AT&T chief data officer Andy Markus forecast fine-tuned small language models (SMLs) will dominate enterprise use in 2026 by complementing large models in agentic workflows, one of several trends expected by the operator for next year.

Markus noted fine-tuned SMLS are key for unlocking value in mature agentic offerings and stated businesses “will understand the importance of their own data in driving AI value”.

“The large language and reasoning models will often handle the master control of an agentic workflow, but the purpose-built SLMs very adequately deliver the required accuracy and efficiency when trained for their dedicated job within the agentic workflow,” he explained.

Markus provided several AI-related predictions for next year in a blog post, all of which are largely due to the combination of AI agents and AI-fuelled coding.

He stated AI-fuelled coding will be the next major methodology for developers, enabling them to cut development phases from weeks to minutes.

“With AI-fuelled coding, developers will start to wear multiple hats in the lifecycle, from product owners to architects, reducing cycle times and time to operation,” Markus stated.

It also helps non-technical teams to develop apps by using plain language prompts to build software prototypes.

Internally, AT&T has used AI coding to curate data products withing 20 minutes while also training it to “adhere to our rigorous code discipline for quality, security and compliance”.

On-demand apps
Also on the 2026 horizon, businesses will start to build on-demand apps supported by AI agents.

“AI-fuelled coding dramatically accelerates software development cycles, making it feasible for a company to build on-demand apps,” Markus said. “Autonomous agents can even independently adapt to new requirements, making redevelopment faster than traditional app cycles.”

He stated while traditional apps will remain, they will be supplemented by agile, AI-driven offerings.

Mobile operators will also offer more AI services, such as fine-tuning, to enterprise customers. Markus noted operators have a long history of working with cloud providers and AI software companies, “giving us a strong foundation in AI services like fine-tuning”.

His last prediction is metrics for AI accuracy, cost and speed will become the focus across every business sector.

“It’s not enough to use AI tools. They have to use them well, with measurable AI results. Accuracy is what drives value, and optimisation is what’s needed for generative and agentic AI to measure up across any use case,” he said.