From Curiosity to Production: Lessons from a Year of Agentic AI

Just over a year ago, we formed Cortex Reply from within Reply to focus on Agentic AI.

A lot has changed since then.

Over the past year we’ve invested heavily in research and development, built prototypes, and delivered production systems designed to harness the transformational potential of Agentic AI. That journey has taken us quickly from experimentation to real-world delivery and, more importantly, to hard-earned lessons about what it actually takes to build systems people can trust.

Agentic AI is powerful. It is also far more complex than many early narratives suggest.

But complexity does not mean chaos. With clear logic, strong architectural principles, and a disciplined approach that works backwards from desired outcomes, production-grade agent systems are achievable.

Over the coming weeks, we’ll share the opportunities, challenges and solutions we’ve encountered along the way.

Some of the key production challenges we’ll explore include:

- Context: Getting the right context to the right agent at the right time is the single biggest practical challenge in real systems.

- Permissions and Identity: Agents with broad, ambiguous privileges create security and trust risks. Copilots and autonomous agents require fundamentally different identity and permission models.

- Observability and Testing: Without traceable inputs, outputs and tool calls, you cannot iterate safely. End-to-end quality assurance and per-agent traceability are essential.

- Long-running and Autonomous Tasks: Tasks that span minutes or days require resumability, retries, and structured human checkpoints that go beyond simple human-in-the-loop patterns.

- Knowledge and Retrieval: Retrieval breaks down without disciplined chunking, metadata, provenance and continuous updates.

This series is for those exploring the art of the possible with Agentic AI, those thinking seriously about governance and scale, and those looking to move from prototype to production with confidence.

If any of these themes resonate, we’d welcome the conversation.