StartUps and AI. What a world!

Hello you lovely, lovely people. I’ve wanted to post a new blog for a while but once again life has been getting in the way. I have been working away on things in the background with side projects kicking away but so much is just grind and not a lot of interest. With these blog posts I always try to make sure I have something to genuinely post about, something that is Start Up centric and I think I have something interesting to share with you now so hopefully you find this of value. 

As many of you may know I run a little AI MeetUp here in Northern Ireland and I have been working with (in more sense than one) AI tools for a while now. I started off as a AI sceptic back when I began this blog, as it seemed a distraction to what I was trying to do. The tools were also a bit, well.........crap. When you’re building something from scratch and everyone is evangelical about new shiny toys that are interesting, but not amazing, it’s a frustrating distraction. It reminded me of people trying to convince others to buy NFT images of monkeys doing something amusing. They were going to make millions weren’t they? (NFT’s are great by the way, just not sure they work in the monkey gif realm). The point is though, I have witnessed enough trends, fads and genuine disrupters in our tech industry to have a healthy sense of cynicism. Cue the Grandpa Simpson shaking his fist at a cloud image.

In Start Up Land I heard some absolute cracking statements.

• “We won’t need Engineers anymore because Loveable is amazing.” 
• “Why would anyone invest in SaaS? SaaS is dead and AI is king!” 
• “Who needs CTOs now Cursor is here. Just tell it what you want.”
• “Stick AI on your deck and it will at least get you through the door.”
• “Nobody is going to look at it unless you use AI.”
• “Why do we need Agile processes when the paradigm has shifted?”
• “Look Amazon has laid off loads of staff as AI now writes their code. Engineers are finished.”

Honestly these were all genuine statements made to me and they were also just……annoying. If you haven't heard about Amazon, AI and AWS going down you should check it out.

This wasn’t just start-up people by the way. Engineers were polarised on what they thought. They either hated AI like it was the Spinning Jenny of old or they embraced it too much and got lost. One thing is true. Casandra Syndrome when any new technology emerges is massive. It happens all the time. It happened with Cloud. It happened with Smart Phones. It happened with DevOps. When new tech is released some people act like the world is going to collapse.

It's not AI's fault.

What I realised was that it was more annoyance with the hype and it wasn’t a problem with AI in itself. So like every curious tech person I started to look for problems where AI could fix things. I developed a few prototypes and saw some potential but nothing that I felt couldn’t be eaten up by Anthropic or OpenAI in a few months. What I could tell though was that the ground was shifting a bit underneath me as I was trying to build out my Foundry Fuel idea. SaaS may not be dead but it was certainly out of vogue and the AI genie was out of the proverbial bottle. As you probably know from previous posts, I shut down what I was doing with Foundry Fuel and realigned my focus. So there I was, no idea and a new landscape of AI awaited.

I built some initial prototypes working with LLMs, RAG, Model training, MCP and Agents. It was cool but again nothing that I would say was solving a massive problem. What I got was to learn the technology, get my hands dirty and know what I was talking about. This is something that is really missing these days in my opinion.

AI tools have come a long way since then and I have continued to explore, experiment and learn what can be done from a seasoned engineering perspective. That’s the key part and a differentiator for me. We all know about Vibe Coding and non-technical people building platforms in five minutes. If that’s your thing then that’s great. Enjoy! Honestly I have no problem with people doing that. Prototyping with AI in this way is completely valid. I’m not saying it’s production ready (e.g. ready for real users) and anyone who tells you it is should probably be handled with caution. Fill your boots though as this really can help you get your idea to some kind of shape and structure really quickly.

What about engineers though? As a seasoned CTO and engineer I am interested in what’s happening back in the real world. How does AI change what we are doing as seasoned engineers and how does that impact existing and greenfield projects? I’m not talking about vibe coding or prototyping small apps.

How does it impact building something big? Something significant. Something enterprise size.

Well in two ways, but both are big disrupters. Firstly there are the AI tools out there you can integrate into your products. The AI APIs, the AWS Bedrocks of the world and the RAG databases like Pinecone. I’ve been using them a lot and they are great. Secondly there are the AI tools that help you build your products. The CoPilots, the Claude Codes, the security tools. These increase velocity and take the load off of developers to craft every single line of code. Both approaches are great, both are powerful, both are dangerous and, if you are an engineer, you should get hands on with them.

