Artificial Intelligence is still hot in 2025

We’re excited to share Dinghy GmbH ’s perspective on how the latest tech news impacts user experience design and digital product building.

This newsletter offers a new format, and we’d be happy to get your thoughts on it. If you’re interested in what Dinghy has been up to, scroll to the end for the usual company updates.

Sustainability & AI

AI still dominates the discussions in the tech and online communities.

About a week ago DeepSeek disrupted the OpenAI leadership in the market and tanked Nvidia’s stock as well as stocks of the US power market.?

DeepSeek AI R1 apparently was cheap in training and required less resources, which could hint at a future with models that have less of an environmental impact. The Green Web Foundation outlines in its report “Thinking about using AI?” how, up until now, there was little you could do with regards to saving energy when using AI. There is potential in choosing data centers that use renewable energy, actively choosing the best model for a task has an impact, too. But overall it boils down to “avoid using AI if possible.”

One argument from the report is “don’t use AI unless your users ask for it”. That resonates with the perception at Dinghy, that in 2024 many software companies carelessly slapped generic AI interfaces onto their product and sold this as innovation, where in reality it often got in the way of a previously well defined and accustomed user experience.

Read the full report on their website

SEO & AI

As expected and seen throughout 2024, AI will have a huge impact on search engine optimization and how content is being discovered on the web.

We’re seeing it on the Dinghy website but also on some of our client websites that traffic from AI-powered search engine is increasing. What’s particularly interesting is that visitors coming from ChatGPT or Perplexity often have a lower bounce rate and stay on the entry page longer.

Plausible is a lightweight and privacy-friendly (cookie-free) analytics tool, that is our tool of choice to track traffic on client websites as well as ours.

They know a lot about traffic-acquisition and published an elaborate write-up about SEO for AI and what it means for website operators. Writing high-quality, in-depth content still is important and there’s technical ways to optimize for being picked-up by AI search engines.

Read the full blog article here

Agents & Agentic Systems

Recent announcements of Anthropic to enable Claude for computer use and OpenAI’s new (and hefty priced) Operator made AI agents accessible to a lot of people.

AI can now control a computer on your behalf. Systems like these, albeit for more specific tasks, have been around for a while and generally are called agentic AI or agents. These agents make use of large language models (LLM) to analyse complex tasks, build a step by step plan and work through it while taking decisions along the way. There are still many open questions and uncertainties about the potential and risks of these systems.

In the meantime we’ve been looking at the user experience of and interfaces for AI powered systems (we talked about this before in The Evolution of AI in UX Design). And here too, are a lot of learnings to be had. Even without agents per-se AI systems will drastically change how humans interact with computers and their devices in general.

One very exciting project we’ve been working on is using speech recognition to automatically fill out a form. While this may sound dull compared to shiny agents, there are a lot of uncertainties and preconceptions to take into account. For example, what happens to form fields that are already filled in? Can they be overwritten by the AI? Can the content be used to give the AI more specific context? The prototype is still in early development and is currently undergoing a record-breaking rapid usability testing sprint.

Agents can only be as good as the data they have at their disposal. We put an LLM to test by embedding content from the Dinghy Data Warehouse (DDW – Double D, Double U). Not only was this a fun endeavor, it was surprisingly easy as well. Thanks to the Sanity Content Lake, its HTTP API and its powerful query language GROQ we could “chat with our blog” near instantly.?

What this does is:

• Download blog posts from Dinghy’s Data Warehouse
• Instruct an LLM to answer questions from the given context by using a system prompt
• Answer a user prompt using the information from the Dinghy blog as context

This is the simplest form of retrieval augmented generation (RAG). It is not an efficient way to go about this in any way though (and as outlined above this is very important).

To turn this proof of concept into a production ready feature some more sophisticated technologies come into play, like content chunking and embeddings, which help finding relevant information in large bodies of content. This will increase the RAG’s accuracy and speed, thus reducing computing load in practice.

Here’s the question and full answer of the LLM based on Dinghy’s blog:

"How can usability research help save money?"

