Everyone’s using GenAI, few are truly benefiting

Computing research reveals the continuing gulf between hype and reality

In the third part of our coverage of IT trends in 2026, we look at the LLMs used in business and how they are being deployed.

Many professionals are now using AI chatbots and tools to help with tasks ranging from software development to drafting emails, but which models are they using most and what are they using them for?

The most used AI tools

The secret of Microsoft’s success has long been its ability to bundle new tools into existing product lines. So it is with Copilot, which is being unsubtly levered into everything – even Paint and Notepad at one point, although the company seems to have backtracked on that for now. Its tools don’t necessarily have to be the best, just the most convenient and closest to hand.

Copilot follows Azure and Office 365 as the most widely used solution of its type among Computing’s readership. Seventy‑nine percent of the 138 UK IT leaders polled said they were using Microsoft’s personal assistant.

After Copilot came the bot that started the GenAI boom: OpenAI’s ChatGPT (70%). (Arguably, OpenAI’s creation should be in the top spot, since it is in part what powers Microsoft Copilot behind the scenes.) ChatGPT benefited from an early‑mover advantage, but whether it will be able to hold on to its share, given the big guns now trained against it, is debatable.

Google, caught on the hop when ChatGPT arrived in 2022, is now regarded by many observers to have caught up with OpenAI in terms of functionality, with Alphabet’s vast revenues also providing a more sustainable funding model. Gemini, which was mentioned by 45% of respondents, may also be a good fit for those using Google Workspace.

“I’m finding Copilot less than accurate and am trusting Gemini more,” remarked a director of IT in the energy sector.

Another tech-giant offering, Meta’s Llama range, is apparently not faring so well though; Llama was mentioned by just 6% of users.

OpenAI’s closest rival, fellow startup Anthropic, saw its Claude chatbot used by 24% of respondents. Anthropic focuses on the business-to-business marketplace, particularly developers, whereas ChatGPT is positioned more as a mass‑market assistant.

“We use Claude a lot for coding,” confirmed a software director in a tech firm.

Other developers swore by specialist coding platforms like Bedrock (5%), Cursor (7%), Replit (1%) and AlphaCode (1%).

Perplexity, the AI search engine, was used professionally by 16% of respondents, while image generators such as Midjourney (3%) and Stable Diffusion (1%) were less commonly deployed.

Hyperscalers’ cloud-based AI platforms were also in use in 80% of organisations polled, including from IBM (5%), Oracle (6%), Amazon (13%), Google (28%) and Microsoft (68%).

Real-world use cases are still limited

While most respondents were using AI tools in their work, their comments indicated that they weren’t finding them particularly valuable yet.

A business applications manager in property services said: “We are trialing Copilot but so far finding little real benefit other than meeting note taking and summary.”

“Our contracts actually say ‘no use of AI in projects’, so it is only used for internal admin, reports, etcetera,” offered an IT manager in the technology sector.

“Only for rough research, getting a first idea before digging in manually because AI systems have a limited knowledge-base, the visible web only,” said a CTO in a design firm.

“Although they can be helpful, they are inherently dangerous in FE as they are often slightly simplistic,” added an IT manager in further education.

The most promising use cases

Rarely has there been such a wide gulf between the revolutionary visions peddled by the industry and the mundane reality on the ground.

We asked respondents about the most promising use cases for GenAI that they had seen, either in production or as proof-of-concept, and their answers were notable for their ordinariness.

Development

The most frequently mentioned uses were in software development, including generating frameworks, translating code to other languages, “vibe coding” simple ideas and documentation.

“It’s the ability to create things that I wouldn't normally be able to, and in a fraction of the time,” said an IT manager at a legal firm. “Where I don't care about how the code is written and I just care about output, it's a life saver.”

Documentation

The next most common category involved documentation: summarising, reviewing, transcribing, rewriting, augmenting, updating and improving written materials.

“Auto transcribing meetings and calls and providing a summary is invaluable,” said IT portfolio manager in a utilities company. “In one area alone this has saved up to 20 minutes per call on a customer call wrap-up.”

AI is also being used to speed up the tendering process - including discovery and generating bids - for creating training materials, and also to provide legal advice, although as one respondent pointed out, “care is needed as errors do happen.”

A CIO in manufacturing said: “It's OK for documentation, but an LLM is, by definition, flawed; we have to create our own internal SLM [small language model] to derive meaningful value from AI.”

People-facing operations

Customer support, sales automation, HR support and information provision are all places where AI can streamline operations.

A strategist in government saw value in the hyper-personalisation of work, speaking of a proof-of-concept “virtual PA for each individual, assisting with routine tasks and lightening the self-service workload to the extent that a good human PA would.”

Cybersecurity and SOC automation

Cyber defence is an obvious use case for AI, although generally machine learning rather than the generative kind. But GenAI is used to come up with scenarios in red team/blue team operations and for explaining issues in ways that are easier for the non-technical to understand.

So, while there are use cases out there and potential for many more as both the technology and understanding of its capabilities mature, in most organisations the reality still appears to be lagging well behind expectations.

“There really aren't that many use cases,” remarked an enterprise architect in government. “Generating materials more quickly results in a lower quality and accuracy.”

Of course, it’s still early days and it’s perfectly possible that AI is both overhyped and underestimated.

Previous articles in this research on IT trends looked at real-world use cases for GenAI as well as opinions on the stability of AI amid speculation that it could be a bubble. The next one will cover trends in security.