The big topic from AWS Cloud Day Dublin was pretty clear: AI. AWS is making big moves in making it easy and seamless to add LLM functionality backed by an à la carte menu of models to your app. It was so dominant in terms of the topics that it felt like a AI conference at times (mixed in with the usual sales pitches 😃).

Here are my hot takes 🔥 on both the hype and substance of AI.

The marginal effort to implement LLM-based AI features has dropped to a very small level

Unlike classic machine learning problems, which potentially involves training a model for your particular business inputs and outputs, an out of the box LLM based solution is almost comically easy to implement.

The technical implementation to implement AWS bedrock is literally a API call, and since the input and output payload is just plain text, its a pretty easy one at that. Qualio’s head of product Kevin Duggan gave a great talk at cloud day on Qualio’s adventures in AI & bedrock, and how it can be done with nothing but traditional programming skills.

How is this going to impact the industry

When the barriers to adding LLM based features drop to almost nothing, it makes for a very competitive environment for ‘AI’ companies. It means that incumbents in different software sectors need not fear being ‘left behind’, as the pivot to add functionality for any reasonably modern software company is minimal. When you combine this fact with the fact that lots of VC capital has been pouring into anything with .ai in the web address means we are likely to see a lot of bankruptcies in this sphere when the VCs get bored of the hype.

On an individual level, the idea bandied about that ‘prompt engineering’ will be a key skill in the future is massively overblown, as a competent person can learn those skills in an afternoon, without even knowing how to code.

Where the opportunity lies

Don’t get me wrong, I still think LLMs are an amazing tool, that will change things in a wide variety of industries, but I think the key differentiators for software companies to succeed will be the following

• Knowing what LLMs are good at, and what they are not good at
• Using LLMs in a way that solves a core problem for customers, not adding them as a novelty feature
• Having a development organisation that can execute well and at a good velocity

AI Agents

A similar thing applies to AI agents, where the LLM ‘part’ is largely solved, but that the key differentiator will be who can wrap the best ‘rules engine’ or guardrails around these agents to prevent unwanted outcomes. This is still an area of frenzied investment, but the winners will not be the ones with the best LLM, but the best traditional software around it