A few weeks ago, I sat down to a Christmas lunch with some senior technical colleagues. After the jokes and stories, we got onto the subject of AI: “It’s overhyped…”, “My team deals with it…”, “It’s not quite there, more of a parlour trick…”, etc. This was astonishing to me; many of them were dismissive of AI and were more or less just getting on with their day jobs. This sentiment isn’t unique to them; while seeing friends and family over the Christmas break, I noticed a broader scepticism about AI, even a fear that “it could run away and take over.”
This is not how I see it. For the last year, I have endeavoured to “live in the future,” applying AI to nearly every task I could imagine. From simulating customers and building software to writing resumes, generating images, organising my schedule, understanding forms, planning trips, researching problems, guiding and teaching—everything. Living like this, day in and day out, has helped me gain a deep understanding of the nuances of these future tools. I have insights that guide me in ways I hadn’t appreciated before. Here are the things I’ve learned that I hope might help you navigate a future sprinkled with intelligent agents.
A Personal Journey with AI
It won’t be worse than it is now. Regardless of whether you are an AI optimist or pessimist, it is moving faster than we can fully comprehend. There are feedback loops leading us to a Cambrian explosion of intelligence. You will find that standard LLMs like ChatGPT, Claude, and Gemini will be good enough for most tasks we humans do on computers. Intelligence is digital, and like a digital photo, you can duplicate it hundreds of times. Imagine having an AI to help you write your articles, allowing you to create variations, voices, and add more research effortlessly.
Most of us might just dabble with a prompt like “Write me a LinkedIn post on AI taking my job,” only to find the output tepid and tasteless, leading us to conclude that “it’s not quite there yet.” However, if you bend that prompt, combine it with other prompts, and add some checks and balances, you can soon have a LinkedIn post that requires almost no changes, written in your style, using your language, and optimised for SEO, catering to your audience. This is possible now; the only difference is the magic of the words and how they are used.
The Future of Work with AI Agents
We are back in the early days of the Internet, but with AI. Back in the 90s, my father asked me to create a website for his company. I took the company brochures and created HTML tables, stitching together what I thought was a reasonable and professional-looking website—a shop front on the World Wide Web. We find ourselves in a similar situation now; we are building AI tools and agents that do “shop front” types of things. We haven’t fully explored what they are capable of and how others are using them effectively.
I think we are doing it all wrong. When creating a website, for example, you might come up with an idea, do some research, and then start hacking together something cool. You may use Cursor.com, v0.dev, or loveable.ai, perhaps adding images from Midjourney and getting Claude to write your copy, feeling pleased with how you’ve halved your development time. This is the “shop front” approach. AI agents are optimisation tools that can help you do your job. They’re little helpers that give you a productivity boost, enabling you to share how much time you’ve saved or money you’ve made.
But this misses the mark. What would the future really look like if we were there right now? We wouldn’t work in the same way as before; we’d do it differently. Our project might be one massive mono repository, including not only the website we’re building (assuming websites still exist) but also markdown docs and prompts integrated throughout the project. Text is a first-class citizen in the future, and formats like markdown, JSON, and CSVs will likely be the most important we deal with. You will write your daily developer diary, capturing activities and learning, adding opinions and resources.
Redefining Productivity
You’ll create your own local AI agents that will read your diary, knowledge base, and code, functioning as a coding assistant to improve it. Not just the AI in your IDE but an actual agent that makes suggestions on improvements and understands the broader context of your project. This type of agent is more like your own CTO, CMO, COO, etc., helping you build out your project and providing the guide rails and insights you might have missed.
Agents are inexpensive to create; if we have a CTO agent, why not develop a Chief Marketing Officer (CMO) agent too? It can create tweets, blog posts, and LinkedIn articles that typically distract us from our development tasks, often resulting in a rushed and lacklustre attempt to build an audience. Don’t stop there; build agents for productising different aspects of your project, create reports on your progress, and automate tasks like deploying and managing DNS records.
Think of it as one large project that encompasses everything: all your notes and documents, the project you’re creating, the day-to-day automations you need, and agents that help you build code, write copy, conduct analyses, and communicate with the rest of the team. Consider how Pieter Levels creates his products entirely by himself. He doesn’t even use databases initially, preferring to rely on JSON docs. He gets things done quickly, then fixes what’s broken—fast!
Agents as Integral Team Members
This is how I believe we should approach software development as we enter the “year of the Agent.” It should be woven into the very fabric of your project, nestled closely with all the markdown, JSON, and other text-based documents. Perhaps, certainly to start with, it’s best to focus on speed, avoiding databases and other infrastructure tools. These can quickly become a distraction, pulling you away from iterating quickly and, above all, engaging with your customers (see this insightful blog article for inspiration: https://www.tmaker.io/adding-dark-mode-added-10k-revenue).
So why not build this now? Maybe I should stop waffling and go ahead and build it. Well, I have… this is the beginning of something I’m calling Agent Web Factory, and I will be sharing more in the coming months. Until then, follow me on x.com/emson.
Thank you!
Ben