier too. All I’m bringing to the table… is knowing what questions to ask.

I feel like this is a point which has been made ad nauseum already by people way further down the rabbit hole than me: The job you did last week doesn’t exist any more. Your new job is knowing how to get the same results out of AI without doing all the grunt work yourself. That said, this post isn’t about that.

My big realisation out of all of this is that the Claude Cowork console has, for this gig at least, become almost the only thing I interact with meaningfully. I still read my emails in Outlook, open spreadsheets in Excel, and browse the web with Edge (ew)… but when I need to actually get something done, I’m only working in Claude.

In other words, Claude has become the universal interface by which I get everything else done. Maybe not EVERYTHING just yet, but we’re not far off it, I wager. Most of us have already stopped searching with Google. Now when you need answers to something, you just ask an AI. Why? because it can find, read, parse, and analyse 20 sources in the time it would take you to open another tab to load Google. Word will be next… who needs a word processor when an AI can write the whole 50 page report for you based on your badly organised and barely comprehensible notes?

So my take away to you is this: If you’re not already working in an agentic platform - start now. My finger in the air prediction is that in under 2 years, that’s going to be table stakes for any knowledge work related job.