Build AI That Works

Build AI That WorksFull Show Noteshttps://www.theintelligenceagepodcast.com/840Mark Smth talks with James Diekman about how Accelerate Tech has scaled, where AI is actually creating value in government workflows, and why most successful projects are targeted workflow changes rather than broad rollout tools. They also dig into Australia’s growing caution around AI governance, data residency, and the practical push toward local models and sovereign infrastructure.We discuss how AI is being used inside engineering and marketing, why many Copilot-style deployments lose momentum, and how project management itself may evolve as agents take on governance, reporting, and coordination tasks. James shares what’s working in production, what keeps failing before launch, and where the next 6 to 12 months of AI adoption is heading.Key topicsJames Diekman shares that Accelerate Tech has grown to about 32 staff in the last six months, driven by demand in government projects and AI-enabled solutions.The conversation contrasts AI use in engineering and marketing versus more passive consumption in functions like HR and finance.James explains that the strongest results come from embedding AI into specific workflows, rather than treating it as a standalone chatbot or a broad deployment.They discuss why many Copilot rollouts see mixed adoption and usage drop-off when the tool sits outside the daily workflow.James describes the most successful approach as identifying a business process, pulling apart a sub-workflow, and then applying AI, automation, or judgment-based reasoning to that narrow area first.He notes that many AI projects never reach production, and says that out of 30-plus AI projects delivered, only five or six have made it fully into production.The discussion turns to local government systems, with James outlining how his team often acts as the integration layer, or “the plumbers,” between older council platforms and newer software.They cover Australia’s increasing AI governance maturity, including state frameworks, federal requirements, and the rise of dedicated AI roles inside agencies.Mark and James debate public concerns around data centers, water use, and energy, with James emphasizing the need for better policy and the practical constraints of local compute.They explore the move toward local models, onshore hosting, and reserving compute capacity as organizations seek more control, lower risk, and better throughput.James makes the case that teams do not always need frontier models like Opus for every task and that model choice should match the job, cost, and risk.The episode closes on project management, where James outlines a “project brain” concept using agents, shared knowledge, registers, ticketing, and workflow automation to support or partially replace manual PM effort.If you want to get in touch with me, you can message me here on Linkedin.Thanks for listening 🚀 - Mark Smith