AI ate my StartUp

So I mentioned before that I was looking for a problem and this is where things circle back round to my Start Up experience. So many vibe coding projects I have seen get swallowed by an AI provider when they release their own tool. I wanted to build something significant. I have spoken before about Selazar and the story is probably well known by now. When we were wrapping up Selazar, AI was really just emerging. We could have used it in so many ways but it was still at the stage of helping people generate text for presentations and documents. We weren’t at the level we are at now with Claude Code for example. Nowhere near. I was also approached when Selazar collapsed to consider building it again from scratch and as tempting as that was I wasn’t really interested as we had solved a lot of the problems in the 3PL industry and lost the business I loved. Rebuilding the same again didn’t appeal. Especially when you are heartbroken. You don’t want to go back to your ex when they have just ripped out your heart over the last 12 months.

Things have changed.

So where was I. I wanted to apply what I had learned about AI tools in and out of product but I needed something significant. This is where a conversation with a local retailer came to the fore. About six months ago I had a chat with someone and I landed on an idea. An idea to take what I had learned at Selazar and apply AI to it to solve new problems. Now all problems need to have a commercial benefit and I had to check that out. Initial discussions were good and when I found myself unemployed I tested the water for this new idea. The feedback was good but there was an issue. The market was really in flux. AI had taken a lot of investment, AI was seen as a bubble getting ready to burst, AI was losing its lustre and investors were not seeing the results they had expected. So conversations were good but not the warmest. So I started thinking. How could I validate this idea? How could I show it would work. I am working with AI tools anyway but I wanted to build something and show it working. So Fufil was born.

Why is this relevant? Why should you care? Well this is where I think it gets interesting.

AI in the product.

Selazar digitised an industry where technology was not hugely present. We took warehouses from bits of paper to mobile apps. We processed orders in seconds not minutes. We increased accuracy dramatically and processed delivery options across multiple providers to find the cheapest. We could get the analytics of Pick and Pack and we could save customers money by streamlining their whole operation. And we did it at scale. At one point we were processing tens of thousands of orders a day at one warehouse alone. It wasn’t perfect though. There were bottlenecks in the process. There was human interaction and a lot of intervention when things went wrong. AI can solve so many of those problems and there are some crazy benefits to this. I’ve worked with companies who throw AI at things. Some vibe code new features and products and backwards engineer them into their platform. Some build AI features to say they have them but they are useless as nobody needs or wants it. Both of these ideas are dangerous as they don't protect data, they create gaps and leave holes to be patched. They are both false economies. A new product like Fufil has to have AI in a way that works. So I started building but introduced agents, LLMs and more to certain places to streamline the operations, make a difference to the users and their pockets.

AI in the build

A large challenge like this is a platform and it takes time to build. We built the first version, the MVP, of Selazar over a year with a team of three engineers scaling up to six towards the end. It’s not a modular app. It’s a beast of a build. You have so many domains to build and cross check. It has multiple code layers and a level of infrastructure a standard application doesn’t need. This is a new build, not existing code, not existing infrastructure. New. In a different language using different tools.

So how can AI help here? Well it can increase the velocity of the build dramatically. I am one person building a huge platform but I have achieved so much already. Can't anyone do this? Isn't this vibe coding? Well here’s the issue. AI is a tool and the tool is only as good as the craftsperson who wields it. Tools like Claude Code are great, they really are, but switching between code projects, data model instances, UI platforms and getting the tools to know what the wider context is a challenge in itself. Tools know the bounds of the context you put them in and you don't want to go free range with AI agents.

It's just the start

This is just the start of what I have experienced so far and I will continue to document this over the next few posts.

I am also going to speak about this in a lot more detail at the next NI AI Meetup. I will explain the product, what it does, where I am bringing in AI but more importantly what it is like to build an enterprise size platform from scratch using these tools. I will share the challenges, the learning experience and some of the snags I came across and I will tell you how I approached them.

I will post another blog after the MeetUp in May but reach out to me on LinkedIn if you want to know more or want to find out how to come along and see what is happening.