Summary
This article explores the shift toward sovereign AI infrastructure and agentic systems designed to prevent the 'hollowing out' of corporate institutional knowledge by offshore frontier models. It highlights the development of DecidrOS, an Australian-based 'operating system' that routes tasks between humans, private models, and frontier models to ensure data sovereignty. While primarily focused on economic and intellectual property security, the move toward localized inference and private models addresses emerging concerns regarding the concentration of AI power and the loss of human oversight in complex decision-making workflows. The partnership with SCX.ai further emphasizes the growing importance of domestic compute for maintaining jurisdictional control over sensitive AI deployments.
Body
Decidr is building AI infrastructure designed to capture and protect company knowledge and make sure it compounds inside the customer’s own AI systemsDecidrOS routes work to humans, private models, frontier models or automation, helping companies control AI spend while building an intelligence asset they ownSugarwork, Rumi and SCX.ai acquisitions support a strategy built around knowledge capture and sovereign infrastructure Special Report: Decidr Ai Industries (ASX:DAI) is targeting what it sees as the next major enterprise AI challenge.The ASX-listed agentic AI company is building AI infrastructure designed to protect company knowledge as an owned asset, as businesses confront the rising cost and strategic risk of relying on offshore frontier models for everyday work.A viral essay by Microsoft CEO Satya Nadella titled A Frontier Without an Ecosystem is Not Stable recently argued businesses risked being “hollowed out” by foundation models, similar to globalisation: not of factories or jobs alone, but of the institutional knowledge that gives whole industries their edge.Telstra chief executive Vicki Brady has made a similar point, describing the next frontier of AI as the ability to capture a company’s human “secret sauce”: the judgement, interaction and decision-making that often sits outside formal systems.“Businesses are spending heavily on tokens, but they’re not necessarily building an asset they own,” Decidr executive chair David Brudenell said.“The work isn’t captured, the learning is lost and the most valuable part, the tacit knowledge, workflows and decision logic that make a company different, can end up being exploited by the frontier models they’re paying to use.”Cost is forcing the issue into the openThe corporate AI race has largely been about access: which model to use, which vendor to pick and how quickly staff can start experimenting.Many businesses have spent the first phase of generative AI buying access to powerful models for employees to use unconstrained.But as usage grows, and AI moves more deeply into operations, the economics become harder to ignore.Decidr’s approach is that not every business task needs to be sent to an expensive frontier model. Some work can run on private models trained on a company’s own knowledge. Some should remain with people. Some can be handled by deterministic automation.Rather than treating AI as a general-purpose assistant, DecidrOS is designed to map how work happens, break workflows into executable tasks and route each task to the right execution layer: human judgement, private model, frontier model or automation.Frontier models have a role, particularly for complex reasoning and high-value work, Brudenell said. But many everyday workflows can be handled by private models, humans or deterministic systems at lower cost and with greater control.The cost saving comes from using the right intelligence for the right part of the task. The compounding value comes from capturing what the business learns each time that task is done.The result, Brudenell argues, is not only lower AI spend. It’s better control over where the company’s knowledge goes and whether the learning created by AI usage compounds inside the business or goes somewhere else.Maximising value, without giving away assetsRather than acting as a standalone AI assistant, DecidrOS is designed to become part of a company's operating system by routing tasks to the most appropriate execution layer.A business might use a frontier model for complex reasoning or high-value work, while repeatable internal tasks may be better handled by private models trained on company knowledge, by people or through deterministic rules.Building the learning stackDecidr’s acquisitions and partnerships sit behind that market position.Last year’s acquisition of Sugarwork gave the company a way to map how work is really done inside an organisation, including the informal paths, people and decision points that often sit behind official processes.Brudenell has described Sugarwork as an “archaeological dig” into the real way work happens inside a company.The output is a structured view of workflows and tasks: who does the work, how often it is done, what it costs, where decisions are made and which parts of the workflow could be executed differently.The recent acquisition of Rumi adds the continuous knowledge layer.Where Sugarwork maps workflows at a point in time, Rumi is designed to capture live workplace signals from meetings, conversations, Slack, Teams and other environments where knowledge is constantly being created, revised and patched.Brudenell describes that as fine-tuning telemetry.“Think about Slack messages, think about your Teams, think about Google Meet, think about all this stuff that’s happening that’s contextual to the work that you’re doing.“You’re now essentially creating a brain for the business.”Together, Sugarwork and Rumi are intended to help companies build a living map of how they work, then use that knowledge as the basis for private AI systems they own and control.Sovereign infrastructure completes the strategyThe third layer is sovereign infrastructure.In May, Decidr signed a strategic commercial agreement with SouthernCrossAI, naming SCX.ai as its first sovereign inference partner in Australia.The agreement is designed to give Decidr customers access to Australian-based compute for deploying their own fine-tuned models, with models and data remaining within Australian jurisdiction.Brudenell said the partnership gave customers a domestic base for their own AI systems.“The agentic economy runs on inference, and inference runs on infrastructure,” he said.“Partnering with SCX.ai gives Decidr customers a domestic, high-performance foundation on which to deploy their own fine-tuned models with confidence.”This article was developed in collaboration with Decidr AI, a Stockhead advertiser at the time of publishing. This article does not constitute financial product advice. You should consider obtaining independent advice before making any financial decisions.More related storiesSponsoredEVG advances US fluorspar huntEvion Group is advancing staged exploration at its Carp Fluorspar Project in Nevada, targeting drill-ready fluorspar zones across and beyond the historic system.Read moreSponsoredLion retains oil field upsideLion Energy has locked in a 45% stake in the East Seram PSC after Indonesian regulators approved the transfer of 15% to its operating partner.Read more