Haunt Digital and AI

I’m pretty overwhelmed by all the AI content out there. And it would be easy to assume that it’s either too complex to integrate or that the useful tools are only available from big tech companies or startups.

But in reality, it's pretty much the opposite. AI can do powerful things and support bespoke solutions with far less effort than most expect.

Two key reasons why:

1. The magic of AI models is easily harnessed and open to use. This is in stark contrast to previous digital tech where gains came only after a ton of effort. Things needed to be built, where AI just needs to piped in, given a few instructions, and deployed carefully.

2. Related to above, AI doesn't require a complicated user interface to work, which was always a really expensive part of any web project. Conversational models means we can interact with AI, and make it do really powerful things, without having to design and build rigid and carefully defined user experiences (click around-y things).

AI agents are great. A lot of energy is going into training and deploying them, unlocking benefits for organisations (including for us at Haunt Digital). But they have issues that limits their usefulness, and these limits are becoming increasingly clear.

Key among these are the limits are:

• Constraints on their ability to perform more sophisticated work. A really obvious (and immediate) example of the use of AI for web development, where the early enthusiasm / belief that AI could usefully code things more or less unsupervised is being re-examined, including by our own devs who are finding the tools very useful, but still needing careful management. This story is being played out across a lot of sectors where fears of mass redundancies are being pushed out as we discover how stupid AI can be.

• Security and privacy risks. This is a well-documented limitation of off-the-shelf-AI models and services which can (or should) place hard limits of their usefulness for organisations dealing with private data.

EchoLeak: The Zero-Click Microsoft Copilot exploit that changed AI security

Hackers exploit prompt injection to tamper with Gemini AI's long-term memory

There are going to be lots of cases where organisations will want or need to go beyond agent limitations and towards their their own AI systems via AI engineering.

Importantly, it is apparent that this will be relatively practical for organisations, especially considering the power of the tools available.

This could look like some of the AI engineering work we do at Haunt Digital, which includes:

API development / orchestration: Wrapping AI models like OpenAI into APIs for use in various applications – basically taking the AI brain and applying it on a custom use-case with an organisation.

AI infrastructure: Standing up and managing AI infrastructure – developing and hosting MCP servers and LangGraph agents, hosting and managing vector databases (which are a key ingredient for many AI applications) and the document pipelines they require. This is a return to “owned” infrastructure (although probably not servers under the desk), which might seem retro but is necessary because of the overall lack of maturity in the AI stack.

Integration with existing systems: Connecting tools to internal systems like CRMs, ERPs and CMSs (which will only mean something if you have one of these). Even if you’re using off-the-shelf copilots or agents there is often a lot of technical “glue” required to achieve meaningful results.

Owned AI services: Combining elements of the above, this involves building in-house components and AI services. For example, internal or public facing chatbots, private tool sets, RAG tools, and other AI-powered or augmented workflows.

A LOT of time and money has been spent in recent years on information architecture, but things are swiftly shifting towards a frictionless user experience, with new (and renewed) importance of tools like back-end content design and search. Generative Engine Optimisation (GEO) is part of this picture.

This change massively on our radar because it affects a lot of our clients, whose sites deliver resource-driven content (vs. more "experiential" sites) such as safety and educational resources, and legislation.

This has been in train for a while, with search already moving toward deeper links, and AI is accelerating this signficantly. Navigation and site structure are becoming less central to good web design, which is moving toward conversational interfaces, which bring new challenges around content quality, relevance and integrity.

In the recent past, many solutions were: a) tried and true, well understood, and just needing to be applied to the right problem; and b) skewed towards visual design, where a strong design background was a crucial skillset.

Increasingly, relevant web solutions are: a) emergent, in that they're novel, presenting new challenges both in terms of understanding their benefits as well as around how they can or should be implemented; and b) more technical or non-visual in nature. A lot of the more impactful work we're seeing in our industry right now are more about information than how things look.

This shift will reward organisations' ability to apply emerging technology to problems effectively.

No prizes for guessing we’re excited about the opportunities AI presents — many of which sit in a “sweet spot” for us.

We’ve always tried to help our clients by combining design and design thinking with real technical depth, making the benefits of modern web technology accessible, and making sure it acutally delivers on its promise.

Our processes – from discovery, solution design through to scoping and implementation –are well suited to the kinds of AI projects we see coming down the pipeline.

Overall, we take a pragmatic approach that we know is valued and will (we hope) help us to work closely with organisations to deliver AI systems and solutions that make a difference.

It's also a time for collaboration, learning from each other and sharing what we've learned. This sounds warm and fuzzy, but is really important if we're all going to meet the challenges and opportunites presented AI.

Our views on AI continue to evolve. There is so much absolute dross about AI being spread around at the moment, attenuated by increasingly toxic and unhelpful social media algorithms. When I see this stuff I'm reminded of Lord Melbourne in that I wish I was as certain about anything as they seem to be about everything.

Also, there are genuine concerns around the implications of AI, for businesses, the environment and society at large. We share some of these concerns but I chose to save them for another article. Although, I'd note we are a lot less doom-y than we were six months or so ago.