Usability research can save money in several ways:

• Minimize Product Failures: By identifying flaws early in the development process through user testing, businesses can avoid costly fixes post-launch.
• Enhance Product-Market Fit: Understanding user needs helps tailor products to market demands, reducing rejection risks.
• Reduce Development Costs: Iterating based on user feedback prevents resource waste on features that aren’t necessary, avoiding expensive redesigns.
• Avoid Costly Rebrands: By ensuring alignment with user expectations from the beginning, businesses can prevent disruptions that necessitate expensive rebranding efforts.
• Increase Customer Lifetime Value: A focus on user satisfaction through testing fosters loyalty, leading to repeat business without heavy customer acquisition costs.
• Optimize Marketing Spend: Insights from usability research refine marketing strategies, ensuring funds are directed where they will be most effective.
• Boost Conversion Rates: Optimizing user journeys based on testing results leads to increased sales and higher revenue.
• Decrease Customer Support Costs: Identifying and addressing usability issues reduces the need for extensive customer support.
• Mitigate Legal Risks: Addressing accessibility and usability concerns early minimizes the risk of legal issues.
• Strengthen Investor Confidence: Demonstrating a commitment to user-centric design can make it easier to secure funding.

In essence, investing in usability research leads to informed decisions that enhance user experience, streamline processes, and ultimately drive down costs.?

We were impressed how easy it is to get started and achieve helpful results. We learned a lot during our recent deep-dives, about LLMs in general and also about the daunting nomenclature and concepts behind it. For example, we finally understand what weights are, how embeddings work and how simple and advanced RAG work.

Are you interested to understand which technologies come into play when building AI based tools in your organization?

Please let us know by contacting us.

Data-Ownership, De-Regulation and Chaos in Social Platforms like TikTok and X

Nils’ stance on the current political influence on social and the web:

We’ve seen the signs in 2024 already but with Trump’s inauguration and the rise of the “Tech Broligarchie”, 2025 started with whole bunch of big events:

• TikTok was suddenly banned and effectively offline for an entire weekend in the US, which sent content creators on TikTok into an existencial crisis.
• Marc Zuckerberg, Meta’s CEO, announced that all of Meta’s platforms like Facebook and Threads will stop content checking, which sets those platforms on a path to follow X (formerly Twitter) after the acquisition by Elon Musk — which resulted in a massive increase of hate speech AND user churning off the platforms.
• New and open platforms like Mastodon, Bluesky and Pixelfed (the Fediverse in general, see below) saw a massive surge in sign-ups, in consequence.

So what does all of that mean for business in the EU; what does it mean for all of our businesses?

We at Dinghy believe that it’s paramount that companies own their content and are 100% in charge of their intellectual property. Knowledge and information are worth more than ever before and therefor it’s important to make sure, that none of that knowledge is suddenly taken away because a trusted platform pivots spontaneously, like we’ve now seen, again, with tech giant Meta.

All this has two direct implications:?

• We persist everything we do in open formats or in data warehouses that allow us to extract all our content as structured data.
• All our blog posts are published from the data warehouse to our website as the primary source.
• Then we cross-post to social platforms (LinkedIn, TikTok, YouTube, Instagram but also Fediverse platforms like Mastodon). That way, we own the content and don’t care too much if a social platform seizes to exist.

What is the Fediverse and what’s going on over there?

The Fediverse is an ecosystem of various social platforms that are connected by a protocol called ActivityPub. The big difference compared to the established, big-tech social media platforms is that the servers on which the Fediverse runs are federated (hence the name). Which means everybody (who can administer a server) can contribute to the Fediverse (self-hosting), but more importantly that also means all the information shared on the Fediverse is distributed and not owned by a few big-tech companies.

As users can decide freely where to open their account (on servers run by people they trust or on their own) it is much more difficult to take control of and shift the narrative on the federated social networks than it is on, e.g. what used to be Twitter.

The Fediverse comes with a couple of alternatives to the common social platforms:

Twitter → Mastodon

Instagram → Pixelfed

TikTok → Loops

YouTube → PeerTube

Facebook → Friendica

Reddit → Lemmy

A full list of Fediverse alternatives can be found on jointhefediverse.net

What’s more: many Fediverse platforms are developed open-source (so you can check what happens with your posts in detail, theoretically) and as the Fediverse is largely run on donations there’s no ads.

@Elena Rossini has a write-up on the advantages of the Fediverse over the big-tech’s walled gardens: Your data, your power. In another post she outlines how she learned to self-host Fediverse services.

TheVerge explains the Fediverse in detail.

What's new from Dinghy

Good UX can boost e-commerce revenue, enhance customer satisfaction, and drive long-term growth. Here’s a business case for investing in usability design.

In this conversation, Nils discusses the advantages of a centralized content warehouse for organizations, particularly those with multiple companies under a holding structure. He emphasizes the benefits of content reuse, security, and digital sustainability, as well as the transformational impact on team collaboration and access to information.?

This playlist offers a complete introduction to using Sanity Studio. Whether you're setting up for the first time or exploring its features, these step-by-step videos will help you confidently navigate and make the most of the platform.