OKR Generation Agent — Mark's AI Strategy Tool

Overview

Mark has built an AI agent that generates client OKR drafts by analyzing all available client data — SEO performance, ad accounts, website health, and historical work. The agent is designed to replace the manual, time-intensive process of writing quarterly OKRs from scratch, which the team consistently lacks bandwidth to complete.

The tool is not a one-shot generator. It's designed for iterative refinement through a human feedback loop, with account managers providing directional guidance (e.g., "more of this, less of that") that the agent integrates in a second pass.

How It Works

Data Sources

The agent ingests all available client signals:

• SEO data (rankings, technical health, content gaps)
• Paid ad performance
• Website analytics and structure
• Historical work and prior strategy documents

Based on this, it surfaces recommended focus areas for the quarter.

Generation Process

1. Mark runs the agent for each client and posts the resulting OKR draft in Slack
2. Account managers review and provide directional feedback — no need for precise edits, just high-level guidance ("focus more on X," "deprioritize Y")
3. Mark re-runs the agent with the feedback incorporated
4. Iteration continues until the account manager is satisfied

Feedback can be sent as a Slack message or email — it does not need to be structured or technical.

Scale: Agent Swarms

For high-volume use cases, the agent can be run as a swarm of parallel instances. In the meeting where this was discussed, 25 agents were running simultaneously to generate strategy documents for PaperTube's 200+ accounts — processing roughly one account every four minutes.

"It's a crazy amount of research that you get in a short time." — Mark Hope

This swarm approach makes it feasible to produce strategy documentation at a scale that would be completely impractical manually. At the time of the Q2 OKR meeting, only ~50 of PaperTube's accounts had existing strategy docs; the swarm was filling in the remaining 150+.

Why This Matters

The team had not completed manual OKRs for Q2 due to time constraints. The agent addresses this directly — it handles the first draft entirely, leaving account managers to apply their client relationship knowledge as a refinement layer rather than doing the full authoring work.

This also creates a reusable, scalable process: the same agent and feedback loop can be applied each quarter without rebuilding from scratch.

Limitations and Human Judgment

The agent works from data signals, not relationship context. Account managers are expected to correct for:

• Recent strategic pivots the client has communicated verbally
• Relationship dynamics that affect what's realistic to propose
• Changes in client priorities since the last data sync (e.g., Citrus had shifted focus since prior OKRs were generated)

The agent's output is a starting point, not a final deliverable.

Related

• ^[1] — primary test case for agent swarm at scale
• ^[2] — example of client whose priorities shifted, requiring feedback-loop correction
• ^[3] — related AI automation tool for account-based marketing outreach
• ^[4] — broader OKR process context