Progress forecasts orchestration-led AI shift in 2026

Progress Software has set out a set of forecasts for how organisations in Australia, New Zealand and the wider Asia-Pacific region will use artificial intelligence in 2026, with a shift away from pilots and towards orchestration, governance and scaled deployment.

The company's Chief AI Officer, Ed Keisling, said many businesses now use generative AI, but fewer see a material earnings contribution. He linked the gap to operating models and scaling practices that do not match early experimentation.

Keisling described 2026 as a year when organisations reassess the foundations that sit under their AI deployments. He cited retrieval, governance and audit as areas under pressure from regulation and security threats.

Orchestration focus

A central theme in the predictions centres on "agentic RAG", which Keisling positioned as the next stage of retrieval augmented generation. He framed it as a move beyond classic approaches that rely on trusted data sources. He said agentic RAG adds multi-step reasoning, tool use and secure coordination.

The predictions argued that companies would look for ready-made structures rather than assembling complex systems internally. Keisling said the approach has particular relevance for mid-market firms and small and medium-sized businesses, where engineering resources and budget constraints can limit custom development.

He said platforms that combine retrieval, reasoning and auditability could reduce exposure to risk. He also pointed to a need for organisations to update governance and audit systems in line with new regulation and emerging threats.

Widening gap

Keisling also forecast a widening productivity gap between "innovators", early adopters and laggards. He attributed this to an increased ability among leading organisations to automate and orchestrate more complex work with AI.

He said organisations would need to find methods for internal growth and learning that keep the broader workforce aligned with AI changes. He also pointed to structured learning and platform adoption as factors that could narrow the gap between leading groups and the rest of an organisation.

The predictions place emphasis on workflow redesign and measurement, rather than deployment alone. They also describe "hard-wiring" trust and controls into the technology stack, as organisations move from experimentation into production environments.

Smaller models

While much of the recent AI market has focused on larger models, Keisling predicted a swing in 2026 towards smaller, specialised models. He said organisations would increasingly train or fine-tune compact models on their own data for specific business needs and internal processes.

He argued these models would require less computing power and storage, reducing run costs. He also said organisations could deploy them across a wider range of environments. He linked this to privacy and security considerations. He said sensitive data could stay within an organisation's infrastructure rather than moving to cloud services or third-party providers.

Trust demands

Another prediction focused on governance and accountability as expectations mature and "AI everywhere" enthusiasm fades. Keisling forecast increased demand for transparency, auditability and fairness, particularly in regulated or sensitive industries.

He gave a financial services example that contrasts a "black-box" approach with a platform that provides explanations for decisions, maintains an audit trail, and allows review. The scenario reflects ongoing scrutiny of AI decision-making in areas such as credit, insurance and fraud.

Data integration

Keisling also placed emphasis on integration and the handling of unstructured corporate information. He said organisations that succeed over the next 12 to 18 months would not necessarily run the largest models. He said they would extract value from unstructured data and convert internal documentation, protocols and knowledge into retrieval pipelines.

He described this as a competitive issue, where teams that cannot unify and contextualise internal knowledge risk falling behind. He used a security operations example, where staff search manually for incident-response playbooks, regulatory requirements or architectural documentation. He said a unified retrieval layer could surface information in real time.

"If 2023-2025 were the years of pilots and prototypes, 2026 will be about orchestration, governance, and scale," said Ed Keisling, Chief AI Officer, Progress Software.

Progress said organisations would take a more disciplined approach to building and deploying AI systems in 2026, with a shift from pilots to scalable platforms and increased focus on governance at scale.