How AI changed the way I work (and what it didn’t)
Published August 1, 2025. Updated October 3, 2026.
This article describes AI tools as they stood in fall 2026. Tool names, prices, and capabilities change from month to month. The lessons about method hold up. The technical details don’t. Check where things stand before you choose anything.
Last year, I wrote a first look back after a year of using AI every day. I compared tools, I marvelled at image generation, and I talked about AI as a good writing assistant.
A year later, the difference fits in one sentence: AI doesn’t just write for me anymore. I build with it.
I work regularly with Claude, ChatGPT, Gemini, and Perplexity, and a bit with Manus. But this article doesn’t compare tools. It’s about what changed in the way I work, whatever the tool.
Talking instead of writing
I’ll be honest: I hate writing and I hate reading. For 20 years, that was the price of documenting, planning, and organizing anything.
Talking to AI out loud changed that more than anything else. I think out loud, and the AI puts it in order while I talk. I read over the result, fix what needs fixing, and it’s done.
It’s not perfect. It cuts out, it gets a word wrong, sometimes you have to say it again. But for someone like me, it took a huge load off. If writing holds you back, start there.
Being able to trust it (a bit more)
A year ago, every procedure the AI gave me, I checked from top to bottom, line by line. Whatever time I saved on the writing, I lost on the checking.
Today it goes much more smoothly. There are still mistakes, but a good share of them get caught along the way, because the AI now checks its own work before handing it over.
That doesn’t mean you stop checking. It means you get to a level of trust where you can build on top of it. And that’s what makes everything else in this article possible.
No more starting from scratch
For me, the biggest change this year is memory.
Before, every conversation started from nothing. You had to explain the project all over again, the client, the decisions already made. Today you can keep a history in plain text files: what was decided, why, and what’s left to do. The AI reads them at the start of each session and picks up where you left off.
It’s like a brain for the business. Everything that used to live in my head, or scattered across emails, is now written down somewhere and actually used. For a small business where everything rests on one or two people, that’s a whole new level.
Handing off specific tasks
Last year, I asked AI for a piece of writing. This year, I hand it a task.
Some AI tools can now work directly on your computer or inside your apps: sort files, go through a batch of documents, pull information out of one program and drop it into another. No more copy-pasting between the conversation and the work. The AI does the task.
The key is the word “specific.” A clearly defined task, with a result you can check, hands off well. “Take care of my admin” doesn’t.
Building your own tools
That’s where this year surprised me most.
With tools like Claude Code, ChatGPT, or Base44, you can now build complete web apps from nothing. I’ve built my own internal tools: a task manager, a booking calendar, a system to track requests, and a tool that keeps an eye on forums where business owners talk shop. Each one does exactly what I need, without the useless features of off-the-shelf software. It isn’t free: count on around $20 a month, depending on the tool. But it’s one subscription, not one for every program and every user.
For a small business, that means you no longer have to bend the way you work to fit the software. You can build the tool around what you actually need.
But there’s a catch, and it matters: a tool you build, you have to maintain. Document it, back it up, fix it when something changes. If nobody looks after it, you’ve just created a new problem. This is where 20 years of project management make the difference, far more than the technical side.
What you can’t hand off yet
Last year, I was excited about image generation. Less so now.
It works, but not consistently. One day the result is impressive; the next day, with the same request, it isn’t. There are queues, bugs, and days when you’re better off letting it sit and trying again tomorrow. Creative work and design can’t be fully handed off yet. It still takes an eye, judgment, and time.
My advice: never count on image generation for something due under pressure. The speed is real, but you can’t get it on demand.
Starting from an existing design is another story: it goes very fast. A document or presentation built in Canva from a template you like, with AI filling in the content, is done in a few minutes and the result is predictable. Starting from scratch is a gamble. Starting from a base is a sure win.
What hasn’t changed
AI doesn’t replace being on the floor, judgment, or experience. It speeds up the doing. It doesn’t decide what to do.
And last year’s advice still holds: a tool you adopt because everyone’s talking about it costs more than one you adopt because it fixed something specific.
Where to start
Pick a task you redo every week. Try doing it with AI, ideally by talking instead of typing. If it saves you time after a month, keep it.
Once that’s running, take the next step: give the AI a memory. A simple document that sums up your business, your clients, and the way you do things completely changes the quality of the answers.
Want to bring AI into your business?
Book a 30-minute meeting, free and with no strings attached. We’ll look at where AI can really help you day to day, without paying for tools you won’t use.
Note: Everything I describe here, I’ve tested myself. I don’t recommend anything I don’t actually use.
Four questions people ask me about AI
Is AI worth it for a three-person team?
Yes, even more than for a big company: you don’t have anyone to absorb the repetitive tasks. Start with just one thing, the one you redo every week. If it doesn’t save you time within a month, stop.
Do you need to know how to code to use AI?
No, not to get started. Even when you’re building tools, the AI writes the code. What takes some learning is knowing what to ask for, and being able to check that the result does what it’s supposed to do.
Can you trust what AI tells you?
Much more than a year ago, but not blindly. AI still gets things wrong, sometimes very confidently. The more a mistake would cost you, the more you need to check.
How much does AI cost to get started?
Count on around $20 a month, depending on the tool. For simple use, one subscription is enough. If you want to build tools or hand off tasks, the advanced plans cost more. The real question isn’t the price, it’s whether you’ll use it. Pick a tool, try it for a month, and keep it only if it saved you time.