Codeless AI Assistants and Automation

Aabha AI Academy • Course development edition • 14 October 2026

Course purpose

Build and evaluate a supervised no-code assistant and a small workflow that turns a fictional equipment-loan request into a checked review record. Learn where AI interpretation helps, where fixed rules belong, and how to stop, recover and hand work to a person. The outcome is a portable assistant configuration, a workflow design or optional live no-code prototype, and an evidence-backed operating guide.

This course follows AI for Everyday Work and Prompting. It assumes you can write a bounded brief, use supplied evidence, check an output and follow a manual procedure. It does not repeat workplace writing, general prompting or simple data analysis as separate units. The new learning is persistent behaviour, knowledge lifecycle, structured interfaces, event handling, review states, duplicate prevention and operational testing.

Entry and equipment

Use a text editor or paper, an ordinary browser and the supplied fictional packs. No programming, purchase, new account, API key, real customer data or connected workplace app is required. Optional live paths need an existing permitted account with the documented feature and available quota. A facilitator may demonstrate them. The no-account route is explicitly a design and rehearsal route; it does not prove a live model or vendor workflow ran.

Complete this readiness check: given a two-paragraph policy, identify one unsupported promise, rewrite it without inventing facts and name a stop condition. If that is unfamiliar, complete the prerequisite course first.

Learning outcomes

By the end, the learner can:

• Write a narrow assistant contract with inputs, outputs, authority and stop conditions.
• Separate stable instructions from changeable, permission-appropriate reference material.
• Define a structured record and map trigger fields without guessing missing values.
• Design mutually exclusive routes and an explicit human approval state.
• Handle duplicate events, temporary failures, uncertain outcomes and exhausted retries.
• Test normal, boundary, missing-data, conflicting-source and adversarial cases.
• Measure safe completion, review load and cost using honest denominators.
• Deliver a versioned capstone with a runbook, test evidence and a clear release decision.

Sequence and planned time

1. Choose the job and its boundaries — 75 minutes. Deliver a one-page service contract and a decision about what may be automated.
2. Configure a reusable assistant — 120 minutes. Deliver stable instructions, an output contract and three behaviour tests.
3. Build a trustworthy knowledge pack — 120 minutes. Deliver a source manifest, authority rules and an updated answer across two policy versions.
4. Turn an event into a structured record — 150 minutes. Deliver a data dictionary, field map and manual-trigger prototype or trace.
5. Route work into human review — 120 minutes. Deliver mutually exclusive rules, a review record and the optional Make routing lab.
6. Recover without duplicating work — 135 minutes. Deliver a duplicate ledger, failure matrix and bounded recovery procedure.
7. Test and operate the workflow — 150 minutes. Deliver a test suite, labelled results, a capacity estimate and an operating plan.
8. Build and defend the capstone — 180 minutes. Deliver the full supervised workflow package, change test and release decision.

Total: 1,050 minutes, or 17.5 hours. This is a planning estimate, not a measured completion promise. Account setup, breaks, waiting for quotas and optional extensions are additional.

Assessment and completion

Modules 1–7 are formative: attempt the exercise before opening the separate solutions book. Each has applied checks with explanations and a concrete artifact to retain. The capstone is assessed out of 100 with a proposed pass threshold of 80, plus mandatory gates for privacy, human authority, source accuracy, safe failure handling and honest evidence labeling.

Two completion descriptions are available. Design and rehearsal completion requires the supplied offline cases, trace logs and all rubric gates. Live prototype completion additionally requires learner evidence from an actual assistant run and actual no-code workflow run, including a boundary case and safe stop. Neither is production certification. All sample outputs in this edition are authored illustrations unless explicitly marked otherwise.

Tool approach checked on 14 October 2026

The portable configuration is the primary learning artifact. Optional Gemini Skills or Gems can host it when available in the learner's existing account. Google is transitioning capabilities, so the guide uses availability checks and an ordinary-chat fallback. Make is the optional no-code workflow tool because a built-in-tools scenario can demonstrate routing without connecting a mailbox or other external app. No AI API connection or automatic sending is required.

The Make Free plan is documented with 1,000 monthly credits; this is a quota, not a guarantee that every account or feature is free. AI features, third-party services and higher limits can have additional costs. Google lists Gems without a Google AI plan and describes Skills without a subscription, subject to availability and account requirements. Current OpenAI documentation restricts new custom-GPT creation on personal accounts and describes migration, so this course does not depend on it. Recheck the linked official sources before teaching any live lab.

Files and teaching method

The package contains a learner reader, eight standalone lesson files, a workbook, synthetic source and event packs, separate solutions, an instructor guide, original SVG diagrams, tool notes and a verification report. Every diagram has a text explanation. Publication and enrolment are outside this content-development task.

Public course brief: https://aabhaai.com/courses/codeless-assistants/