What will you build?

Website cards, a dashboard model, document folders, and a miniature shop arranged on a project workbench.

Build with AI · Sources reviewed 2026-11-19

Start with something you want to exist: a useful website, a clearer dashboard, a checked document, or a business task that reaches the right person with less preparation. Then choose how much of the work AI should perform and what the surrounding system must guarantee.

The projects below offer several levels of commitment. Some begin inside an assistant you already use. Others need a workflow platform, code, or an operated service. Difficulty describes the integration and ownership involved; it is not a promise about how quickly a particular person will finish.

On this page

Explore twelve useful builds

Starter to intermediate · Platform starting point

A website for a real project

A responsive site with your own content and a reviewable deployment.

For: Individuals and small teams

Approach: Coding assistant plus HTML/CSS or the existing site framework

Inputs: An approved brief, copy, images, and navigation

Output: A working site preview

Intermediate · Platform starting point

A dashboard from a CSV

A small dashboard that makes a few important measurements easier to inspect.

For: Individuals and operations teams

Approach: Code with deterministic data calculations

Inputs: A clean, fictional or approved CSV and metric definitions

Output: An interactive dashboard with source totals

Intermediate · Platform starting point

A small utility application

A focused calculator, checklist, or file-organizing tool.

For: Individuals and technical builders

Approach: Coding assistant and a small application

Inputs: Example inputs, rules, and expected results

Output: A runnable utility with meaningful checks

A person compares research cards with a laptop display and an open notebook at a warmly lit desk.

Starter · Worked build

A research notebook

An evidence-linked comparison and a useful list of unanswered questions.

For: Individuals and independent professionals

Approach: Existing assistant; optional scripted validation

Inputs: A bounded packet of public or self-authored documents

Output: A versioned briefing and source register

An open card-catalog drawer, locked document drawers, and a source card sit beside a laptop and magnifying glass.

Intermediate · Worked build

An assistant for approved knowledge

Answers from a small trusted collection with sources attached.

For: Teams and service desks

Approach: Source-grounded assistant or retrieval application

Inputs: Approved documents, versions, and access rules

Output: Answer, evidence, and unresolved questions

Source cards, a structured form, a calculator, and an exception tray illustrate document extraction and review.

Intermediate · Worked build

Documents into checked fields

A reviewable record extracted from a document.

For: Operations and administrative teams

Approach: Workflow or code with schema validation

Inputs: Synthetic invoices or other approved sample documents

Output: Structured fields linked to source evidence

Support request cards are sorted into trays, with a red-bordered exception card beside network tools and a laptop.

Intermediate · Worked build

IT ticket triage

A proposed queue, urgency, and next step for a technician.

For: MSPs and IT teams

Approach: Rules plus bounded AI interpretation

Inputs: Fictional tickets and an approved routing catalog

Output: A recommendation with an explicit review state

A business owner reviews a card beside a laptop and organized document folders in a workshop office.

Intermediate · Worked build

An inquiry-to-quote workflow

A quote draft with checked arithmetic and missing details made visible.

For: Small businesses

Approach: Visual workflow or application with a pricing function

Inputs: An inquiry and a versioned price book

Output: A private quote draft for an employee to review

An employee uses a laptop beside an onboarding kit, with colleagues and connected team symbols in the background.

Intermediate to advanced · Worked build

An onboarding coordinator

Owned tasks across departments with durable case status.

For: Medium businesses

Approach: Workflow orchestration with validated model output

Inputs: An HR-approved case and role catalog

Output: Task plan, exception queue, and completion evidence

A person reviews a card index beside document cabinets and server racks in a warmly lit operations workspace.

Advanced · Worked build

An enterprise knowledge service

Current operational guidance within each caller’s access boundary.

For: Large organizations and enterprises

Approach: Identity-aware retrieval and an operated application

Inputs: Approved procedures, permissions, and source versions

Output: A supported answer or a clear escalation path

A healthcare worker and colleague review a document folder beside a laptop in a clinic office.

Advanced · Industry build

Healthcare referral preparation

A review packet showing present and missing referral information.

For: Medical practices and health systems

Approach: Administrative document workflow with qualified review

Inputs: Fictional referral documents and an approved checklist

Output: A draft checklist and evidence references

Two colleagues review a row of document cards at a warmly lit desk in a law office.

Advanced · Industry build

A dated evidence table that keeps assertions attached to their sources.

For: Law firms and legal departments

Approach: Document extraction, matter access controls, and attorney review

Inputs: Fictional matter documents with stable source IDs

Output: A draft chronology and unresolved conflicts

The first three links introduce suitable development environments; they are starting points for your own application. The worked-build links provide a defined example and its expected behavior. Industry routes add the responsibilities and information boundaries particular to that setting.

Pick a first version you can inspect

Write one sentence describing the deliverable. “A dashboard showing overdue work from this CSV” is easier to test than “an AI operations platform.” Identify the source of each displayed number and keep calculations in formulas or code you can inspect.

For a website, supply the actual content, page structure, images, and interactions you need. Review navigation, forms, mobile behavior, and deployment settings. A generated page is a starting artifact; a finished site must behave correctly for its visitors.

For an application, provide normal inputs, missing inputs, and a deliberately invalid case. Ask the coding assistant to implement the smallest complete path and explain how to run it. Inspect the change, execute the meaningful checks, and keep a version you can restore.

For a connected workflow, begin with fictional records and a private output. Add the real source only after you know what data the workflow reads, which account performs each action, and how you will recover a failed run.

Choose a route through the building blocks

If you are unsure what connects to what, begin with plugins, connectors, and skills. If the job repeats a known sequence, use workflow design to define the steps and exceptions. If the system must choose among tools while working, inspect agent anatomy.

That distinction changes the design. A fixed extraction-and-review process may need one model call. An investigation may need several tool calls and a record of what has already been tried. More components should earn their place by solving a specific problem. Anthropic’s architectural guidance provides background for this choice.

Use choose your stack to compare approaches, then visit the platform guides for product-specific context.

Plan for the work around the model

Estimate cost from the whole job: eligible accounts, connector plans, model usage, storage, hosting, monitoring, and maintenance time. Include retries, large documents, and human review. Check current vendor terms for the configuration you actually need; a free prototype does not establish the cost of a shared business service.

Give every connected project a way to recognize an already-handled request, a place to store unresolved work, and a person who can repair it. Keep an example of correct output and a small set of failure cases. Those become useful regression checks when you change a prompt, model, connector, or source.

Match the project to your setting

For personal work, start with the individual build. A small business needs a clear owner and manageable review. A medium business needs dependable handoffs across teams. An enterprise needs service ownership and access boundaries that hold across its users and systems.

Add an industry lens when the work involves specialized records, professional judgment, or distinct operating conditions. Keep the governance guide beside the build so responsibility is part of the design from the beginning.

Share this page