As the year comes to a close, I wanted to keep this short and forward-looking.
2025 made one thing clear: AI is no longer just an assistive layer. The real leverage is emerging in agentic workflows: systems that plan, act, evaluate, and persist over time. What started as prompt-and-response has evolved into orchestration: coordinating tools, memory, environments, and feedback loops to produce real outcomes.
In 2026, I am most focused on long-running agentic tasks because of the business value they unlock. These are not agents that run for minutes. They are systems that operate across hours or days and handle context shifts, interruptions, and changing objectives without constant supervision. This year, my longest continuous session reached 22 hours, 2 minutes, and 2 seconds. The goal next year is to double that. This is not an endurance exercise. It is evidence that long-horizon work can reliably drive real outcomes.
That shift changes how we build, test, and trust these systems. Orchestration becomes infrastructure. Reliability matters more than cleverness. Reproducibility starts to look less like determinism and more like resilient progress.
If you've been experimenting at the edges, building agents, wiring tools together, or stress-testing workflows, 2026 is the year this work compounds.
Wishing you a restful holiday season and momentum in the year ahead.
Subscribe for new essays
Get updates when new writing goes live.
Mark Agate is a flight controls engineer focused on agentic systems, automation, and building test-driven workflows that make powerful tools safe to operate.