Notifications
Clear all
Topic starter
20/08/2026 5:59 pm
TL;DR: Agents often fail on complex workflows not because they lack tools, but because planning is left as a prompt-level suggestion, according to Arize. The practical shift is to treat planning as a structured, enforced control with visible task state and finish gates, which is increasingly relevant as AI agents act more like non-human identities in production systems.
NHIMG editorial — based on content published by Arize: How to Build Planning Into Your Agent (The Architecture That Actually Works)
Questions worth separating out
Q: How should security teams govern AI agents that run long, multi-step workflows?
A: Security teams should require durable execution, full event history, and clear ownership for every multi-step agent workflow that touches sensitive data or privileged tools.
Q: Why do AI agents fail when planning is only a prompt instruction?
A: Prompt-only planning creates text the model can ignore, forget, or bury under later tool output.
Q: What breaks when an agent cannot track task status during execution?
A: Without explicit task status, the agent cannot distinguish active work from finished work or blocked work.
Practitioner guidance
- Define planning as a governed runtime object Store plan state outside free-form conversation history so the agent can inspect, update, and validate tasks across turns.
- Add an explicit in_progress state Use at least four statuses for agent work: pending, in_progress, completed, and blocked.
- Pin the current plan to every iteration Inject the active plan immediately after the system prompt and before session history so the agent always sees task intent ahead of noisy tool output.
What's in the full article
Arize's full article covers the implementation detail this post intentionally leaves at the architectural level:
- The exact prompt and tool patterns used to create todo_write, todo_update, and todo_read in production
- The PlanMessage formatting approach, including how the plan is pinned ahead of session history
- The finish-gate enforcement logic that rejects incomplete task closure
- The heuristics used to decide when planning should trigger for multi-step requests
👉 Read Arize's analysis of planning architecture for multi-step AI agents →
Agent planning architecture: why multi-step workflows keep failing?
Explore further
20/08/2026 6:38 pm
Planning state is now an identity control problem, not just a UX problem. When an AI agent can select tools, carry context, and continue work across iterations, its planning layer becomes part of the control plane. If that state is only expressed in free-form text, it is neither enforceable nor auditable. For NHI and agentic AI programmes, the practical conclusion is that plan persistence belongs in governed runtime state, not conversational memory.
A question worth separating out:
Q: How do you know if an agent planning system is actually working?
A: Look for reliable multi-step completion, fewer looped tool calls, and successful finish calls only after all tasks are resolved. If the agent regularly forgets the last step, drifts after several tool calls, or needs repeated resets, the planning layer is not being enforced strongly enough.
👉 Read our full editorial: Planning architecture is what makes agents complete multi-step work
