Choose a step to improve—and a responsibility to protect Leading Deeper with AI — Jasmine James https://jasminejames.dev/leading-deeper/choose-ai-assistance Skill source (GitHub): https://github.com/jasminelsjames/leading-deeper-ai-skills/blob/e1d2e65d3922c5e376fc0c6e8cfbb2f85722b1a7/skills/find-low-leverage-workflows/SKILL.md Adaptable template based on the talk; not the original internal prompt. Version 1 / 2026-11-04 START HERE: ONE FIRST RUN Choose one step to improve with AI. Use the AI tool approved for your work with supplied material. Wording and layout will vary by model; look for the evidence, uncertainty and review decisions shown here. Connections and recurring runs depend on your setup. All examples use invented sources and results. They are illustrations, not reports from Jasmine’s workplace or measured outcomes. FICTIONAL INPUTS **[A1] Fictional weekly status workflow:** Mira owns the weekly customer status report. Analyst Sam gathers dated ticket updates, groups them by customer, proposes an interpretation of risks, and reviews the report with Mira. Mira decides priorities and what to tell customers. **[A2] Intended improvement and growth to protect:** Sam wants less repeated copying and searching, but is learning to recognize delivery risks. Keep Sam's first interpretation and Mira's coaching discussion. The customer outcome to support is an accurate status report; there is no evidence yet that assistance improves it. **[A3] Baseline and available evidence:** For three comparable weekly reports, Sam recorded 18, 22 and 20 active minutes gathering updates. Review, correction and maintenance time were not measured. Approved excerpts can include ticket IDs, update dates and status text. No live connector access is needed for a first trial. WORKFLOW 1. Human — Describe work and responsibility: Supply the steps, people, intended outcome and baseline. Explicitly name the risk interpretation and coaching that should stay with Sam and Mira. 2. AI — Design a bounded experiment: Assess each step separately. Propose assistance with collecting dated evidence, then define the owner, quality checks, measurement and stopping condition without assuming a time saving. 3. Human — Decide whether to try it: Mira and Sam review the plan. If they choose to run it, they measure human effort and corrections and review after two comparable reports; no workflow changes are made by this draft. ILLUSTRATIVE OUTPUT Two-report experiment proposal · draft **One step to assist:** For the next two comparable reports, propose using AI to turn approved ticket excerpts into a draft table: customer, ticket ID, update date and stated status, with a reference for every row. Mark missing or conflicting updates for review. Mira owns the outcome; Sam prepares and checks the evidence table. [A1, A3] **Responsibility and growth to protect:** Sam writes the first interpretation of risks before Mira's review. Mira retains prioritization and customer communication. Keep their coaching discussion. The proposed assistance covers evidence preparation, not those decisions. [A1, A2] **Quality, effort and review point:** Sam checks every row against its excerpt and checks that all supplied tickets are represented. Record active preparation, review, correction and maintenance minutes for both trial reports. The supplied gathering baseline averages 20 minutes across three reports; total baseline human effort is unknown. Compare like-for-like gathering separately, and do not declare an overall saving from that number alone. [A3] Review after the second report. **Stop or revise condition:** Stop using the draft table if it invents a status or source; revise the scope if checking it displaces Sam's risk practice or coaching. These are proposed conditions, not observed failures. Whether assistance improves report accuracy or reduces total effort remains unknown. No automation has been started. WHAT TO LOOK FOR ACROSS MODELS - One step to assist - Responsibility and growth to protect - Quality, effort and review point - Stop or revise condition HUMAN REVIEW - Does this help Sam with repetition while preserving risk interpretation, coaching and Mira's ownership? - Are the two-report scope, source checks and stopping conditions clear enough to assess before continuing? MCP AND TOOL SETUP No MCP is needed when you provide the workflow and baseline. Optional read-only data access can help establish a baseline before an experiment. ### Read optional baseline evidence - Needed when: Only when the baseline must be retrieved from an external system rather than supplied by the owner. - MCP or tool examples: An approved reporting/data connector or MCP; Local CSV or exported report tools - Tools must support: Read the agreed quality, effort or outcome measures; Check definitions and reporting periods; Record what is observed and what is missing - Access needed: Read access limited to the named measures and reporting window. This