34 Agents and a Production Mindset: What Running Nova Actually Looks Like

<p>Nova is in production. That phrase sounds clean, but the reality is messier and more interesting than the label suggests. It's not deployed in the sense that a SaaS product is deployed — there's no signup page, no billing, no uptime SLA I'm liable for. It's in production in the sense that I actually depend on it now. It runs. It fails sometimes. I fix it. I depend on it again. That cycle is what production means to me.</p><p>The system currently runs 34 agents. That number surprises people, but it makes sense when you look at how it's structured. There are general chat agents, pipeline execution agents, ERP specialists that understand AIREP's data model, language translation agents, and a handful of others I've built to handle specific recurring tasks. They're not all running simultaneously — Nova orchestrates which agents are invoked based on context. But they're all live, all callable, all part of the same system.</p><p>Getting here wasn't a single build sprint. It was a slow accumulation of decisions — some good, some that I had to undo. The architecture that works now is one where agents are composable and narrow. Each agent does one thing well. The orchestration layer decides what gets called and in what order. That separation of concerns sounds obvious in retrospect, but I spent time early on building agents that tried to do too much. They were brittle. Narrow agents are easier to test, easier to replace, and easier to reason about when something goes wrong.</p><p>The thing I keep coming back to is the difference between AI as a tool and AI as infrastructure. Most people are still using AI as a tool — you open a chat window, ask a question, copy the answer. That's fine, but it's not leverage. Leverage is when the AI is woven into your workflow so deeply that removing it would break things. Nova is at that point for me. It handles memory across conversations, it routes tasks, it drafts, it searches, it summarises. It's not augmenting my work — it's load-bearing.</p><p>That shift changes how you think about building. When AI is infrastructure, reliability matters more than capability. A flashy model that hallucinates 10% of the time is worse than a simpler system that's predictable. I've made deliberate choices to keep certain agents conservative — they do less, but they do it consistently. Consistency is what lets you actually trust a system and hand it real work.</p><p>The current phase I'm calling production orchestration is about hardening those patterns. It's less about adding new agents and more about making the existing ones more robust, more observable, and more self-aware. One of my active goals is building a self-improvement loop — agents that can review Nova's own code, flag issues, and propose refactors. That's not science fiction at this point. It's an engineering problem, and it's the next concrete thing on the list.</p><p>I'm also thinking about how Nova connects to my other projects. AIREP is the obvious one — there are ERP-specialist agents already, and the natural next step is tighter integration where Nova can answer questions about live data, surface anomalies, or draft reports from within the ERP context. Find a Sign is another candidate. The marketplace has a transparency-first model — no pay-to-rank — and AI can reinforce that by helping customers find what they actually need rather than what a supplier paid to surface. The values and the technology point in the same direction.</p><p>None of this is to say Nova is finished or even particularly impressive from the outside. It's a personal system. It's rough in places. But it works, and it compounds. Every agent I add makes the next one easier to build because the infrastructure is already there. Every task I hand off to Nova gives me back time I put into the next thing. That compounding effect is the real argument for building this kind of system — not the demos, not the agent count, not the architecture diagrams. Just the fact that it makes you faster, and the gap widens over time.</p>