Your agents lack context: Here's how to fix "You're absolutely right!" — Brandon Waselnuk, Unblocked
In one same-prompt test, a context engine nearly halved agent token use. The design goal is targeted, permission-aware retrieval that resolves conflicts across engineering systems.
Unblocked ran the same prompt and model with and without its context engine. The unassisted run used **21 million tokens**; the context-backed run used **10.8 million** and reportedly saved **2 hours** of wall-clock time.
For larger coding-agent jobs, retrieve organizational context before search and rework compound. Combine semantic retrieval with deterministic queries, rank conflicting sources, personalize by team and repository, and enforce the requester’s permissions.
Unblocked ran the same prompt and model with and without its context engine. The unassisted run used **21 million tokens**; the context-backed run used **10.8 million** and reportedly saved **2 hours** of wall-clock time. For larger coding-agent jobs, retrieve organizational context before search and rework compound. Combine semantic retrieval with deterministic queries, rank conflicting sources, personalize by team and repository, and enforce the requester’s permissions. These figures come from one vendor-presented task, without enough detail here to generalize the reduction. Curated files can become stale, while an MCP connection alone does not ensure the agent calls the right source or continues searching.
This adds a vendor-reported efficiency signal to the prior argument that coding agents need organizational context, while narrowing the design requirement beyond simply connecting an MCP server: retrieval must combine semantic and deterministic methods, reconcile conflicting sources, respect permissions, and keep searching when evidence is incomplete. The single-task result does not establish general savings or solve context freshness.