Like many engineers, I grew up on science fiction, and the story it told about artificial intelligence was usually a bleak one. Skynet becomes self-aware, the T-800 arrives, and humanity spends the rest of the franchise trying to survive the thing it built. For a long time, that was the frame many of us brought to AI, whether we admitted it or not.
Six months on, I find myself firmly in the Star Trek camp. On the Enterprise, the ship's computer is part of everything the crew does and makes them better at their work, while the captain remains in command throughout. That is the future I now expect for AI in the enterprise, and the closing panel at PlatformCon Sydney, which ClearRoute was proud to co-host, reinforced it. Over 400 people came together, and the conversation was dominated by one practical question: how do we put agents into production safely? Guardrails, identity and security came up in almost every session. That is what will keep us safe from the T-800.
When we founded ClearRoute in 2022, we made a significant bet on platform engineering. It now matters more than ever, because AI amplifies whatever it is given. Take the PHP vulnerability a few weeks ago. An organisation with a mature platform could raise the fix, test it in an ephemeral environment and release it with little manual intervention. An organisation that releases every three weeks, or every six months, loses time it cannot afford. Enterprises in that position need to resolve how they deliver software before they think about agentic workloads.
In Australia, CPS 230 brings this into sharp focus. Any team running agents should be able to show what each agent has done, stop it in a way that fails over safely, and produce the evidence APRA will ask for. Identity and access management must be built into the platform, with each agent given the scope it needs and no more.
Platform teams will increasingly serve business users as well as engineers. Many people already have applications they have built with AI sitting on their laptops. If we do not give them a safe route to production, we will face something harder to manage than shadow IT, which is shadow AI.
New Relic's latest research found that 22% of financial services firms are running AI agents in production with no monitoring at all. If you have an agent running anywhere, map every system and data source it touches, then check it against those three tests. The Star Trek future is available to us, but we have to engineer it.
Like many engineers, I grew up on science fiction, and the story it told about artificial intelligence was usually a bleak one. Skynet becomes self-aware, the T-800 arrives, and humanity spends the rest of the franchise trying to survive the thing it built. For a long time, that was the frame many of us brought to AI, whether we admitted it or not.
Six months on, I find myself firmly in the Star Trek camp. On the Enterprise, the ship's computer is part of everything the crew does and makes them better at their work, while the captain remains in command throughout. That is the future I now expect for AI in the enterprise, and the closing panel at PlatformCon Sydney, which ClearRoute was proud to co-host, reinforced it. Over 400 people came together, and the conversation was dominated by one practical question: how do we put agents into production safely? Guardrails, identity and security came up in almost every session. That is what will keep us safe from the T-800.
When we founded ClearRoute in 2022, we made a significant bet on platform engineering. It now matters more than ever, because AI amplifies whatever it is given. Take the PHP vulnerability a few weeks ago. An organisation with a mature platform could raise the fix, test it in an ephemeral environment and release it with little manual intervention. An organisation that releases every three weeks, or every six months, loses time it cannot afford. Enterprises in that position need to resolve how they deliver software before they think about agentic workloads.
In Australia, CPS 230 brings this into sharp focus. Any team running agents should be able to show what each agent has done, stop it in a way that fails over safely, and produce the evidence APRA will ask for. Identity and access management must be built into the platform, with each agent given the scope it needs and no more.
Platform teams will increasingly serve business users as well as engineers. Many people already have applications they have built with AI sitting on their laptops. If we do not give them a safe route to production, we will face something harder to manage than shadow IT, which is shadow AI.
New Relic's latest research found that 22% of financial services firms are running AI agents in production with no monitoring at all. If you have an agent running anywhere, map every system and data source it touches, then check it against those three tests. The Star Trek future is available to us, but we have to engineer it.