AI & Your Work: Examine the Work Before You Automate It.

Teaching

Choose one recurring piece of work that consumes more attention than it should. Keep the example private and nonconfidential. You need an observable process, not a product idea.

The finish line is a defensible next move—including redesign, human-led work, or more investigation—not an automatic AI recommendation.

Teaching

Tool-first thinking compresses several decisions into one purchase. Ask what is happening in the work, where it is actually stuck, and what—if anything—technology should do next.

Novelty, vendor preference, and competitive pressure are not diagnosis. Automation is one possible intervention, not the objective.

Teaching

This public course teaches the complete decision method through a prepared fictional case. It does not diagnose a confidential case or authorize implementation.

Individualized assessment, case research, collaborative construction, and client-specific outputs require the Clinic or a separately scoped private engagement. The developing Decision Record Review package is not available for booking.

Teaching

Describe the shape of work, not sensitive contents. Do not enter names, employee or health records, payment data, passwords, tax records, trade secrets, privileged material, or confidential information.

A model does not replace authoritative sources or qualified professionals. Professional decisions retain their professional boundaries.

Teaching

Adoption statistics do not create an implementation mandate. Definition, wording, firm size, sample, date, and self-report limits travel with every claim.

Uneven adoption means urgency cannot substitute for a clear work problem and testable value hypothesis.

Teaching

Customer demand, capacity, hiring difficulty, and rising costs justify examining work; they do not prove AI caused or will solve the problem.

Choose a constraint important enough to fix, bounded enough to test, and safe enough to learn from.

Teaching

‘We need a chatbot’ says almost nothing. ‘Qualified inquiries stall because nobody owns the handoff after the first call’ names trigger, failure, and consequence.

‘Automate proposals’ becomes examinable only when approved notes, drafting, scope/price review, owner, and time burden are visible.

Teaching

Define the work and desired change. Examine friction, inputs, judgment, risk, value, capacity, and context. Decide among four legitimate findings. Build the smallest safe next action.

Build comes last. Understanding and deciding earn the right to build.

Teaching

Friction locates recurring loss. Inputs identify what correct performance requires and who may use it. Judgment names decisions involving interpretation, approval, empathy, negotiation, taste, duty, or responsibility.

Risk identifies possible harm and detection. Value defines observable improvement before a demonstration can impress you.

Teaching

Kōkua Studio is a fictional two-person Hawaiʻi service studio. Its owner spends about six hours each week converting discovery-call notes into customized proposals.

The tempting answer—use AI to write proposals—is held back while the work is examined. This is not a client story or claimed result.

Teaching

The process is: call ends → notes scattered → services checked → price checked → draft written → owner reviews → proposal sent. Six hours and delay are symptoms.

Causes may be scattered notes, unclear ownership, excess variants, stale language, tacit pricing, or drafting. Faster text can move instability downstream.

Teaching

Correct work needs an approved call summary, current services, real availability, pricing rules, and approved language. Existence does not establish currency, completeness, authority, or tool permission.

A safe first test uses fabricated or deliberately redacted inputs and controlled approved references. Convenience is not permission.

Teaching

The owner decides client fit, realistic scope, price, supportable promises, exceptions, and final send. Her value is not typing sentences.

These judgments are the responsibility structure. AI may organize or draft but must not silently inherit undelegated authority.

Teaching

Failures include exposed notes, invented deliverables, stale prices, omitted limits, and polished unapproved promises. Reduce likelihood and design detection before harm.

‘A human reviews it’ is incomplete without a reviewer, checklist, evidence, escalation route, and missed-error consequence.

Teaching

Kōkua Studio’s baseline is about six owner-hours weekly plus current turnaround. Measure draft time, correction rate, turnaround, and unapproved scope, prices, or promises.

Time saved is not success if substantive errors rise. Precommit the baseline, measure, and guardrail.

Teaching

The map permits four equal findings: bounded experiment; redesign first; keep human-led; or insufficient information.

‘No AI’ and ‘not yet’ are successful when evidence supports them. The method evaluates work; it does not force adoption.

Teaching

Move only as far as needed: ownership → template/checklist → ordinary automation → AI-assisted step → integrated workflow → custom system.

Complexity creates setup, testing, training, maintenance, security, and exit costs. Do not buy architecture because a demo hides labor.

Teaching

Test ten fabricated or redacted historical cases against an approved source packet. Produce drafts only; the owner checks fit, scope, price, promises, missing facts, and prohibited language.

Compare time and correction burden with baseline. Stop for confidential input, an unapproved promise surviving review, or correction burden erasing the benefit.

Teaching

Connect an input boundary, approved references, structured instruction, output, review checklist, decision owner, destination, evidence log, and maintenance owner.

A prompt can create a moment. A workflow must survive ordinary use, exceptions, turnover, vendor change, and wrong output.

Teaching

Hawaiʻi context can change tax, licensing, shipping, staffing, infrastructure, bandwidth, relationships, and failure consequences. These are requirements, not decoration.

Route jurisdictional questions to current authoritative sources and qualified review. Generic generated checklists must not flatten distinct obligations.

Teaching

Document trigger, owner, inputs, steps, decisions, exceptions, destination, and time. Mark sensitive inputs and human judgment. Define one reversible improvement with fabricated, historical, or non-sensitive material.

Before buying, name user, permitted data, output, reviewer, total cost, measure, stop evidence, and review date.

Teaching

Use the public method independently when you can own a bounded analysis. Use the Clinic for collaborative construction around a nonconfidential case.

For confidentiality, research, jurisdictional evidence, or written analysis, begin with a nonconfidential fit conversation about a separately scoped private engagement. The developing Decision Record Review package is not available for booking. Paid value is individualized judgment and build work—not withheld education.

Teaching

General questions address the method, evidence, or fictional case. If an answer depends on missing facts, identify needed information rather than inventing certainty.

Response frame: the general problem is ___. AI may help with ___, while ___ stays human-owned. A safe test requires ___. The main risk or missing fact is ___.

Teaching

Find one worthwhile constraint. Protect people and information. Preserve judgments carrying responsibility. Run one bounded test—and keep permission to stop.

Completion means you can defend a next move with evidence. It does not make AI win, authorize deployment, or replace professional review.