Lees dit artikel in het Nederlands op siskin.amsterdam: “Een systeem dat werkt voor jou”.
This is part two of a three-part series. In Part 1: ‘Stop chatting, start doing’ I wrote about the limits of the chat paradigm.
The prettiest app nobody opens
Two months ago, I built a complete CRM system in a weekend. With an interface I was genuinely proud of: client cards, search functions, a timeline of contact moments. The kind of application you would normally hire a team for or pay for with an expensive subscription. And it was not the only app I built: in the same period I built an invoicing system, a knowledge base for storing and analysing articles, and a portal for analysing international trade data. All with Claude Code, all in days instead of months.
I was pretty proud. Rightfully so, I think. This genuinely felt like a new superpower: the ability to build professional software as a non-developer, tailored exactly to my own work process.
Two weeks later I found myself typing: “give me everyone I know in financial services, with name and phone number.” And then I realised I had not even opened the screen I had so carefully designed. That is when it hit me that it is not about building new tools, but about a redesign of what we call work.
Three phases, looking back
The first phase of my collaboration with AI was about building. AI as executor. I thought about what I needed, AI made it. Quite impressive, but still the old paradigm. I was the architect, AI was the contractor.
The second phase was about directing. AI as an assistant sitting next to me. “Get the emails about this project and list the action items.” “Schedule these appointments.” Individual assignments, ad hoc, but more frequent and more natural over time.
The third phase is where I am now: AI integrated into the systems that organise my workday. Not “do this for me” but “think along with me.” The difference between an assistant that does what you say, and a partner that has context.
The rest of my story is about that third phase. Not as a blueprint; what works for me does not have to work for you. But as an illustration of a principle I see more clearly every day: the most powerful interface is not a screen, it is language.
Why language beats buttons
The CRM tool I built has a fantastic search function. I can filter clients by industry, by location, by when I last spoke to them. Nice. But try this with buttons:
“Give me a list of everyone I know in financial services, with name, phone and email. Create a follow-up action for each of them. Schedule them on the Wednesdays and Fridays of the coming three weeks, maximum five per day.”
That is CRM, task management and calendar in one sentence. No user interface can do that, no matter how well designed. You would have to open three apps, copy and paste manually, and work out the planning yourself. Or you say it in one sentence, and it happens.
Or take something closer to actual content work. Say I write an article and reference a number of sources: reports, studies, news articles. After a few pieces I type: “Take the sources I referenced last week, find recent, widely-read articles that reference the same sources, see what new angles are in there, and save a summary as a basis for a follow-up article.”
That is not just logistics anymore. That is thinking along on substance: analysing sources, making connections, suggesting a direction. And it connects five systems: my articles, my knowledge base, the web, the analysis, and my notes. No form in the world can do that. But a sentence can.
The apps I built have not become redundant. They have shifted: from interface to infrastructure. The AI talks to them. And the nice thing is: that infrastructure (MCP) is not tied to one AI tool. The connections are built on an open protocol. If I want to use a different model tomorrow, I can. I build on my own systems, not on those of a vendor.
The importance of systems
Knowledge work demands more and more from our brains. Every day we have to process an enormous amount of information, and we are constantly context switching: from one role to another, from one project to the next. That is also true in my situation, especially when different projects run in parallel for different clients. Then you need a solid system: a process to plan your day, to keep track of your priorities and deadlines. Only when that runs smoothly can you spend your time effectively.
AI amplifies your systems. If you have no systems, it amplifies your chaos.
When I integrated AI into my work process, I discovered that your system becomes even more important when you work with AI. Because AI thrives on systems. If your tasks are in three different places, your notes are in your head and your email is a mess, then AI cannot fix that. It amplifies what you have. Or what you don’t have.
The most important question is therefore not “what can I let AI take over” but much more “how do I create an environment, a system in which human intelligence and artificial intelligence reinforce each other.” My biggest gains were not in speeding up individual tasks, but in connecting the systems I already had. Calendar, email, tasks, notes, CRM, invoicing. All separate islands that now talk to each other through language.
A day in practice
My workday starts with a daily review: AI pulls information from all my systems and turns it into a briefing for my day. Not a summary of separate parts, but an integrated story with a clear thread.
My calendar says I have a meeting at ten with a potential client. My email says that same person asked a question yesterday that I still need to get back to. My task list says I was supposed to prepare a proposal that is not there yet. Those three things are in three systems. The AI connects them in milliseconds. Not because the AI is so clever, but because it has access.
Throughout the day I use the app Drafts as an inbox for everything that comes in: ideas, notes from conversations, links. Deliberately simple: a text field and a button. No categories, no folders, no decisions at the moment something comes in. At the end of the day the AI processes everything in that inbox: what is a task, what is a note, what is an idea for later.
I don’t process email in my email client but in conversation with the AI. Claude reads, I decide, Claude executes: draft a reply, create a task, schedule an appointment, or archive. The email itself is not the work. The decision about what should happen with it is the work. The rest is handling.
And then WhatsApp. Not the first thing you think of with productivity. But if you have kids who play sports, WhatsApp is where half your planning comes in. Match times, meeting points, cancellations, spread across group chats with sometimes dozens of messages per day. The AI reads along in the relevant groups and processes what is needed into the family calendar. I no longer have to scroll through twenty messages to find out where our youngest is expected at quarter to six tomorrow.
Memory as foundation
In part one I wrote about the memory problem: the black box you have no control over. In my setup, memory is a collection of files on my own computer. Readable, searchable, editable.
The AI knows who my clients are, which projects are running, what the status is of open proposals. It knows when my kids have practice and which school they go to. It knows that I don’t schedule meetings on Tuesdays because that is my thinking day.
All that context I don’t have to provide every time. It is just there, like with a colleague you have been working with for a couple of years. And if it remembers something that is not right, I can open it and change it.
The trap
Of course there are traps. It is tempting to keep automating more, connecting more, letting AI do more. There is always a next integration, a next system, a next workflow.
But the goal is not optimisation. The goal is space. Space to think, to be creative, to do the things that only you can do. If you spend more time building your systems than doing the work you built them for, you have lost your way.
Tiago Forte uses the image of AI as a cognitive exoskeleton: technology that does not replace you but makes you stronger. That image is exactly right. An exoskeleton does not do the work for you: it lets you lift more.
The goal is not a smarter machine.
The goal is to strengthen your own thinking power.
That is what I will write about in part three.
What is coming
In part three I want to explore what happens when the friction falls away. It is not about the tools or the systems but about the effect on your thinking. Because that is ultimately what this is about. Not what AI does, but what it frees up in you.