Fluidstack Engineering: The Real Work in AI Agents is Designing Handoffs
MxMnr · x · 2026-08-02
Fluidstack's network automation team shared their engineering experience building an internal AI agent, agent-deploy. The article emphasizes that the real challenge in agent development isn't picking the model, but designing rigorous handoffs and control logic.
- Core Architecture: Instead of using a general coding agent, they built a custom agent loop. An independent analysis revealed that only 1.6% of an agent's code handles AI decision-making, while 98.4% is dedicated to permission gates, context management, tool routing, and recovery.
- Safety Mechanisms: General agents often keep safety rules as polite requests in system prompts. This team encodes safety directly into the code, covered by tests.
- Practical Application: The tool is in daily use. Operators can ask questions in natural English, and the agent gathers fleet-wide state to output grounded conclusions, recommendations, and auto-files Jira tickets with evidence. Structurally, it never alters the network without human approval.
More from coding & agent
- ERP•AI launches headless SaaS with $0 seats that bills machines: $1 buys 4,000 record writes — rohanpaul_ai · 2026-10-16
- FDE-Gym benchmark shows huge gap on real AI delivery tasks: top model scores 0.563 — DeepLethe · 2026-10-16
- Vibe-coded dashboard leaked franchise financials before anyone logged in — shauntrennery · 2026-10-16
- Fintech HR built a sourcing agent on 60,000 past applicants — 30 qualified candidates in a month — G0NZ0BOY · 2026-10-16
- Orchestrating a 100-image portrait of humanity for aliens with ChatGPT, Argon and p5.brush — dx8xb · 2026-10-16
- 3x DGX Spark cluster runs GLM-5.3-Flash locally behind a proxy, 1.75M tokens/day at 119W — Teknium · 2026-10-16