Development of Generalist Commercial AI Agent

Sundae Bar Plc (AIM: SBAR), today announces the launch of the training and evaluation of its Generalist Commercial AI Agent ("Generalist Agent"), marking a significant milestone in the Company's product roadmap. This phase focuses on training and evaluating the Generalist Agent's reasoning, decision-making and workflow capabilities through structured benchmarks, rather than training underlying foundation models.

The Generalist Agent is being developed as a single commercial agent designed to scale across many enterprise workflows without requiring bespoke AI development for each one. This approach is intended to preserve the marketplace outcome of selling fit-for-purpose AI products, while improving iteration speed, scalability, and capital efficiency.

Training and evaluation are being conducted through Subnet 121 ("SN121"), the Company's decentralised training and benchmarking environment built on the Bittensor network, where global developers compete to improve the Generalist Agent against structured, real-world benchmarks. Demonstrated performance improvements are expected to be incorporated into a single, continuously improving production agent deployed through the sundae_bar platform.

Subject to successful training and validation, the strongest-performing version of the Generalist Agent is expected to be made available to customers through sundaebar.ai, where businesses would be able to rent and deploy it as part of their operational workflows. Monetisation is expected to be driven through repeatable subscriptions and usage-based pricing, with commercial usage intended to provide a direct feedback loop into ongoing development. As the Generalist Agent's capabilities enhance over time it is expected to support higher-value use cases and broader enterprise adoption without proportional increases in development cost.

The Company's enterprise platform, including integrated payments, is operational and is intended to support the future commercial deployment of the Generalist Agent through subscriptions and usage-based pricing.

Jill Kenney, CEO of sundae_bar, stated:

"This marks an important step in our strategy to build a single generalist commercial agent that can be deployed across a wide range of enterprise workflows. As customer usage grows, we expect commercial demand to support ongoing development, ensuring that the strongest-performing version of the Generalist Agent is continuously improved and deployed, while maintaining capital efficiency as capabilities scale."

The Company believes that decentralised training through the Bittensor network provides a more capital-efficient approach to developing commercial AI products than traditional closed development models. By rewarding only demonstrated performance improvements against objective evaluation criteria, this model is designed to concentrate development effort on measurable outcomes rather than fixed internal resources, supporting faster iteration and lower marginal development costs over time.

This initiative represents a strategic step in sundae_bar's evolution towards a product-led enterprise automation platform, centred on the development and deployment of a core generalist commercial agent. By aligning decentralised innovation with commercial deployment, the Company aims to build proprietary AI capability while maintaining capital efficiency and platform-wide scalability.