Nova Is in Production. Here's What That Actually Means.

<p>Nova is now in production orchestration. That's the milestone I've been working toward for a while, and it's worth being precise about what it means — because "production" for a personal AI system is a different thing than shipping a SaaS product to paying customers.</p><p>It means Nova is no longer a prototype I run occasionally to test ideas. It's running as the backend AI service for my own work — across projects, across conversations, across days. When I need something drafted, researched, summarised, or acted on, Nova is the first system I reach for. That's the real threshold: not deployment, but dependency.</p><p>Right now Nova has 34 agents. They span general chat, pipeline execution, ERP specialists for AIREP, language tasks, and more. The number isn't the point — I'm not listing 34 agents as a flex. What matters is that the system has enough specialisation that routing a task to the right agent actually produces a meaningfully better result than throwing it at a general-purpose model. That's when multi-agent architecture earns its complexity cost.</p><p>The orchestration layer is what I've been building toward. Individual agents are relatively straightforward to write. The hard part is the coordination: how does the system decide which agent handles a request, how do agents hand off context between each other, and how do you avoid the classic failure mode where a pipeline silently degrades because one step in the chain returned something plausible but wrong? These are engineering problems, not AI problems. They require the same discipline as any distributed system — clear interfaces, explicit state, and failure modes you've actually thought about.</p><p>One thing I've learnt building this: the value of a personal AI system compounds slowly and then quickly. For the first several months, Nova was useful but not indispensable. I could have done most of what it helped with faster by just doing it myself. The shift happened when Nova started having enough context about my projects — AIREP, Find a Sign, Sweeper Parts, client work — that it could operate with less explanation from me. That's the compounding effect. The system gets more useful the more it knows, and the more it's used, the more it knows.</p><p>The next goal I'm working toward is a Nova self-improvement loop — agents that can review, refactor, and improve Nova's own code. That's not science fiction at this point; it's an engineering problem with a clear shape. You need agents that can read a codebase, identify specific improvement targets, propose changes with testable outcomes, and hand off to a human (me) for review before anything is committed. The autonomous part is the identification and drafting. The human part is the judgement call on whether to apply it. I'm not trying to remove myself from the loop — I'm trying to make the loop faster.</p><p>What I find genuinely interesting about building Nova is that it forces clarity about what "useful" means. A lot of AI tooling is built around impressive demos. Nova is built around my actual workflow, which is less glamorous — managing multiple projects simultaneously, keeping context across long-running decisions, drafting technical documentation, handling the administrative surface area of running a consultancy. None of that makes for a good conference talk. All of it is where real leverage lives.</p><p>If you're building something similar — a personal AI layer over your own work — the advice I'd give is this: don't start with the architecture. Start with the one task you do repeatedly that you hate, and build something that handles exactly that. Get it working well enough that you trust it. Then expand from there. The 34-agent system didn't start as a 34-agent system. It started as something much smaller that I actually used.</p><p>Production, for me, means it's load-bearing. Nova is load-bearing now. That's the milestone.</p>