The year began with a breakthrough in China from Deepseek, which seriously undermined US ambitions to dominate the large language model (LLM) market. What Deepseek showed the world, besides causing a major disruption in the valuation of the financial markets of US AI tech giants, was that China, which the US had tried to undermine by only allowing the export of less powerful AI acceleration hardware, was able to create a model that could outperform US LLMs that benefited from Nvidia's most powerful chips.
Its significance goes beyond geopolitics: Deepseek's R1 model demonstrated that you don't have to throw huge amounts of computing resources and spend huge amounts of money on hardware to accelerate artificial intelligence to achieve good results. Hyperscalers' financial results show the trend is towards large investments in the gigawatt-scale data centers they expect will be needed to support the most powerful hardware to accelerate artificial intelligence.
But for everyone else, including enterprise IT, such infrastructure is certainly overkill, especially since smaller AI models are able to combine the expertise of government LLMs with more focused training to achieve results that can outperform the big players when deployed in a business context.
Agentic AI has become the most hyped technology trend of 2025. Enterprise IT professionals and business leaders are having to deal with the consequences of enterprise technology vendors' mad rush to sell AI-enabled products.
There are many reports showing that AI provides low return on investment (ROI); Most projects fail to make it past the pilot stage, and more and more AI is being implemented into corporate IT systems. This meant that although corporate Artificial Intelligence Strategy perhaps based on the standardization of multiple AI engines, every piece of enterprise software is sold with autonomous AI capabilities.
Given the low ROI achieved by most enterprise AI projects, the industry has shifted to agent-based AI to connect the dots between enterprise AI systems that were attached to commercial enterprise software. The goal is to increase efficiency by allowing disparate AI systems to act as AI specialists configured to handle specific parts of a business process.
The question then becomes what happens to the parts of the workflow that the human worker must perform. It is this interface between workers and AI systems that is receiving a lot of attention right now. If AI sells increase efficiencythen at some point people's jobs will change, and some may find that they do not meet needs. Those who remain employed will have AI agents as employees.
Business leaders are considering how to balance human work with tasks that can be easily completed by AI agents. Instead of just being digital tools, there is discussion about seeing an AI agent as a resource that improves and gains experience over time through learning, i.e. machine learning. There will be social implications as agent AI moves beyond the hype and evolves into something that can do useful work within an organization.
Here are Computer Weekly's 10 best artificial intelligence stories in 2025.
When used correctly, large language models (LLMs) promise revolutionize software development – but they are not easily suited for some enterprise IT use cases, as natural language features pose some challenges. Most programs are written in English-like programming languages, which are deterministic, meaning that the programmer effectively tells the computer what it needs to do. However, using natural language in Vibe coding can lead to problems when trying to describe something unambiguously.
Availability DeepSik-R1 LLM shows that AI can be deployed on modest hardware. Matthew Carrigan, a machine learning engineer at Hugging Face, suggested that a DeepSeek-based AI inference system could be built using two AMD Epyc server processors and 768GB of fast memory. The system, which he demonstrated in a series of tweets, can be assembled for about $6,000.
The Ada Lovelace Institute explores how 'market forces' can be used to boost professionalization artificial intelligence guarantee in the context of a wider political shift towards deregulation. It recommends that AI regulation differentiate between AI systems in general and those used in a more specific context, in terms of both the practical technical and legal competence required to provide each type of system, as well as the standards that should be applied to each.
Companies promoting AI forget to mention that it is often based not on code, but people label data and viewing questionable content – AI could not exist without cheap labor, largely outsourced to countries in the Global South. Then there is the “cloud,” which has a larger carbon footprint than the airline industry and is distinctly physical, as seen in resource-intensive data centers and mining in environmentally challenging locations.
We spoke to Chris Lowke, Group CIO at Hiscox, about implementation of Microsoft Copilot and how to achieve success in AI projects. For Loak, an AI strategy is like the North Star, which broadly envisions an AI-powered business.
“We believe that AI is a generational technology that will underpin many, many things,” he says.
The phrase “don't believe the hype” has never been more appropriate – with more and more warnings about AI investment bubble it could affect everyone if it bursts. For example, Thinking Machines Lab, an artificial intelligence startup, recently raised $2 billion in funding at a $10 billion valuation—the company has zero products, zero customers, and zero revenue. The only thing the company has provided to its investors is the resume of its founder Mira Murati, the former CTO of OpenAI.
We'll explore how organizations can take automation to the next level using agent-based artificial intelligence. Analyst firm Forrester uses the term “process orchestration” to describe the next level of business process automation: Using agent-based AI in workflows allows ambiguity to be handled much more easily than the programming scripts used in RPA.
AI glitch was one of the hot topics at the Gartner Symposium in Barcelona. We spoke with Gartner's Helen Poitevin about the chaos at work in AI. Poitevin said employees will see some of the tasks they do begin to disappear. She recommends that IT and business leaders take a human-centered approach when designing AI systems that people want to use to do their jobs more effectively.
We talk to security experts about how IT departments and security leaders can keep them going. artificial intelligence systems safely and reliable. If you think of an AI model as a new employee who just joined the company, do you give them access to everything? No no. You trust them gradually over time as they demonstrate trust and ability to complete tasks.
Organizations are starting to think about where they can use artificial intelligence. business processes. IT leaders can prepare their organizations for workflows that can be shared among internal staff, external contractors and AI agents by capturing knowledge using structured data ontologies to make expert knowledge machine-readable.
