Preparing a strategy review in 30 minutes instead of an afternoon
Almost none of the time spent preparing a strategy review is spent on strategy. It goes on collection: chasing eight owners for a status, opening Linear to see whether the project actually moved, retyping numbers into a deck, and discovering an hour before the meeting that two initiatives have no owner and one indicator has not been updated since March.
That work is mechanical, and it is exactly what an agent with access to both your strategy and your execution tools can do unattended. The point is not to save the afternoon — it is that when collection is free, the review can be held monthly instead of quarterly, and the meeting starts at the disagreement instead of at the status round.
What actually eats the afternoon
Four things, in our experience and in the reports of the teams we have watched: finding out what moved since the last review; reconciling what people say with what the execution tool shows; noticing structural problems (an initiative connected to no objective, an objective with no measurement); and writing an agenda that puts the two decisions that matter at the top instead of walking every line in order.
None of that requires judgement about the strategy. All of it requires access to several systems at once — which is precisely what an agent has and a slide deck does not.
The loop, concretely
hoshin.app ships this as a guided prompt called `weekly_sync`, so the protocol is not something you have to invent or re-explain each time. It encodes the sequence and, importantly, the routing rule.
The agent reads the initiatives in the South quadrant of the X-Matrix. Each one can carry an `external_ref` — the URL of the Linear project, Jira epic or GitHub milestone that implements it. The agent follows that reference through the execution tool's own MCP server, reads the real progress, and writes it back with `update_element_status`. Then it recomputes the completeness score, and drafts the agenda from what the score flags rather than from the order of rows.
The routing rule matters more than it sounds: strategy and reviews belong in the strategic memory, tasks and issues belong in the execution tool. Nobody should be maintaining a copy of the sprint in a strategy tool. The `external_ref` is the join, and it is the only duplication we allow.
- Read the initiatives and their external references.
- Pull real progress from Linear, Jira or GitHub through their own MCP server.
- Write statuses and indicator values back with `update_element_status`.
- Recompute the completeness score and let it name the structural gaps.
- Draft an agenda ordered by what drifted, not by the order of the matrix.
The three questions a review should answer
What is drifting, why, and what decision is needed now. A review that answers those three has earned its hour; a review that reads out twelve statuses has not.
The completeness score is what makes the first question answerable without opinion. It reports coverage, correlation density, orphan elements, complete chains from vision down to a KPI, and the balance of leading versus lagging indicators — and it names the specific elements at fault. An initiative wired to nothing. A vision that never reaches a measurement. An objective tracked only by lagging indicators, which means you will find out too late by construction.
The second question — why — is where humans are irreplaceable, and where the meeting should spend its time. The tool's job is to get you there by minute three.
For a consultant running several clients
The value multiplies with the number of strategies you maintain. One leadership team has one review to prepare; a strategy consultant has eight or twelve, each with its own tooling, cadence and level of hygiene.
Two things matter at that scale. First, the preparation must be reproducible without you: a prompt your agent runs against any client's matrix, not a personal ritual. Second, client material must be able to stay in your environment — which is why the sovereign local mode exists, one data directory per engagement, no account and no network.
Frequently asked questions
Does this require connecting hoshin.app to Jira or Linear?
No integration to build on our side. Your agent already talks to your execution tool through that tool's own MCP server; we only store the `external_ref` pointing at the right project. That is deliberate — an integration we maintain would break, and it would tempt us to duplicate your backlog.
What if our initiatives are not in a tool at all?
Then the loop is shorter but still useful: the agent updates statuses and indicator values from what you tell it in the conversation, recomputes the score, and drafts the agenda. The `external_ref` makes the pull automatic; it is not a prerequisite.
Can the agent write, or must someone update the matrix by hand?
It writes. `update_element_status` moves statuses, progress and notes, and since July 2026 `create_matrix`, `upsert_element` and `set_correlation` let an agent build and restructure the matrix itself — including turning a strategy document into a scored matrix in one conversation.
Thirty minutes is a claim. Where does it come from?
From the shape of the work rather than a benchmark we can hand you: collection is the part that takes hours and it is the part the agent does. What is left is reading a diagnosis and deciding what to argue about. If your matrix has never been maintained, the first pass will take longer — it will surface everything at once.