Covate convenes Meta Operations Conference 2026 at the National Press Club

CANBERRA, 27 May 2026 — Covate convened the inaugural Meta Operations Conference (Metopcon) on Friday 22 May 2026 at the National Press Club of Australia, bringing together twenty-seven practitioners from federal government, defence, cyber, software engineering, digital agencies, financial services and media to examine how AI agents are reshaping the way organisations build, operate and go to market.

Held under Chatham House Rules, the day was deliberately cross-disciplinary. Participants ranged from people who have been building large language models since 2018 to a retired engineer who had been coding with AI agents for just three months. The facilitator set three goals for the day: to explore what the group called Agent Driven Hyper Development, to test which business models will persist as AI capability compounds, and to identify what this network of practitioners can do together.

The morning session worked through the architecture of modern LLM-based systems — from foundation models and inference engines through to the Model Context Protocol (MCP), command-line agents, and full agent frameworks such as Hermes and OpenClaw. Contributors walked the room through how agentic systems are now assembled from interchangeable parts: model providers at the base, orchestration and memory in the middle, and conversational channels such as Slack, Teams, Discord and WhatsApp at the edge.

Michael Gately presented agentaus.ai, an Australian sovereign LLM released in May 2026 that keeps data in-country for defence, NDIS and other data-sensitive customers. Gately observed that Australia has adopted Claude faster than any other country as a proportion of population, and argued that the next twenty-four months will see a shift from "apps with AI bolted on" to AI-native processes where applications are the output, not the centrepiece.

Two case studies from digital agency JourneyHorizon anchored the afternoon. The first, ChimeDeck, is an open-source Trello replacement built in three weeks by one engineer using a disciplined "Wagile" pipeline: a detailed business requirements document, 172 sprint definitions generated by Claude Opus 4.7, code execution by OpenAI Codex 5.3 through Microsoft Copilot for IP indemnity, and full regression testing via Playwright and MCP after every sprint. The equivalent traditional build was estimated at one year for a team of ten. The second case study described an eight-person AI marketing team running on Hermes, which has compressed an SEO content cycle from 180 hours per month to three.

A recurring theme across presentations was the three-model stack now emerging as a de facto pattern: a large reasoning model for planning, a code-trained model for execution, and a separate model — often self-hosted and open-weight — for review. "AI-generated code works beautifully, but it isn't maintained unless you impose the same quality discipline that existed pre-AI," one contributor noted. "Don't be lazy with the excitement. Hold the discipline of code quality that has existed since the punch-card days."

Coax founder presented an omni-channel inbox for small business that vectorises every customer interaction — voice, SMS, WhatsApp, Instagram, web chat — so the AI responds with full context regardless of channel. The team's honest reflection on the commercial frontier resonated with the room: token economics are still unsettled, and managing the transition from subsidised pricing to sustainable margins is the live challenge for every AI-native business.

The discussion of federal government opportunities was particularly direct. Participants observed that ICT operating models inside Commonwealth agencies have not materially changed in thirty years, that the major consulting firms continue to bid twelve-month engagements for work that smaller AI-enabled teams can deliver in six weeks, and that two decades of neglected data management has created an unusually large opening for firms willing to do the unglamorous work of cleaning, governing and productising agency data.

The closing open discussion ranged across the economics of compute, the energy constraints on AI infrastructure, the strategic divergence between US and Chinese model development, and the alignment question that sits beneath any honest conversation about artificial general intelligence. The room landed firmly in neither the utopian nor the dystopian camp. "Both scenarios are live," one participant summarised. "Navigating the transition is the work."

"What made Metopcon valuable was the mix in the room," said Tony Winmill, Director at Covate. "We deliberately put people who have been building LLMs for eight years alongside people who picked up Claude Code three months ago, and alongside leaders who are accountable for delivery inside agencies and enterprises. That is the conversation Australia needs to have right now, and it isn't happening at the big conferences."

Covate intends to run Metopcon as a recurring forum. A summary transcript of the day has been prepared for attendees; recordings will be destroyed in line with the conference's Chatham House commitment.

About Covate — Covate is an Australian advisory and delivery practice working with government and enterprise clients on strategy, enterprise architecture, AI-led transformation and major program delivery. The firm has supported Commonwealth agencies and large enterprises across more than two decades of practice. More at covate.com.au.