planning exercise does not need workflow edits, messaging, administration or automation permissions. - Without the connection: Provide an approved aggregate export, record a short manual baseline, or state that the baseline is not yet known. Describe the workflow, ownership and evidence you already have. If no baseline is available, define a small manual measurement rather than assuming a benefit. MODEL STARTING POINT Reviewed October 30, 2026. Start with GPT-6 Luna (gpt-6-luna). An economical first pass through a structured worksheet; keep the experiment small and the output short. Start with low reasoning effort. Move up to GPT-6.1 Sol (gpt-6.1-sol) when: Ownership, growth opportunities and competing constraints need a more careful examination. ALTERNATIVES - Claude Haiku 5.5 (claude-haiku-5-5): An alternative for the first-pass plan in your approved Claude-based setup. HUMAN RESPONSIBILITY The leader makes the choice. The model helps examine the tradeoffs; it does not decide whose development or responsibility to remove. Suggested fit based on official provider documentation, not our own speed or quality benchmarks. Test on representative examples before relying on a model. A model does not connect your tools or grant access. MCP support, available tools and model selection depend on the agent host and your organization's approved setup. API PRICE REFERENCE USD per one million input and output tokens, using standard first-party API pricing. These are not subscription prices or total workflow costs. Tool calls, reasoning, retries, caching and different tokenizers affect the final bill. GPT-6 LUNA Provider: OpenAI API model ID: gpt-6-luna Context: 1.05 million tokens Standard API: $0.10 input / $0.50 output per 1M tokens. Model documentation: https://developers.openai.com/api/docs/models/gpt-6-luna Pricing: https://developers.openai.com/api/docs/pricing GPT-6.1 SOL Provider: OpenAI API model ID: gpt-6.1-sol Context: 1.05 million tokens Standard API: $2 input / $10 output per 1M tokens. Model documentation: https://developers.openai.com/api/docs/models/gpt-6.1-sol Pricing: https://developers.openai.com/api/docs/pricing CLAUDE HAIKU 5.5 Provider: Anthropic API model ID: claude-haiku-5-5 Context: 1 million tokens Standard API: $0.10 input / $0.50 output per 1M tokens. For requests with prompts up to 100,000 tokens. Standard API: $0.50 input / $2.5 output per 1M tokens. For requests with prompts over 100,000 tokens; the higher rates apply to that request. Model documentation: https://platform.claude.com/docs/en/models/haiku-5-5/overview Pricing: https://platform.claude.com/docs/en/about-claude/pricing Machine-readable catalog: https://jasminejames.dev/leading-deeper/model-guidance.json PROMPT Help me choose a bounded AI-assisted experiment using ownership and leverage. PREFLIGHT Verify which tools are actually available and which source locations they can access. Do not assume that naming an MCP server means it is connected. Local file and terminal tools are valid and do not necessarily use MCP. If required access is missing, ask for an approved export or excerpt; state what cannot be checked or run. INPUTS Workflow and steps: [fill in] Outcome owner / current contributors: [fill in] Customer outcome or recurring burden: [fill in] Judgment, relationships or growth to protect: [fill in] Available baseline evidence: [fill in or state none] ASSESS Consider the steps separately, not the whole job. For each, ask: who should own the outcome, and where does continued human attention change it? Identify familiar repetition that may benefit from assistance and work where judgment, relationships, practice or coaching adds value. A step that is routine for one person may be useful development for another. Low leverage does not mean low importance. OUTPUT Recommend one small experiment. Name the step, intended improvement, bounded AI contribution, necessary context, owner and reviewer. Specify what stays with a person, how quality will be checked, a review date and a concrete stop-or-revise condition. Name the customer outcome or development opportunity the change should support. Separate observed facts from assumptions. If no baseline exists, propose a lightweight way to establish one. STOP Return a plan for review; do not automate or change the workflow. Do not assume that assistance frees time or that freed time automatically produces a better outcome. Define the evidence that would let us assess each claim. WHAT A PERSON REVIEWS - Is this the right step to assist, given the person and their development needs? - Are ownership and the human review decision clear? - Is the experiment small enough to test and stop? EVIDENCE TO CHECK - Compare the same quality, recurring effort and customer-outcome measures over comparable periods. - Record where attention actually went, rather than counting generated output alone. - Review the protected responsibility or development opportunity alongside efficiency.