PretectalStart here
Open the prototype workbook and finish this sentence: when [person] needs to [task], they currently [steps], and the difficult part is [observed friction]. Ask that person whether you have described the task accurately.
Your finished output: One person has attempted the core task, you have checked the output and a failure case, and the saved prototype includes its brief, test notes, limits, and a decision about what comes next.
Pause and revise when: You cannot name the intended user, the core task depends on unconfirmed access, or the prototype needs an explanation to complete its main path. Resolve that gap before adding features.
Before you start
Start with a task you understand and a person you can learn from. A prototype should help you test one useful change before you invest in a larger product. Using AI to build does not mean the finished tool needs an AI feature.
For: First-time builders, domain experts, and people improving a team workflow.
The deliverable: A working prototype, a focused build brief, and user-test notes.
Gather these inputs
- One recurring task and a person who performs it
- An example of the current input and desired output, using sample data
- A way to build and preview a small app, plus a place to record decisions
The workflow
01Find the friction
Watch or walk through the current task. Record the trigger, steps, handoffs, and points where someone repeats work or loses context. Choose one person and one moment to improve. Capture the current time or error pattern if you can observe it; otherwise leave the baseline unknown.
- AI can help
- Organize your notes into a task map and suggest questions about missing context.
- You decide
- Confirm the problem with the person doing the work. Choose the useful change yourself; an AI suggestion is a hypothesis to test.
- Keep
- A problem statement and a description of the current task.
02Write the smallest useful brief
Define one input, one user action, and one useful output. Write what a successful attempt looks like and what the first version excludes. Decide whether an existing tool already meets the need. If you build an app, choose the interface, framework, and backend needs separately; a frontend choice does not decide where data or authentication lives.
- AI can help
- Compare small implementation options against your constraints and turn the chosen path into a short build brief.
- You decide
- Keep the first task complete but narrow. Add an AI feature only if its variable output helps the task and you can evaluate that output.
- Keep
- A brief with acceptance criteria and explicit scope boundaries.
03Build one complete path
Give your coding assistant the brief, sample input, expected output, and acceptance criteria. Build the path from opening the tool to receiving the output. Save a known working version before each change. Include a clear empty state and recovery from a bad input. Keep repeated instructions, good examples, and checks in a reusable skill or project instruction file; review what the assistant actually follows.
- AI can help
- Draft the interface, implementation, and explanations of the changed files. Make one bounded change at a time.
- You decide
- Run the result yourself. Check that each requested behavior works and that the prototype makes its limits clear. Keep real team data out until you have agreed where it can go.
- Keep
- A runnable prototype using sample data, with setup instructions.
04Try it with a person
Give the intended user a realistic task and sample input. Watch without explaining the interface. Record whether they finished, where they hesitated, and what they expected. Check the resulting output against the acceptance criteria, including an incomplete or unexpected input.
- AI can help
- Draft neutral test tasks and group your observed notes into issues after the session.
- You decide
- Keep observations separate from interpretation. A successful demo or a positive comment does not establish ongoing usefulness.
- Keep
- A task record with observed issues and the next change to make.
05Choose the next version
Fix the issue that most prevents task completion, then repeat the task. Record what changed and decide whether to stop, revise, or prepare a limited team trial. Preserve the brief, sample data, source files, and unresolved questions so another person can understand the prototype.
- AI can help
- Implement the selected fix and draft a change log from changes you verified.
- You decide
- Decide whether the task is useful enough to continue. Keep the prototype labeled as a prototype until its data handling, access, ownership, and maintenance have been addressed.
- Keep
- A tested version, a decision to continue or stop, and a clear next owner.
A useful starting prompt
This prompt scaffolds a task. It does not verify its own output or replace the review described above.
Use material you are authorized to share. Replace the bracketed inputs and review the result against the original sources.
What changes in practice
A teammate collects the same project details through several messages, then copies them into a handoff document.
A small form collects the required details, shows what is missing, and generates an editable handoff for the teammate to check.
This is an illustrative prototype, not a reported Pretectal deployment. The useful first test is whether the teammate can complete the handoff with fewer missing details; measure any time difference during the trial.
Use the getting-started guide ↗Ready for handoff
- An intended user can attempt the complete task with sample data.
- The expected output and at least one failure case have been checked.
- Observed results are separated from assumptions and desired benefits.
- The prototype has a saved version, setup notes, and an explicit next decision.