In 2023, after using ChatGPT for the first time, I decided to step away from the ghostwriting agency I had built.
I could see very clearly how AI had the potential to change the way we work. It made me want to shift my focus and explore something else. To me, that was being realistic about where things were heading.
Fast forward to the end of 2026, and a lot has changed.
When I first started using AI, my experience was mostly prompting it and seeing what it returned. Even then, it often struggled to do what I wanted. Today, working with ChatGPT and Claude feels very different. It’s a time to stretch my imagination about what I can do.
Back then, I thought mainly about writing. That was my skill and my strength, so naturally, I wondered whether AI could help me write faster.
Today, I can build agent workflows and put much more of my product and growth marketing experience into practice. I’m getting to do things I feel I never quite had the chance to do before.
I started out writing press releases in a marketing agency. Later, I moved into product, operations and growth, including at Alibaba. It’s fun to pull from skills throughout my career and make them relevant to what I’m doing today. I get to play around with them and hone them further.
It feels like these different parts of my career are coming together.
Over time, I’ve experimented with internal workflows and content systems. Those systems helped me hand over work and step away from the agency. I also built detailed systems with Claude to improve and refine writing.
At one point, I thought that might be as far as I would go.
But in recent months, I’ve been working on two Telegram bots.
The first was a job searcher I helped a friend build while she was looking for work. We customised it around the exact roles and types of companies she was interested in, with relevant news as well.
More recently, I started building a resume writer that people could use directly in Telegram to tailor their resume to a job description.
But things move so quickly. Before I could even launch it, there was already another announcement from ChatGPT.
Thankfully, these projects are experiments. I’m building them because I enjoy it and want to explore what’s possible. Still, they give me a sense of how quickly things can change.
I think 2027 could be the year of personal agents, with more of this kind of work finding its way into the workplace.
I’ve been trying Instinct, Muse and Dots. In my experience, they still feel early. They tend to work on one task at a time, and sometimes they’re slow compared with using a normal chat window. Instinct, in particular, has become so slow that I find it unusable. I’m interested to see how it develops from here.
There’s plenty to improve, but I’m looking forward to seeing more people build agents for their own work. I also wonder whether talking to a colleague’s agent will become more common.
That raises questions about what we know in our own areas of expertise, what we want an agent to help with, and how we build and maintain it so it works well within an organisation.
For me, the next challenge is to get more utility out of agents at work. I want to explore how they can be more proactive, handle more scope and take on more of the work I could delegate.
I don’t feel anywhere close to the ceiling. It feels as though I might be using only 10 or 20% of what I could do with them.
But some of the challenge is personal.
I work across many different projects and threads. I want agents to help me switch between them and reduce the decision fatigue that comes with that.
At the same time, I’m a micromanager. I find it hard to let go. The agents haven’t necessarily performed poorly; I just want to keep a close eye on what they’re doing.
Scheduling is one example. Sometimes I have preferences about what kind of work I want to do, depending on my mood. I’m a subjective person. A schedule can make sense, and I can still feel like doing something else.
So I’m thinking about how to take baby steps towards trusting agents more. How do I give them room to work and make some mistakes, so I can see what needs fixing and gain confidence over time?
If I always keep such a tight leash, can they ever earn my trust?
That feels like a psychological challenge as much as a technical one. It’s something I want to work on in 2027.
There’s still so much potential to explore, and I’m looking forward to it.