Should You Build an AI Prototype or Fix the Workflow First?
By Zechariah Myrick · August 31, 2026 · 8 min read
Do not start with ‘Which AI tool should we buy?’ Start with one repeated work moment and ask whether a small version could make a real decision easier. If the work is not yet clear, repeatable, safe to discuss, and owned by someone who can act on the result, the better first move may be a manual workflow fix, a short research step, or specialist review—not an AI prototype. For a Naples or Collier County decision-maker, this check protects a serious idea from becoming an expensive collection of features.
A prototype can clarify one uncertainty. It does not prove demand, accuracy, security, compliance, cost savings, revenue, or that automation is the right answer.
The question is readiness, not enthusiasm
A capable professional, owner, adviser, consultant, manager, or serious founder may bring real expertise, notes, a sketch, a website problem, or a long ChatGPT conversation. Those are useful starting points. They are not, by themselves, proof that a prototype should be built. The practical question is narrower: ‘Can we describe one work moment, one responsible person, one safe input, and one decision that a small version could help us make?’
The U.S. Small Business Administration describes market research as a way to understand customers and improve an idea. A workflow decision sheet is smaller than market research. It helps you choose a responsible next step, but it cannot establish that customers want the result, that information may be used, or that a proposed system will work.
Naples and Collier County are relevant here because a local owner or adviser can bring direct knowledge of the work and the authority to change it. That local context is not evidence that any particular AI project is needed or will succeed.
Use the four-path first-version decision sheet
Use public, fictional, or material you are authorized to discuss. This is a planning aid, not legal, security, financial, or professional advice. Write one sentence for each line, then choose only one path.
- 1. Name the recurring moment: ‘When ___ happens, the responsible person needs to ___.’ Keep it to one moment, such as explaining a public service, preparing a human-reviewed draft, or sorting an approved inquiry—not every step in the business.
- 2. Name the decision: ‘After seeing a small version, ___ will decide whether to ___.’ The decision might be to continue, revise, pause, ask for specialist review, or keep the work manual.
- 3. Name the safe material: Use an approved public summary, fictional example, or other permitted material. If the central value requires client, employee, credential, payment, health, legal, proprietary, or regulated information, stop and establish the needed permission and controls first.
- 4. Name today’s manual baseline: Describe what a person does now, where they hesitate, and what is missing. If no one can explain the current work consistently, fix that workflow before asking AI to reproduce it.
- 5. Choose one path: Build a bounded prototype when the moment, safe material, owner, and decision are clear. Run a manual test when the question is clear but the workflow has not been observed. Fix the workflow first when ownership, handoffs, or the current process are unclear. Pause or seek specialist review when consequences, permissions, security, accessibility, or regulation are material.
What each path produces
- Build a bounded prototype: A public explanation, clickable concept, human-reviewed checklist, or demonstration using safe material. Its job is to let someone see, try, or review one thing before a larger commitment.
- Run a manual test: A person follows the proposed checklist or draft process without automation and records where context, correction, approval, or exceptions are needed. This can expose a bad premise before software makes it harder to change.
- Fix the workflow first: A one-page description of the current handoff, decision owner, input, output, exception, and stop condition. A clearer process may be the useful outcome even if no AI is added.
- Pause or seek specialist review: A documented question for the person qualified to address permissions, privacy, security, accessibility, legal, financial, health, employment, safety, or other consequential requirements. A pause is a responsible decision, not a failed project.
A simple example without private details
Imagine an adviser with a public service page whose visitors often ask the same basic question. The decision sheet might say: ‘When a visitor reads the page, the owner needs to know whether the next step is understandable.’ The safe material is the approved public page. The manual baseline is a short, human-reviewed explanation sent after common questions. A bounded prototype could be a revised public page or clickable explanation—not an intake bot connected to private records. The next decision is whether intended readers can explain the next step clearly enough to keep refining the page.
Now change the example: the proposed value depends on interpreting private client files or making recommendations that affect someone. The same sheet should choose pause or specialist review. A polished chatbot answer does not grant permission, make the workflow appropriate, or transfer accountability.
Keep accountability and uncertainty visible
NIST's voluntary AI Risk Management Framework emphasizes documented roles, responsibilities, human oversight, scope, and feedback. A small organization does not need an enterprise program to use that principle. It does need a named decision owner, an honest boundary around the first version, and a clear handoff when the result is uncertain or consequential.
Label every note as a confirmed fact, an assumption, or an unknown. ‘This public page is approved’ is a fact to verify. ‘A shorter page will reduce repeat questions’ is an assumption to test. ‘Whether any client material may be used later’ is an unknown until the appropriate permission and controls exist. This separation keeps an AI-generated draft from becoming an invented requirement or outcome.
What ChatGPT can help with—and where it stops
ChatGPT can organize an approved, non-confidential summary into a four-path decision sheet, identify missing questions, and draft neutral wording for a manual test. A bounded request could be: ‘Using only this approved public summary, list the recurring moment, responsible decision, safe material, current manual baseline, and the questions that would determine whether to prototype, test manually, fix the workflow, or pause. Do not invent permission, demand, cost, technical capability, compliance, or results.’ A responsible person still needs to check every answer.
ChatGPT cannot decide whether your organization may use information, validate customer demand, make a high-stakes judgment, establish security or compliance, or accept responsibility for the result. Do not put confidential, regulated, client, employee, credential, payment, health, legal, or proprietary details into a public form or general chatbot. Start with a fictional example, approved public material, or safe summary.
Who this fits—and who should not start with a prototype
This approach fits a decision-maker with a real business or creative idea who wants to make something tangible with patient one-to-one help. It is useful when the person can describe a narrow moment and wants to choose the smallest useful version before authorizing broader work.
It is not a fit for someone seeking a guarantee, a set of free prompts, an immediate full launch, or a shortcut around a missing permission or required review. If the first meaningful version would process sensitive records, make decisions about people, access systems you do not control, or affect a regulated or consequential setting, begin with the appropriate human review instead.
Turn the chosen path into a useful conversation
For a careful builder's approach, see Zechariah's background and working approach and the related guide on what to include in a first AI build scope. The scope guide helps after you choose the prototype path; this decision sheet helps you decide whether that path is warranted at all.
If you can bring a safe summary, a real work moment, and the decision that needs clarity, bring them to an idea-to-prototype conversation. You bring the experience and material you are allowed to discuss. Zechariah helps choose and build the smallest useful version—or identify when a manual test, workflow fix, or specialist handoff is the more accountable next move.
Sources and local context
← Back to the AI Guides